Audiophile nonsense has entered the chat. If your amp didn't cost more than my house, and if the knobs aren't made of bubinga, are you really trying hard enough?
Audiophiles are like flat-earthists, they don't believe in phase cancelation and when you talk about a null test they say it was filmed on a set by Stanley Kubrick 😁
@@iwancarnifex No and Yes. First, with Apple Music you get now Lossless Music in CD Quality. Before Apple music only offered AAC. Yes You dont need to listen to Highres Music but CD Quality is an improvement. Second just because a High-Res Headphone makes no sense doesnt makes it worse for listening to CD quality music.
I noticed when I converted some of my masters to mp3 the mp3 file sounded more scratchy bright than the same track with a 24 bit wav. It was noticeably different in the high frequencies. After some trial and error I eventually fixed the problem by applying a high cut filter to the inaudible high frequencies in the track. From this I surmised that these inaudible frequencies were being mapped to audible frequencies by the mp3 codec. I’m not saying that high definition audio is superior because it handles frequencies, but rather curious why this happens? Why does the mp3 dithering not handle the mapping more gracefully?
MP3 uses some psychoacoustic tricks, it essentially always loses details not only bc. of the high cut filter. First of all, the Version of the Converter matters! MP3 Converter Software has been improved dramatically over the years, but also the Bitrate is extremely critical. When you really want to use MP3, and i personally am completely have abandoned this format more then a decade ago, except when i want to share a song and don't have either enough bandwidth or Storage Capacity. Beside that, there's no point of using MP3 any longer in 2024! It was invented when the Internet still was on 64 Kbit DSL and you had to pay a ton of money for a subscription plan. Simply switch to FLAC. It has all the comfort of Tagging whilst being completely transparent and it saves you 40 to 60% of Storage Capacity. Beside that, you can ALWAYS retrieve tthe Original WAV. File back and burn a new CD for example.
If you're a music listener, you're not going to hear an improvement going from 320 kbps MP3 to lossless FLAC. Save the hard drive space and invest in better equipment (headphones, amps, speakers, etc.)
Exactly correct. As a digital signal processing expert, I deal with this situation every day. With modern antialiasing filtering, reconstruction algorithms etc higher sample rates does not matter. 16 bit vs 24 bit resolution only changes the effective signal to noise ratio the systems can support. 16 bit supports 65535 slicing levels. 24 bit supports 16,777,216 slicing levels. Hence 16 bit supports 96 dB SNR where 24 bit supports a SNR OF 144 dB.
Get yourself better listening devices. stop using cone and dome drivers with spindel spring coil and rubber etc suspention = high moving mass slow stop and start. .Go for stax lambda best or their top model 009 as far i know. i have stax lamda signature mk2 from 92 No dynamic earphone has been able to beat it. still working fine. real earpads and go for a prize that is more than when i got them new still.
Modern antialiasing filters still cause audible phase shift at 44.1/48 kHz, that's why 24/96 is considered minimum for archiving - phase shift from the low pass filter is less audible.
I will say though, one advantage of using higher bit depths is that it can save a take if you realize (too late) that one of the mics wasn't amp'd enough, and you can't take another take, having that lower noise floor, from what I've seen anyways, allows you to really amplify the signal in the mix without getting any noticeable noise. But admittedly, that is a pretty specific usage, that happens because of user error to begin with. (edit) the whole thing about audio being better when treated less preciously is precisely why people seem to prefer vinyls & tape, even in today's day and age, I work in digital graphic arts, and it's really the same, often in digital artworks you want to add 'noise' and irregularities into artwork, that usually always makes it pop more, and gives it character
Yeah, if you were to record a sensitive source at something crazy low, like -60dBFS or lower in 16 bit you could MAYBE run into noise floor issues. That’s an insanely low record level, so it’s unlikely to be a problem very often. But because of this unlikely possibility it is definitely a best practice to do 24bit when recording instead. That seems reasonable to me. But even 24 bit is actually overkill for that, because none of your analog front end can use 24bit’s 144dB of dynamic range. 20 bit probably would have worked just as well for this purpose, because that’s about where really nice analog gear is going to max out for keeping the noise floor down, but it’s no big deal either way. For consumer audio though? No way! :-) Even 16bit is probably overkill for all practical purposes. Hope that makes sense! -Justin
Exactly. When you use 24 Bit your A/D Converter has a S/N Ratio of 146 dB instead of 'only' 96 dB with 16 Bit and that certainly gets rid of the noise floor.
I always record in 48 kHz/24 for two simple reasons. 1. All my gear supports 48 kHz, so whatever interface I choose to use it’s working with 48 kHz. 2. If the recordings will be used in a video I will have a perfect sample match. If you use 44.1 kHz with a 24 FPS video each frame will have 1837,5 samples, so if I cut both the video & audio when the grid in PT is set to frame the cut will be between two samples. This is avoided with 48 kHz. So, that’s my approach, I use 48 kHz for practical reasons, not audible.
Same for me, I work in 48 for practical reasons, the TV/media business works with this sampling rate and expect you deliver your tracks/stems in 48k/24b.
If you guys are that kind of person whose doesn't hear any difference between "Apple Digital Master" versus CD, you may be definitely deaf. Even from the former Mastered for iTunes quality, from the same AAC quality it DOES A BIG DIFFERENCE over CD quality. Why the people doesn't notice any difference? The MASTER it does matter at all.
Excellent, excellent video explaining the trivial quest for high res. I've played with various rates in my DAW and have not heard differences. Now I understand why. Thanks, Justin; my hard drives appreciate you....
It's worth repeating, as you mention, that capture and delivery are very different things. For example, Zoom just released a sound recorder that records 32-bit float (the F2 lavalier recorder). The huge benefit of that is that it's impossible to blow out the recording as it doesn't peak at 0dB, and levels don't need to be set. You just hit record and you're golden. Once the sound is captured, that 32-bit float just becomes 32-bit bloat (you heard me). Recording headroom is just recording headroom. The end listener doesn't need to hear the empty headroom, and won't benefit one iota from doing so.
You are right. There's not much a point for super Hi res other than selling really expensive gear to people who don't know any better. The one advantage I'm seeing is you have a lot of very affordable devices coming into the market which offer tremendous quality. I'm speaking of desktop dacs costing less than 200$. And current audio is getting way better than it was 10 years or 20 years back
Yeah, and here are two conceptual things that are often misunderstood: 1. There are no stair steps. Digital audio is entirely smooth, continuous, and analog once it goes through the low-pass filter. That’s the whole point of the filter, is it turns everything at the upper frequencies into pure, smooth sine waves, filtering out all the squareness (which are just the upper partials). 2. It only takes two samples to accurately reconstruct any given sine wave at or below that frequency, perfectly in terms of frequency, amplitude, and phase. Sine waves have a particular shape such that you can mathematically reconstruct the whole thing just from the two samples.
Bingo! You are exactly right. Either a given frequency can be reproduced perfectly at a given sample rate (minus some noise, which is determined by the bit depth) or it can’t be reproduced at all. It’s a bit of an all-or-nothing affair. Thanks for the comment, Justin
@@SonicScoop Thanks for the video! Clearly explained, and your 10-year challenge is exactly the kind of real-world experiment I keep using to point out how people really can't hear the difference (though I wasn't aware of yours until now).
What's misunderstood is anti aliasing/mirror effect requiring low pass filters which he mentions in this video, but doesn't go through much detail. If you pay attention he is staying that the mastering process could benefit for higher frequency rate on the recoding session in terms of having a easier time filtering high frequo based on equipment used and recording in higher bit rate gives you much more headroom so there is less likely clipping while recording. I think the majority of people missed the boat on this. In other words, if you actually record in a higher bit rate with higher frequency rates where higher frequency noise with high enough amplitudes could enter the mix and a higher chance of clipping, it would make a difference.
I'd like to point out a big missconception here: quality matters if it's higher or lower sample rate, the lower sample rate the lower the quality that is a scientific fact judging by how the quantization works. The big point that all seem to miss is that beyond 44.1kHz one would not be able to tell the difference and basically is pointless to have so much storage space spent for nothing.
Exactly! 22K is pipe dream for most people; most adults hear in the 10 - 16K range, dependant on age, and the music doesn't sound 'different' just because we've lost a few KHz here and there! If cymbals still zing, and ting like they should, then excessive high frequency recording is a waste.... I know in some circumstances that there are higher frequencies that work with others to give a certain order of distortion that resonate well (certain harmonics) that if missing would destroy the feel of the music and maybe even notes being played, but these are all recorded within the 'lowly' 16/44.1 redbook solution, so it's not an issue! As you say, storage space for needlessly high 'resolution' files is just pointless! I love good old analogue, but I also love my (well recorded/mastered) CD's... :-)
Thank you for interjecting some reason and science into this area. On a somewhat related note, I've noticed that many people equate "pristine" audio with detail. But audio that is very revealing of detail is not necessarily perceived as musical or all that pleasing to the ear. (I came across at least one blind study that arrived at the same conclusion).
True. Witness the love of the sound of vinyl - which has a smaller signal to noise ratio than 16bit digital audio, has more stereo crosstalk, and changes its frequency response and harmonic distortion characteristics according to how far towards the centre of the disc the needle happens to be. Not pristine at all by comparison to digital audio, but people love it.
Very good summary, I totally agree!👍 Especially concerning our golden time of audio quality! Unfortunately I can't imagine what we can expect from future developments🤔. The great leaps in improving audio quality are behind us. But luckily we are now able to concentrate fully in our creativity without be concerned about quality - even at home! I remember recording on noisy tapes in the 80ies. Everything today was science fiction at those times (a DAW running on an tablet!!!)
I really appreciate you taking the time to explain this in a way a layman like myself can understand. I was finding myself obsessing a bit on trying to get these higher bitrates and it was frustrating. I am at peace now with how things truly work. Especially someone like me who listens to music on bluetooth!
I was running this video in background, and that 15kHz startled me, I thought something was wrong with my head. Anyways, it was mathematically proven that you only need double sample rate of what your maximum recorded frequency is, and that would give you accurate reproduction. Any errors in audio are not likely and negligible. Search for Nyquist theorem. 44.1 sample rate can be used to reproduce frequencies up to 22.05kHz. Only a few people, if any at all, can perhaps hear that. Noise for dogs.
Man, I'm 34 and I still use a tascam 4 track cassette recorder and a Alesis 3630 compressor for my music, and people that hear it don't tell me, nobody will like your music because it was recorded on a tascam cassette recorder and a Alesis 3630 compressor
Becuz: the resolution of good quality tape and agood cassette machine and analog outboard gear is better than a lot of digital stuff. Especially when you start using plugins.
Great point, also if the music is great all the techy things don't matter as much. LIke, what is the best mic? The one with a real artist in front of it. WHIle i've heard some crappy recordings done in garageband. I've heard some great stuff as well. Its more so the music, and the person turning the knobs.
Thanks so much for clearing this up. A while ago I did the 320kbps mp3 vs wav test and could not hear a difference at all... With high grade converters and headphones! I thought my ears were to blame.
You are not alone! Being human is to blame, not your ears :-) Good on you for actually doing the blind listening test. A lot of people with surprisingly strong opinions on this never actually do that.
Listening to Spotify premium 320kbps Vs CD quality music (Tidal) is a big difference in musicality. I can EASILY hear the difference between the two. With lower end equipment though that detail that you gain may not be as obvious to your ears. Going back from Tidal CD quality and higher to Spotify premium actually sucks it's so noticeable
@@georgearrows7701 also remember that the sound coming out from your headphones/speakers is going to be affected by everything that it goes through between the source and the end point. So if your pc has a shitty DAC it will lower the quality, a worser power source will also affect it. And also, the mastering of the track will affect it too. Start a free trial on a hi-fi service and listen to some different tracks. Do a side by side comparison
@@zachunter2357 That's the thing. I am using a pretty expensive semi-pro interface (RME Babyface Pro) + pretty good headphones (Slate VSX) and don't really hear it. Anyway, I will do the comparison again because this has been bugging me for years.
I think there is a better low bitrate formats than the mp3 like m4a aac, ogg, because the spectral analysis sometimes shows that some mp3 converters screw up the songs by cutting too much frequencies. So you still get 320kbps but in reality it's less.
Try as I might I have yet to be able to hear the difference between cd quality and the high res audio, which could explain why sa-cds never took off. I’ll have to try the MP3 320k to see, but I know that at 256k I can tell but it’s really difficult.
The other thing is that although we do have the best audio formats available to consumers now, they don't listen to music in any way to take advantage of them. Phone speakers? Laptop speakers? Alexa speakers? People aren't really even listening in stereo.
It’s true, people can listen on some pretty garbage speakers these days! But on the other hand, if they wanted to, really great sounding speakers and headphones are available at a much better fidelity and a much lower price than ever before in history. That’s kind of awesome. Let’s not romanticize the past too much either though. Back in the day, most consumers also listened on total garbage speakers and hardware, often as bad or worse than the laptop speakers of today :-) But people have very fond memories of listening to music on their terrible tiny old transistor radios, or even today on their phones and laptops or even a single shared earbud with friends. As cool as good audio is and can be, music is ultimately way cooler and way more powerful. Hope that makes sense, Justin
@@SonicScoop Hey Justin. You're right of course. Transistor radios were shrill and distorted. I inherited a basic little reel-to-reel machine when I was about ten, which only had one speaker (even though it was a stereo tape head). But then we also had a radiogram! The turntable in it ran at 33rpm (or less), not 33/3rd. I only realised when I took my albums over to a friend's house, and at first thought their machine was running fast. But everyone else's was the same, so it was our radiogram that was wrong.
I would somewhat argue against that, I would say that a lot more people have headphones that sound good, compared to how many people had good sounding headphones or stereo systems in the past. So while some listening, does take place on sub par playback equipment, a lot of people actually also listens to equipment that is far superior to the typical listening equipment just a decade or more ago. However, that is an issue, that mixing and mastering focuses so much on speakers rather than headphones, as most listening with quality equipment these days is actually with headphones, compared to the past when stereo speakers were the dominant listening environment of quality. This causes a lot of issues with the stereo representation, where speakers provide a channel crosstalk that doesn't exist in headphones, and the fact that we can't tell the direction of low frequency sounds in a room, but if if is only sent to one headphone we can.
I'm going to be getting theoretical here but what the heck; I get aggravated, too. Much of this presentation is based on a common misunderstanding of Nyquist's sampling theorem regarding the reconstruction of a signal that has been sampled at anything strictly more than twice the highest frequency present (the "Nyquist rate"), and a common misunderstanding of the frequency content of audio as based on Fourier analysis. Nyquist reconstruction only works if the signal has a limited frequency spectrum but, according to Fourier, any signal that is of finite duration has an infinitely wide frequency spectrum, thus Nyquist reconstruction does not apply to music. Even a pure sinusoid of limited duration has an infinite frequency spectrum (I suspect many audio folks don't realize this); only a sinsuid that started at -infinity and runs to +infinity has a frequency spectrum that looks like a spike. Just imagine any three evenly spaced samples of a single cycle of a sinusoid, around the Nyquist rate. According to the video, you can reconstruct it but there's not a chance. Further, Nyquist's reconstruction requires the availability of every sample to reconstruct any point of the original waveform. So, even if the sampling rate wasn't an issue for a signal that somehow had a limited frequency spectrum, you couldn't reconstruct anything in real time from a digital input stream. You'd have to wait for the entire song to download before you could even start playback. Our digital audio realities are fine but I get aggravated when I hear theories that don't apply being crushed to explain how things work. And while I agree that our hearing just isn't good enough to benefit from high resolution audio, I think a distinction has to be made between (1) what difference high resolution processing makes when it's used throughout the entire recording chain, and (2) what difference high resolution audio makes just for the final output. I suspect high res makes a significant difference in the first case especially if a lot of processing is used, if for no other reason than cumulative errors (resulting in unwanted artifacts) are minimized, but the evidence is in regarding the second. We just cant hear the difference.
Justin, loved it. Three questions: . A famed producer (I believe it was Aerosmith's producer said that excessively high resolution audio was allowing us to hear details in the original recording that weren't there and therefore were a bad thing. True/false? . Jimmy Page said the Zeppelin catalog was recording at such fidelity it's future-proofed for years (or more to come). Page specifically said for future "higher audio standards," which as they're in the future don't yet exist. True/false? . I thought the sound on VHS was/is pretty groovy. Isn't VHS tape a NON digital format? Hence, the sound is 16 bit which as you noted no one can really distinguish from 24 bit. Hence, analog can reproduce in the instance of VHS sound as close to the source as digital? Not a trick question. P.S. I think one big misconception you blew-up in this video, is that if you record say, a giraffe making a sound at a zoo - whatever sound they make at 16 bits and you record it at 32 bits - most folks, including many professionals, believe that at 32 bits of 'resolution' you'll hear greater fidelity with 32 bits. You'll hear a better sounding giraffe, as well as the sound of garbage rustling on the ground. A better wind sound. Maybe human speech, better, from visitors to the giraffe enclosure.
Yes, I think those first two examples come from a misconception about how this stuff works. There are a lot of people who are great musicians, producers-and in some cases, audio people-who don’t really understand the situation fully. They usually have never done double blind trials on any of this, and often don’t exactly understand what’s going on under the hood. But that’s ok. They can be tremendously great at what they do and be mistaken about this. I was once. Jimmy Page is an amazing guitar player. That doesn’t mean he knows how digital audio files work, or that he’s done blind listening tests at various sample rates or bit depths. That’s not really a criticism. You don’t need to know any of that to be a great musician, or even a great producer or mixer. If you had to pick between being amazing at guitar or production and knowing how digital audio works, the first too are way more badass :-) VHS CAN be digital (see ADAT) but is usually analog yes. (Technically you could record digital signals on a reel to reel take machine, and some companies made such machines early in.) But yeah, generally analog in both cases without getting too nerdy about it :-) Analog formats actually have lower dynamic range and higher noise floor than 16 bit. Good vinyl would be the equivalent of about 11 bit in that regard if I remember correctly. I’d imagine VJS would be around there or lower in effect, but I don’t know right offhand. Generally speaking, any analog formats are going to sound less like the source than a 16/44.1 file. The only thing that comes close is an absolutely excellent tape machine running at 30ips. That might come close to sounding as close to the source as any reasonably well built 16/44.1 system, and would be orders of magnitude more expensive. But analog can sometimes sound damn COOL, even if it doesn’t sound exactly like the source. A less than stellar tape machine running at 15ips could sound more badass than the original source. That’s subjective. But it probably won’t sound as close to it as even a basic stock AD converter built into a laptop. That’s objective. Hope that helps, -Justin
You tube sound seems to cut off at 15kHz. I tried using sound tones n the same way as you did from an internet source in class only to find that those higher frequencies are not available.
Higher sample rates introduce less latency when monitoring natively via the DAW. Once tracked at high sample rate for low latency, there is no need to introduce sample rate conversion, which will cause degradation.
One benefit of using 96 instead of 48 at recording is the AD converters allow me better headroom. Not sure how to explain this in technical terms other than what I see when I clip the sound at 48 compared to 96. Another benefit of higher sample rates is signal latency. Although I've never considered 96 being “super high resolution.”
Headroom and dynamic range is a function of a bit depth, and of the analog components of your converter. Sample rate just doesn’t play into it, in theory or practice. If your converter works differently at the two sample rates for some reason, that could be possible I guess. I’d like to know why. That seems like an odd design. But that seems unlikely. If anything, it could be slightly less easy to clip a lower sampling rate because you are dealing with less signal overall because the reduced supersonics and lower anti-aliasing filter. Have you done any properly controlled tests of this where you can absolutely confirm this is the case? Please share if you can! I’d be curious to know more. As I’m sure you sneaky know, we all think we hear things that we aren’t really hearing. It’s happened to all of us. I’m sure you’ve had the common experience of fiddling around with an EQ knob, thinking you were changing the sound subtly, until you realized it wasn’t engaged, or you weren’t on the right channel. That’s just part of being human. As powerful as we can develop our listening skills, it is true that our minds will always be more powerful still. If you have a test you can share that confirms this, I’d be very interested to see it. Thanks much! Hope that helps, Justin
@@SonicScoop Many thanks Justin! I noticed it when running outboard reverbs both at 48 and 96. At 96 I had more headroom before the signal clipped in the converter, RME ADI-8 DS Mk3 via ADAT to/from Apollo 8. So it was purely by eyes I noticed about signal not clipping at a certain level. Nothing else was engaged or changed so I was a bit surprised myself. Probably more to it than headroom being changed then : ) I'm like you-being aware of the sound while fiddling and thinking yup that's a tad better-and realising the plugin wasn't active : ) I have that perspective when listening to anything these days but trying to explain how the brain fools us is a difficult task but I'd be happy if you talked more about that, blindtesting, do we actually hear the difference we think we hear etc etc.
I listen to MP3s on an almost 20 year old Zune with Beyerdynamic studio monitor headphones and the sound quality is flat out amazing. 44.100/16bit is more than enough to capture and reproduce sound that humans can hear and a good ripped Mp3 sounds nearly identical.
My argument for using 24 or 32 bit float is just headroom, I can record really quiet sources and then gain them up and compress the living shit out of them and the noise doesn't come into that audible range. For consumer side I go 48/24 because it is standard for video.
I wish I knew half of what you talking about. Really just wanted to know if I wasting money with tidal Master plan. If hi - fi would have been adequate. Or if I'm just imaging Spotify not sounding good from my system.. Last two mins most useful for me. Listen to what works for you. Sound advice
Years ago, I was tracking thru 18bit motorola converters. I was dropping the teac four Trac drum tracks down to digital. I noticed it sounded much more like the tape in 24bit than sixteen bit my sampling rate was 44.1 in both cases. I learned 24bit is better for resolution. However I had to mix down to 16 bit digital in this case. My hardware could only handle 16 in the multitrack window.
I use to bounce my music from Cubase (6) to a 44.1kHz/16Bit wave file, then drag it onto my iPod mini, go out to smoke a cigarette and gaze into the distance. It occured more than once to me, that I was bouncing out an 320kBps MP3 and was not aware of it. The funny thing is: That iPod makes it a blind test without my provision - it doesn't show or tell you in any way which format is being played back. BUT: My nearly 40 years old ears seem to reveal it - whenever I listen to one of my productions and I have the feeling it's not uncompressed sound... guess what: In fact it would be compressed, really. I would yet have to fail on that one to conclude an MP3 sounds just as good as a PCM. But don't get me wrong: I agree wholehartedly to what you say in this video.
Here is a blind test that will change your mind, Justin. Listen to a 320k vs. a wav or flac (create both from the best same source) of The War on Drugs' "Lost in the Dream." I have, and for a fact, I can tell you there is a specific character that shows the difference. I challenge you to notice it, it can be deciphered in the first minute of the song. Another stunning mix was Morning Phase, equally decipherable. The sense of detail and SPACE that is destroyed by 320k is obvious to me, at least. Most won't notice, so don't stress, that is true. But when you personally want the best, why not go ahead and mix for that. Try to hear it, try to make it.
Interesting to hear! What are you using to listen to the these properly double blind? I can tell you that I have had similar experiences in sighted listening tests. For instance, when I created a similar listening test for our readers, I could have sworn up-and-down that I could hear the difference when listening sighted, and that so many people were going to get it right. But as soon as I properly jumbled them up so I didn’t know which was which, those differences seemed to evaporate like a mirage. Ultimately, the results of thousands of responses to our online test were no better than chance: www.trustmeimascientist.com/2012/04/02/take-our-audio-poll-do-we-need-higher-definition-sound/ If you can consistently distinguish this, that would be amazing! No one in the world as of yet has been able to show that they can consistently distinguish between 320kbps and any higher resolution format in a proper double blind listening test. I’m not saying that it can’t be done. Just that it never has been yet on record. If you can do it, that would be amazing. I’d happily write a glowing full length article about you and how you are able to do this so we can share it with the world. I’ve had this challenge out for almost a decade now, and so far no one has done it yet. It would be awesome to finally put a bow on it. What software or hardware solution are you using to do proper double blind listening tests on your end? Is it an ABX tester? That could do the trick. Thanks, Justin PS: Here’s the original challenge on my old blog. It’s been seen tens or hundreds of thousands of times, shared all around forums and all that, but no luck yet! Be our champion :-) www.trustmeimascientist.com/2013/09/03/think-you-have-golden-ears-take-the-scientist-challenge/ www.trustmeimascientist.com/2013/10/07/update-on-the-golden-ear-challenge-who-will-conquer-audio-mount-everest/
@@SonicScoop Well I went, prepared to be humbled, but the links for Line 'em Up and Kite of Love are broken. I did find in another of the articles the foobar can have an ABX tester plugin, I'll do that. My previous tests had my wife switching the tracks; a program that does that will be fun, so thanks very much for that. Also, please know I meant no disrespect, I love your channel and learn what I can from it :)
320k MP3 isnt a "hi res" format - its still a compressed (althogh half decent) file. Im pretty sure Justin is referring to higher sample rates and bit depths.
Great video. I’ve had the same concerns myself and you properly articulated why. The best ever hifi upgrade I did was buy a Nord One Up amp. This was a game changer for my PMC OB1s speakers. I have heard differences between formats but not in ABX tests. What I mean is the Linn 96/24 for example sound great but think the real reason is the the audience for these are people who care about music and the producers bother to mix the tracks well in the studio. i.e. don’t dynamically compress the music. Place instrument well etc. It’s not the format it’s the production. Madonna immaculate conception on cd sound just as “interesting” and alive as any 96/24 recording do.
Love your content bro. I'm 28 years old and can hear clearly from 30hrz to 22khz. I tried many app from play store and that's my hearing rage. However that higher frequency I perceive them as noise.
Ok… sound engineer here. While the maths and physiology is correct, you’re missing a couple of critical details: Due to a physical limitation, DACs don’t perform optimally at 44.1KHz, or even 48KHz. Clock a DAC at double that rate and it will perform a higher fidelity conversion from digital to analogue. The result is clearer more transparent sound. That said merely possessing a 96KHz audio file is not sufficient to unlock that performance… for starters the audio has to be read in to your DAC at 96KHz. Now, even if your streaming providers offer 96KHz audio (and some do), if you send that to your transport over Bluetooth or AirPlay, then it’ll be downsampled to 44.1KHz… does your transport upsample it on the fly up to 88.2KHz or better before talking to your DAC? Unless it does, then you need a source outputting audio at 88.2KHz or better, with a physical cable running between that and your transport... or better yet just skip the whole transport stage and plug your source device directly into your DAC! Here’s the cool thing though, the original recording doesn’t even have to be at 88.2KHz or better - you can load any old 44.1KHz wave file into a DAW, then re-render to 88.2KHz or better, and that’ll work just fine. If you’re fortunate enough to be aged 15 or less, with pristine hearing, then starting with a 48KHz source file will sound even better, but for the rest of us that’s completely overkill. So, the first point being that that any improvement isn’t baked into the audio itself, but purely due to the DAC operating in its sweet spot. To built a DAC that could do this at 44.1KHz would be prohibitively expensive, if that were even technologically feasible, and I’m don’t think it is at present. Second point… bit depth plays a much more important role during the recording, processing, mixing, and mastering processes. I could write a whole book on this, but the long and short of it is that the audio is digitally amplified and de-amplified at various stages then layered on top of many other audio tracks, so all those noise floors quickly stack up, to the point they become very audible… therefore you want the original audio recorded, processed, mixed, and mastered at least at 24bit PCM, but there are arguments for going beyond that. As someone who predominantly works on live concert albums, I favour 32bit FP (floating point), since that completely offsets the issue of digital clipping. But… once the mastering is done, that audio can happily be downsampled and rendered to 44.1KHz 16bit, and the information within will theoretically be indistinguishable to most adults, no matter how good they are at listening! So, in most circumstances, the advantages gleaned from playing back true 96K 32bit PCM audio at that rate, are only really useful to mixing and mastering engineers. Sure, you’d be able to hear the music more clearly (often to its detriment), but you’d really have to concentrate! We only need this level of clarity to tune the audio to what we think is perfection. For playing music I’m perfectly happy with a 44.1KHz 16-bit… unless I’m only there to listen, wherein I prefer that DAC to be running in its sweet spot… but mainly just so I can get a hard-on from hearing the processing artefacts ;) Lastly, the biggest influences on audio quality are, in order of importance: The artist and their performance Their instruments The recording environment The mics and mic placement The mic preamps The ADCs The sample rate and bit depth used to record The skill of the engineers involved, and the techniques they employ Your preamp, amp, and speakers Your speaker placement Your room Your DAC The audio will however, only be as good as the weakest link in that chain. Things that don’t matter, at least not for digital audio quality: Your cables Your power supply Your transport (provided it only transports… most don’t) And if there’s any issue with digital audio, then you’ll hear it as a significant disruption of the signal, and never as some barely audible distortion. Lastly… don’t be blowing hundreds of bucks on an audio cable unless it’s for the looks - you should see the crap we use in the studio. If cable was a problem then the 128 or so cables that said audio passed through before it even left our studio, not to mention the thousands coms cables it ran through as it was piped across the cloud before it finally entered the ones leading to your speakers, would have utterly mangled it! As for digital signals, well the audio either works or it craps out. That’s the whole point of digital! If you copy a text document from digital device to device, or expose it to vibrations or EMF… does it ever corrupt or degrade in any subtle way? Does the text become blurry? No. You can copy it a bazillion times across every cable in the cloud and it’ll remain forever a perfect replica. That’s the whole killer use case of digital audio!
The real benefit of high sample rate is moving (most of) the phase shift from the low pass filter out of the audible range. You can hear the phase shift, right? If not, maybe mastering isn't your best career option. 24 bits gives you more s/n for mixing, that extra 8 bits is ~48dB you can throw away and still have CD quality s/n.
Yes, that's one of the main points of the video! That any difference heard in sampling rate is really a difference in the anti-aliasing filter :-) What Viklas says is also correct: Oversampling fixes this issue. But even if that weren't the case, you may also misunderstand "phase shift". EQ doesn't "cause phase shift" so much as *phase shift causes EQ*. For practical purposes, you have it backwards. It's not that you EQ something and you get EQ, plus this side effect called "phase shift". No, it's quite the opposite: A conventional EQ allows you to apply phase shift, and the result of that phase shift is the EQ that's being added. The EQ is the side effect of the phase shift! Not the other way around. Now, this is complicated further when you are EQing two parallel signals that have some variation between them. If that case, you will end up with some additional adding or summing around the corner frequency of the EQ, and this is indeed more significant on EQ hi and low pass filters. But the "phase shift" is not some crazy comb filtering effect that our minds may conjure up when we hear the term "phase". Rather, it's generally, a greater reduction than anticipated in the extreme lows or highs. So really, the "phase shift" just creates a slightly greater amount of EQ than anticipated :-) The additional s/n ratio for mixing from higher bit depths is 100% irrelevant, because the s/n ratio of 16 bit is already 96 dB+ X-D Seeing that most commercial mixes are well under 14dB of dynamic range, and even in old school classical and jazz they are rarely greater than 20 or 30 dB (at MOST), this is just not a problem at the mix stage that needs to be solved. I hope that helps! Very common misconceptions here, and they are all addressed in the video if you get through the whole thing :-)
Most of the difference to music quality appears to be down to the recording and mastering process, in my personal experience. While 320 kbps MP3 might be hard to distinguish from 16/44.1 FLAC... I wonder whether that's not really dependent on the hardware to a good degree. The better setup, the easier the perception of transients and hence space, distance etc., might be. I can for sure tell that the quality difference between streamin of something lossy like through Spotify versus lossless like through Tidal, is night and day, in clear favor of the latter. That is though probably also mostly due to Tidal sourcing the tracks much better. I've been custom-upsampling (like 64M taps sinc filter using SoX) my FLACs to 24/96 (for my phone) and 24/768 (for my home setup) because it gives me a bit more refined, clearer and smoother highs, which I actually hear the difference in (although the differences are kind of subtle, they won't really hit one in the face), and that's comparing FLACs, and even that on mid-fi equipment, nothing that breaks the bank. MP3 is really out of the picture there. Sorry for not having done the double blind test experiments! Will for sure try that though.
Also, today's delta-sigma DACs and even some other DAC tech like R2R tend to oversample themselves, unless you supply them a high-enough input stream. Save for Rob Watt's M-Scaler or something in that league, the small chips of these DACs that need to do their resampling in real time can't really compete with an offline process that takes a few minutes per track on a modern CPU core and a good 6 GB of RAM to finish... And another very clear benefit of oversampling to mention is that today's music that gets released is most often overly loud and already digitally clipped, which means informations is already lost in there, mostly impacting the higher frequencies. A proper upsampling process (after digitally lowering the volume beforehand, sometimes by as much as a whole third), can reconstruct the waveform and partially undo the damaging effects of digital clipping. Perhaps it is this that justifies the oversampling hassle to me, since the space storage and the bandwidth between my home NAS and my living room computer doing the audio playback, is not really an issue.
Thank you Justin, this comforts me in my unashamed attitude of "If it sounds good, it's good"! I have a question about the "Air Band" on Maag plugins, I really like the "brightness-not-harshness" it allows my 58 years old ears to perceive and appreciate on instruments and mixes. I usually use 4-7dB @ 20k and only recently used 15k on the EQ2, should we be cautious of unheard frequencies causing aliasing generated by the plugin at 20 or 40k?
Good to hear! Aliasing should’t be a problem and should be filtered out if everything is working properly. But there’s a chance that to you get ears you might be boosting more HF than you realize :-) It could be good to double check things on a frequency analyzer to make sure the super high end you have trouble hearing doesn’t look to crazy relative to references. Either that or visit a high school classroom, draw your nails across the chalkboard to get their attention, and then ask them for feedback on your mix ;-) Hope that helps! -Justin
So, Justin, I see you've changed back to another JZ Mic which looks like the V67! Nice choice! As for the topic, I really enjoyed listening to this and your perspective. I have to admit that I was somewhat of a numbers-fan and thought that bigger meant better. However, after listening to you it made sense to me in that you really won't hear the difference on the listening end of the sound. And I'm convinced that you need some really high-quality headphones to even attempt it....if you can. I've tried and didn't hear a difference. I am going to have to do some really close comparisons when I do get a chance to see if I can capture the differences between the two resolutions. Again, thanks for this video Justin!
I remember converting the black album cd in the '00 to various kbps mp3, starting from 32kbps to a variable which at that time was up to 220 kbps I think. And the sound depended a lot of the speakers. In the PC starting from 128 kbps sounded good, but in the hifi equipment to a properly bass sound must had be the variable, not less. Even if actual 320 kbps mp3 is said to be enough, I consider quality cd, be WAV, be FLAC, etc, the most honest format to sell music, just in case, the original, nothing removed.
Sounds reasonable to me! Though I understand why people don’t want 4x larger CD file sizes on their devices or streaming sites for practical reasons. -Justin
I use Wave 48K at 32 float because I believe if I yell too close to the mic, and it goes way into the red by accident, it won't distort. When I export to MP3, it still sounds good. Am I wrong?
Good talk. As a matter of fact, nothing touches say, a 2" Studer and a vintage Neve console. Of course also the old outboard gear. There's a reason for all the modeling algorithms of that vintage analog gear. Yes, analog has a "smearing". But, digital lacks some body that analog has. Meant to mention. There's nothing like the saturation sound of hitting 2" tape really hard.
Some good examples of that "tape hit hard" sound: "Best Friend's Girl" by the Cars, "Bicycle Race" by Queen, Both produced by Roy Thomas Baker. He would have the engineer hit the 2" 30ips master really hard and said 'if the recording head glows red, we have a hit.'
Well done on this video! I have been singing this same song for years on forums and with students. As a delivery medium, why do you need more than 100dB of dynamic range? (dithered 16 bit). The part I think many still struggle with, and I can see it in some of the comments here, is not understanding the bit depth of a session. (Speaking about Pro Tools here, but I believe other DAWs are similar). If set my session to 16 bit and fill it with 16 bit files, we are not operating in a 16 bit world. Every DAW these days has at least , a 32 bit floating mix engine. You are mixing and outputting a 32 bit signal IN SPITE OF THE FACT THAT IT MIGHT BE MADE UP OF 16 OR 24 BITS FILES. The bit depth of the session only relates to the bit depth of any files you record into the session or any files you import that require conversion.
I started as an engineer in the 80s. 60db headroom was about what we had. When the 16 bits came in no one could believe how low the noise floor was. All DAWs now have 32 bit internal buses and some even have 64 bit summing. No one seems to understand. I still deliver projects at 24 bits as otherwise, people don't think they are getting their money's worth. This is to make vinyl from lol
I find the only benefit of hi-res releases are that the albums normally receive a better mix, Metallica's Death Magnetic is a great example... the retail CD is a brick walled mess victim of the loudness war and due to the compression it's actually tiring to listen to it, the 24bit/88.2khz hi-res release just sounds better due to the mix and lack of compression, so I just converted the flac files to 16bit/44.1khz and burnt myself a CD that sounds better than the one I parted with cash for..
Yes, that is a valid reason to repurchase etc. It's a shame however that people are being convinced they should buy because of the numbers and not because of the fact you get a better mastering or better recording. If you convince everyone it's the numbers that matter, they will rebut every CD ever owned by them and it also attacks the second hand market. Imagine reprinting a book and encouraging everyone to rebuy because the new print has a HD font that will improve your immersion in the text.
I swear that CD's sound slightly fuller in the low end to me in my car vs. me playing the same album on Spotify on my phone using an AUX cord. I have no idea how much of this is just placebo or if there's something else at play, such as the DAC in my phone or whatever.
Spotify is objectively worse imo, I listened to my CD version of an album then the Spotify stream and the spotify one is so muddy compared to the CD, like there was a blanket over it. I'm pretty sure I have "high quality" enabled on my Spotify too, and was listening to a downloaded album so it should've been optimal quality
That is because of the Aux cord , you will get some armonic distorsion from the plugs themselves and also if you are using a 3.5mm jack the lower freq range is not as acurate ,, /( correct me if I am wrong but from 120Hz down / so yes, you are rigth, you can feel and hear the diference , but it is not because of the file audio qualitty ;)
@@fernandoferreromusic I thought it might be the aux cord, but then I listened to the same album on my MP3 Walkman (same cord) and it sounded much closer to the CD quality than the Spotify version :o I think this might be specific to Spotify, they must do something to the audio files even if they're technically at 256/320kbps I think they do some extra compressing to save space.. although it could also just be my phone's aux jack vs my Walkman jack, the phone jack being inferior?
There are a few factors here. One is that you could be listening to lower resolution mp3s, such as 160kbps instead of 320kbps, depending on what version of Spotify you’re on. Trained listeners may be able to discern subtle differences in that case. (Though most people can’t.) Another is the the DAC, sure, but often more importantly, the analog components of both your phone and your car stereo that aren’t in the equation when playing the CD. Another even bigger factor could be volume differences between the two inputs, including Spotify turning down louder material, but also the audio input (or phone output) being quieter than the CD. Level differences explain a LOT in audio preferences. And Spotify does indeed turn down the loudest material for consistency, which your CD player and your MP3 player likely doesn’t do. Even if none of that was at play, and you had identical levels and audio (which you probably don’t) sighted listening makes a difference as well. I could go on. But I probably shouldn’t! :-) I hope that helps, Justin
I have 100s of classical cds at 16 41k. The biggest difference I can hear is not in the quality of the amplifier, the DA converter, the headphones, the full tube amp, the toroidal transformer... It's in the microphones, mic placement, instruments, studio preamps and mastering. I do think a small difference is gained however in SACD.
This is a very fin topic. Someone should have a discussion with the hi-fi (consumer) industry about this and then with the audiophile listeners using converters at 768 sample rates when listening to lower sample rate files.
I do a lot of sound design and multiple renders of audio. Stretching, bouncing etc and have recorded at 96k for the last 12 years for that reason. Also my clients want 48k almost 100% of the time so the conversion is seamless this way. I think some of the “myths” of higher SR are leftovers from when converters weren’t as good (maybe), like early aughts. But if you really wanna hear a marked improvement, get a master clock. That was an eye-opener. I run 96/24 with Lynx Auroras and a Big Ben-that’s enough! But this video gets into consumer habits, and listening-and I agree. CD is fine and I really prefer vinyl. But it’s apples and oranges comparing processing digital audio with delivery specs vs listening to music. My .02.
Absolutely, for sound design or sample libraries where you’ll be time stretching audio or pitch shifting down significantly, higher sample rates make sense. That’s one of the special cases mentioned near the end of the video. And if you’re working primarily with video, 48k there is the norm, so may as well deliver it that way instead of sample rate converting a 44.1 file. Thanks for the comment, -Justin
Ah I wrote this literally 1-2 minutes before that caveat, haha. I also agree with the many other attributes mentioned as to creating good recordings, or rather, the Things That Make SR And Bitrate Moot. Like bad preamps, oppressive, stark, reflective rooms, poor levels, and questionable choices-the list goes on!
I've been saying for quite a long time, anything beyond 24-bit 48kHz (chose this really only because it's what you'd see on DVD's and Blu-Rays) is useless outside of production environments.
In the early 2000’s, I remember a salesman at a music store tell me I should use 24 bit over 16 bit to avoid audible “stair stepping” in a fade out to digital black. And yeah, I THINK I heard that once, in the last second of a 16-bit file where sound was attenuating between -60 to -90 dB while my interface and monitors were turned up to maximum volume. *I think.*
I think we should all consider that when listening to pure tones as hearing tests it doesn’t reflect the real world where tones are combined. It’s the simultaneous interaction between frequencies, even the frequencies that are considered outside of the range of human hearing because of single frequency hearing tests primarily, that can get very interesting so to speak in this conversation.
That's not how it works. Yes, there's intermodulation distortion - at least in theory - but that's too low in level to be audible over the actual playback material, and probably not present at all in actual recordings. I'm not a fan of test tones, either. But this often cited euphonic interaction between frequencies is a myth. There are interactions, but they're either not audible, or just as audible at normal sample rates.
@@michaelanderwald4179 It also sounds more like something that would result in amplitude peaks rather than frequency distortions. If we can't hear a certain frequency, it's not going to matter much if two sounds at that frequency interact a little and boost in volume, as we can't hear it. Perhaps I'm mistaken.
@@FloatingOnAZephyr Depends on the type of interaction of the frequencies with each other. Intermodulation distortion can go much lower than the frequencies that cause it, but it's low in level and probably not something that would sound good. It can actually sound a bit like aliasing, so very much not harmonic at all. I've sometimes wondered if some people prefer DSD recordings because the energy around 100kHz due to noise shaping adds some kind of biasing effect on the playback system. But that could be added in other ways to a "low" resolution playback system. Also it's pure speculation.
Hold on... if I'm understanding this correctly: I can record a VO using a sample rate of 44.1 kHz, and I can do the same reading in another recording at 48 kHz, and the only real difference is that some of the higher frequencies (which annoy me anyway) won't be there in the 44.1 kHz?? Or have I misunderstood?
Human voice doesn't have much information beyond 18Khz, I mean there is something beyond that in noisy consonants like K, T, S sounds, and in the sound of breathing and other things like mouth clicks. But honestly the part of those sounds that you will loose going form 48KHz to 44.1KHz is not something that you would want to hear, if your ears can actually hear it and your sound system can reproduce it.
@@kelainefes Thank you very much! I mean with VO, most of what you mentioned is something that narrators etc. work hard to eliminate anyway, or rely on audio engineers to remove for us! Again, thanks! 😀
@@HungryForTastyFoodAndComicArt Yes, for a voiceover an engineer would want to remove mouth clicks and greatly reduce the amplitude of breaths to the point you fell them more than hearing them, and the consonants would surely not be receiving boosts in very high end, and depending on the microphone used would be processed to sound a bit "darker" or "smoother" and certainly not brighter.
Pretty much! There is some chance that a given converter could sound slightly better at 48k than at 44.1k if it is using an analog anti-aliasing filter, so you might get a teeny bit flatter response around 20k or so with the 48k mode in that converter (if you can even hear that high) but that’s about the strongest reasonable case you can make, I think. -Justin
Apparently you’ve never heard of noise shaping filters as it applies to bitmapping. Dsd sounds remarkably different than an mp3. Maybe your speakers aren’t able to show you what you’re missing.
FYI, TH-cam processing adds an incredibly steep low pass filter at 15kHz. Some older vids may have one at 18kHz, but it seems like they lowered it a few years ago.
TH-cam does a low pass, yes, but as you say that it is very steep I will speculate that you are uploading MP4 videos with audio in AAC format. Many encoders do a low pass at around 15KHz just like YT does, so if your encoder does not give you the option to disable that filter you can assume it is enabled, and that your audio is being lowpassed twice. The solution to that is to upload .MKV files with MP4 video and wav audio, which are supported by YT.
What’s interesting to note here is that most people not only can’t hear super high frequencies, but they generally ALSO can’t hear a fairly steep roll off beginning at 15k :-) I can hear the difference between 128kbps and higher resolutions double blind, all day long. And this is exactly what I listen for these days. But even I have to admit it’s gotten fairly subtle these days. Someday, I’ll be old enough that I probably won’t be able to. Super high frequency hearing tends to decline with age. At that point I’ll probably be approaching the age of the average person who swears that high res is better! X-D 128kbps does benefit from a fairly aggressive high frequency roll off for sure. 256 and 320 less so. Back in the day it was easier to distinguish 128kbps, back when the codecs were suckier and they didn’t do the high roll off and when I had the ultra high frequency hearing of a teenager! But take any one of those factors away and it becomes more subtle when listening blind, even when you’re trained and can do it reliably.
@@SonicScoop Indeed. Today the only artifact I can consistently hear at lower bitrates is the "digital jingling" at the top-end of hi-hats and cymbals. Even that goes away for me starting at 256kbps and my blind test results 256kbps+ were no better than guessing. My high pass filter note was really more a warning against using those high-pitched "hearing test" videos on TH-cam...they don't work too well :-)
Just ran this video's audio through a spectrum analyzer and it's all there, right up to the 19k using 144p video quality. I don't think audio quality changes with picture tho. Every other day youtube changes. Thanks for giving me something to do.
But two things I didn’t quiet get til now, should I use a low pass filter at 22.1 kHz in my master or my mix to get rid of that stuff as soon as I do not apply over sampling with my limiter? I thought over sampling would just make the chops off more rounded like a analogue clipping does? 🤔 and if I apply more over sampling at high rates, the sound really smooth outs more which is not always wanted, I really tested that with the pro l and layed 5-6 different versions from no over sampling I think up to 16x over sampling and stuff, so should I cut everything off higher than 22.1 kHz if I’m not gonna over sample with my limiter? And the second thing is about dithering... I thought dithering should be used as the final mastering process, when converting 32 bit files to 16 bit to avoid artifacts isn’t that so? Because if I’m not hearing any noise floor and I have to admit that was never the case because I’m producing mostly hip hop and rap where are just no quiet passages in my song, does it makes any sense to apply this? Thanks yours sincere Mathew 🙏🏼
And if oversampling is acting like a brick wall I wonder why this is affecting the higher frequencies because mostly the low end gets limited the most or am I wrong in this case? You threw up a huge question mark right here and now lol, but I heard that oversampling is reducing the aliasing but I didn’t know it has to do with these high frequencies, I thought it would be about artifacts in general
You're sorta right and sorta wrong. When testing my fave synth at 2441 and 2496, I was shocked at the difference in clarity. Always good to start at the highest possible quality and then Downsville, if needed ... that's my 2 1/2 cents.
Thanks for this post! I’d love to hear your perspective on apple digital masters, I feel like they sound better but I don’t know the engineering behind it.
Thank you. Your video helped me staying with Spotify for the UI, multi-platforms (Connect) and international songs selections. Honestly tried Apple Music and Tidal and just not getting any additional benefits.
Yeah, 320kbps is a pretty amazing consumer format. Improvements from there have routinely been shown to be existent in double blind listening tests, so whichever service you prefer for other reasons is probably the more important variable. Glad you found a choice that makes you happy! Very best, Justin
Great question. As a mastering engineer, I prefer to receive whatever bit depth and sample rate my clients were working at. But in the overwhelming majority of cases, the final deliverable resolution of the master is going to be 16/44.1k. That's what most digital distributors accept, and for good reason. If I was still recording and mixing today, I'd probably work at either 24/44.1 or 48. For my tastes, I'd probably go for 44.1k. In the cases where I can hear the difference between the two rates, it's usually because the converter in question uses an old schools anti-aliasing filter that has slightly more super high frequency roll-off at 44.1k than at 48k, and this can actually sound a little more pleasant and "tape like" to my ear in most cases. But there are probably converters where I might prefer the opposite. In other cases, say where the converter is designed with a proper digital anti-aliasing filter, the two are indistinguishable, so in those cases, why not use the smaller one, if there's no benefit to going larger? In either case, I'd probably test my chosen converter or interface in a double blind way to see if I actually got any perceptible benefits either way. Even in cases where there ARE perceptible differences in a given converter, totally blind, the are likely to be extremely subtle, amounting to a very tiny EQ difference in the extreme highs. If the difference is more significant than this, that's a problem with the converter design IMHO. If I was working mostly in video, maybe I'd stay at 48k because that's more the norm there to cut down on the number of conversions needed back and forth where errors can happen some of the time in fairly rare cases. Hope that helps! -Justin
@@SonicScoop Yeah I did a brief project with Producer Greg Wells and he was using 24/44.1 and so I started using it after that LOL other engineers argued with me but I couldn’t hear the difference. I wanna do a session at 16/44.1 but I’m scared lol So the streaming services prefer 16/44.1 submissions? Imagine how much less hard drive space I’d use and how much faster my transfers and bounces will be as well!
End consumers getting the file in the same format it was recorded/mixed avoids a Sample Rate Conversion, which can mean less degradation. Higher sample rates give a more accurate trace of the analog/electrical wave form.
If you want to avoid sample rate conversion, don’t use the unnecessarily high sample rate to begin with :-) As far higher sample rates “giving a more accurate trace of the waveform”, I’m sorry but this is incorrect. That’s the exact misconception that we start off the video with, on purpose. There is just no mechanism in physics by which increasing the sample rate can do anything except for increase the highest frequency you can reproduce. It’s a bit of an all or nothing affair: Either a given frequency can be reproduced perfectly at a given sampling rate, minus some noise (which is determined by the bit depth) or it can’t be. This can be confirmed for yourself with proper testing. Similarly, there is no mechanism by which increasing bit depth can do anything but lower the noise floor. If you believe that there is, that would be an amazing breakthrough discovery and science! I mean, could you explain the mechanism by which anything else could occur? I hope that helps, Justin
@@SonicScoop when you have more samples per second, there is less rounding going on as the digital waveform tries to approximate the analog voltage (input). A hypothetical infinitely sample rate would track the voltage perfectly, to the atomic level or beyond. There would be no rounding, at that point. Its calculus. Approximation of a circle using discrete values. More "steps" equals a closer approximation to the continuous curve. Why not use less noise, noise is not typically desired. Even if not audible, it has interaction with the audio. There is no universal reason to not be technically better. If for no other reason, than archival purposes. If you null test an mp3 @ 320kbs, and a 24/96 file, there is no residual sound? Interesting discussion. -Kyle
@ReaktorLeak im talking about the capture at the ADC stage. Voltage to binary. A "perfect" conversion would track the voltage fluctuations down to the electron. Otherwise your missing information at conversion. ie the information that would otherwise be in between samples, is not recorded, it is ignored, the waveform jumps to the next discrete value, the analog signal is continuously variable. Up to the limits imposed by the hardware. Trace the outline of a circle with straight lines, once with 1" lines, another with 1/4" lines. The shorter more frequent lines give a "truer" representation of the circle. This is akin to sample rate. More samples, more accurately traces the voltage functions aka the analog audio waveform.
@ReaktorLeak you cannot recreate the circle however, because samples are discrete values (straight lines) in your example you would have a triangle. The samples cannot reflect the "inbetweens". The voltage change happening in between samples. "Reconstruction" of the inbetweens ends up in averaging. You don't "know" the voltage fluctuations, your guessing them based on the two points. A momentary transient occuring between two samples, would be ignored. It would go undetected, and get rounded/averaged away. The samples can't dectect things in between samples. They are akin to steps, not curves. Calculus shows us that more discrete values (samples) more accurately trace a curve than less. Its just like frame rate in video. 29fps is fast enough to fool the eye, until you slow it down. Higher frame rate is always truer to continuously varible.
@ReaktorLeak you can perfectly re-create the circle IF know ahead of time its a circle. If you do not know its a circle it could be any possible configuration of lines between those 3 points. A converter doesn't "know" the voltage fluctuations in between samples.
Ultrasonic content in high sampling rate audio can actually be very counterproductive, if the complete audio chain does not support it. Tweeters and amplifiers can generate folding frequencies which can be considered strong distortion components.
I don't know the details of Rubert Neve's listening tests where he claimed people could hear the difference between an 18 kHz sine wave and a square wave with an 18 kHz fundamental, but I suspect this kind of distortion was the culprit.
Last week I switched to high sample rate for my mixing template. Not for that "super high quality" sound (actually when the files were recorded at 48kHz than they CAN NOT be converted to higher samples rates with all that extra information) but for the digital processing I use. Some of them have oversampling but some simply don't. And those plugs which don't have oversampling CAN achieve a more accurate harmonic saturation/distortion. It depends on how well the plug-in was coded. So the sampling rate of my project is high but my exported file is still 16Bit 44.1kHz/48kHz. The only downside is the processing power - but hey with this "trick" you convince yourself using less plug-ins 😉
If it works for you keep on doing it! I would suggest that you eventually do an experiment though, just for fun and certainty. Try this: Bounce your mix from your 48k session. Then, copy your entire session from 48K to 44.1 K. Then, bounce that otherwise identical mix as well, to the same bit depth and sample rate. Now, load both up into an ABX tester and see if you can hear the difference! This would be one way to confirm that you were actually getting the benefit you hope you are getting. Maybe you are, maybe you aren’t. I don’t know! I have recommended this to a friend who worked at 88.2, and they have found that they couldn’t tell the two files a part. In some cases they stop using 88.2, in some cases they didn’t. But in no cases so far did they hear anything like the kind of difference that they thought they would. Maybe your situation will be different. In either case, if you try it, please share the results here! A lot of people would be interested to hear them I think. Thanks and I hope that helps! -Justin
i converted a cd wav song with sox to 4 bit 36khz. what shall i say, i dont hear a difference. The Importance of Bit Depth in Low Dynamic Range 16-Bit and 96 dB Dynamic Range: In 16-bit audio, this means the entire theoretical dynamic range of 96 dB is available. This is divided into 65,536 steps (2^16). However, in practical use, the entire dynamic range is not always utilized. If the music has a dynamic range of only 5 dB, for example, in practice, only a small number of the available 16-bit steps are used - about 2-3 bits out of the 16 available bits. This means that the remaining bits contain no "useful" information, as the music doesn't get any louder or quieter. 4-Bit and Dynamic Range: In 4-bit audio, the dynamic range is limited to 24 dB, and the volume levels are divided into 16 steps (2^4). If the music has a dynamic range of 5 dB, the music will be divided into only 2-3 bits, which is more than sufficient to represent the audio content without noticeable quality loss. Since the music has a low dynamic range, the remaining "bits" of the dynamic range are not required and are not used. Sampling and Bit Depth Bit depth is primarily a measure of the resolution of the dynamic range, i.e., how finely the volume levels are divided. If the music uses only a small dynamic range, only the corresponding bits are utilized. A 16-bit audio format theoretically provides 96 dB of dynamic range (with 65,536 steps), but if the music only uses a dynamic range of 5 dB, it means that only the finer steps of the lower bits are used - this is 2-3 bits out of the 16 available. If the dynamic range of the music is very low (e.g., 5 dB), then 4-bit is more than sufficient because the dynamic range of this music lies within the 24 dB of 4-bit. Quantization noise or stepping will not be noticeable because the music simply doesn't have a larger dynamic range. Summary 16-bit audio provides 96 dB of dynamic range, but with music that has low dynamic range (e.g., 5 dB), only 2-3 bits are used because the volume fluctuations in the music are minimal. The other bits are not utilized. In 4-bit audio, the dynamic range is reduced to 24 dB, which is more than sufficient for music with a small dynamic range (e.g., 5-20 dB). So, it's not about higher bit depths offering finer resolution when the music has a low dynamic range. The higher bit depth is only relevant when the music actually requires the full 96 dB dynamic range or more. If the music is recorded with only 5 dB of dynamic range, then the 16-bit resolution is just as unnecessary as the 4-bit resolution because the music only uses a very small portion of the dynamic range.
This is not right. If you record and playback a sound that is 4k on an 8k system you are subject to huge aliasing issues. There is always a filter to roll off the top frequencies so they don't reflect down
Yes, that is correct. You do need to use an anti-aliasing filter, so technically, you need a sample rate ever so slightly higher than double the highest frequency you want to capture. With modern HD converters that use oversampling for very steep anti-aliasing filters, there doesn’t have to be much of a gap there at all. I purposely avoided discussing that in great detail so I wouldn’t be throwing in too much technical information for people, but you are absolutely correct about that. That is why the standards are 44.1k and 48k, not 40k on the nose. In the early days when analog filters were almost always used, perhaps 44.1k could have been arguably been slightly too low for some filters. But today, that shouldn’t really be the case. Hope that makes sense, Justin
I have to be honest... Im recording in 32bit just because I can, after every is done I export it to 16b anyway... And I really never saw any use in using more than 4x oversampling in some plugins. If someone is using x8, x16 or higher... I'd really like to know in wich cases do u use that much of oversampling?
I use more than 16x only in clippers on my master bus. Not that I can hear it, but at 16x I can still measure the aliasing distortion from high frequencies at around -90dB, and at 64x it goes down to about -112dB. It's probably a minor form of OCD on my part more than anything else, but knowing that the aliasing distortion is below the -96dB noise floor helps me sleep better. Ok scrap probably, it definitely is OCD.
@@kelainefes Yeah to be fair I can see the reasoning behind it. Better safe then sorry 😁 But from my side I never really use clippers on a master, I tend to do it on Bus level... So I can clearly hear the difference switching from native to x2... x4 that depends on the material... Fabfilter recently introduced x32 oversampling... I can't wait to play my tunes to superman so he can hear that buttery smooth top end at 2MHz 😅😁
@@DaveChips My reasoning for a soft clipper before the limiter is this, I use it to lower the highest peaks so that the limiter has to work a bit less. To my ears, this reduces the loss of punch from kicks and snares compared to achieving the same amount of peak reduction with just a limiter. Ofocurse I always A/B with and without the clipper and sometimes I don't use it.
This is super interesting Justin. It means I can start a 44khz project in my DAW instead of a 48khz one and use the CPU and RAM saved for distortion, analog emulation, etc to make it all sound more interesting. My main concern would be the Nyquist frequency: when using certain plugins that may not have a resampling option, working at 44khz would be worse than doing so at 48khz as some artifacts would be more noticeable and hard to clean up? Thanks a lot for this episode!
Oversampling is needed only to reduce the aliasing caused by plugins that can introduce harmonics, so saturators, clippers, brickwall limiters, fast compressors etc. Working at 48KHz doesn't really get rid of aliasing at all, you always need oversampling to do that anyway: if harmonics are introduced to a 15KHz sine wave, in instance, the harmonics produced will be 30KHz, 45KHz, 60KHz, 75KHz, 90KHz, 105KHz etc and that's just up to the 7th harmonic and 192KHz would not be enough anymore so you need to have a sample rate that allows those harmonics to be correctly described. Whatever harmonics go beyond the Nyquist frequency are mirrored back below the Nyquist frequency. In instance, if you're working at 48KHz and you use a non oversampling saturator on a 15KHz sine wave, the 2nd harmonic will be 30KHz, and it will be mirrored back and appear at 18KHz (30KHz - 24KHz = 6KHz, and then 24KHz - 6 KHz = 18KHz).
So, I've done the sine wave test, but one thing that I just don't get is the square wave hearing test. From my understanding, a square wave generator is just adding odd harmonics in ever decreasing amplitudes. So, if that's the case, and with what I can hear from a sine wave, the square wave should start sounding "siney" at (at best) 5.5-5.7 KHz, though when I listen to one, the tone changes quickly starting at 11 KHz or so. Is there someone here smarter than I (not difficult) that can explain what is going on? I mean, from what I understand, it seems like frequencies I can't hear are affecting what I can hear. I think the one thing I disagree with most is the assertion (not yours, but you mentioned) that certain analog gear can't handle the frequencies of the high sample rates. I would imagine if there was a problem on mixing, then the problem would have been there while monitoring on the way in. I also can't really think of components that would suffer from the higher frequencies other than maybe a cheap and dirty op amp, but even those have been rated 50KHz+ for a very long time now. As far as the content, I generally agree. Music is dense enough in the audible range that there is just too much going on for us to hear a meaningful difference other than "well mayyybe." I mean, possibly singular sounds higher sample rates would make more sense, I don't know. I thought I've heard differences before, but I figured that was more on (like you said) getting the filter out of the way if it was an old/cheap/crappy converter. That said, I do default myself to 88.2/24 recording because of at least perceived differences I've heard in the past, but have not scoffed when someone asked for 44.1/24 or whatever. And when I buy music, I go CD quality or better (physically, I have switched to buying vinyl whenever possible). Storage is cheap and processing power is ridiculous now. Good stuff; I've been going back and forth with myself on high resolution audio and audio formats in general since...well, forever. I hope this brings about good discussion and hopefully some other people on the other side of the aisle.
i don't seem to hear anything above 16khz. when i was a boy having equipment that had frequency range up to 12khz was quite good. and everything sounded just fine - percussion etc. for example a sharp boombox. or a turntable with a built in amplifier. they sounded just fine.
I have been using yt music for a long time now, and I saw TIDAL commercial about "high res audio" you probably already know this. I took all the audio gear I had, all the headsets and speakers and I compared and tested.... No difference expect that yt music is louder and had bigger bass. No doubt that TIDAL has studio audio quality but I don't hear a difference.
Genuine question. I heard somewhere that 96k is a requirement for film sound/music. 1. Is that true? 2. What benefits (for film) does it give to have audio in such high resolution?
48K tends to be the standard in video production that is used most commonly in my experience. This is mostly because the math works a little bit better between 48K samples and 24 frames per second. That way, you always have an even number of samples for each frame. That’s pretty much why that standard was adopted. Hope that makes sense!
These discussions are always based on the level of human hearing. Just because we can’t hear greater than 20k hz doesn’t mean it isn’t there... and if the sample rate can’t handle it, it will mess with other frequencies and get into the audible range. What’s the frequency response of your tweeter? Probably higher than 20k hz... this then effects the other frequencies that tweeter is producing. I’m new to this though, tell me what I’m missing here?
Hey Cris, I get what you’re thinking, but it’s actually a little bit of the opposite in most cases, surprisingly enough. Removing excessive supersonics won’t mess with the frequencies down below (you can test and confirm this for yourself with simple null testing). But having excessive supersonics CAN actually mess with the signals below by introducing “intermodulation distortion” to the playback circuits. Yes, there’s a chance your tweeter and analog chain can reproduce 50k fairly flat if it’s really really expensively designed. But if it’s not, then pumping a ton of 20k-50k+ through them that they were never designed to handle can actually cause distortion than can fold down into the audible range. It’s something a lot of people don’t consider, but it’s a true possibility of ultra high sample rates. Hope that makes sense! -Justin
@@SonicScoop Yes that makes perfect sense! So it’s not that it can’t handle those frequencies, it’s that it doesn’t pass along those unnecessary high frequencies, which can cause problems themselves. Thanks for the reply and clarification!
There is something I don't get. Take any audio interface that allows you to record at 96khz and TLM 103 mic. How can you actually record 96khz when this mic goes up to 20khz? isn't there any placebo effect?
No, people always use Nyquist-Shannon sampling theorem to say you only need to sample twice the highest frequency you want to record i.e. in Audio you only need 40kHz. But did not understand the whole theorem. The theorem also said you need to multiple every sample to a infinite function call sinc function to be able to recover the same wave form. The important word is "infinite function". It is impossible to calculate infinite function because that would take infinite time. So you always need to stop some where just like when you use Pi. Higher sample frequency will make calculation many times near to "infinite function". So same DAC chip with same computing power suppose to have easier time with higher sample rate in theory.
Well illustrated, on top of all that we should consider the frecuency response of microphones (good old SM57 does not go over 15 kHz), preamps, etc. Just saying,
18:40 Sorry, but that "battery draining 10-20 times faster" is way overblown. Practically every modern multimedia capable device decodes audio and video on dedicated chips, and as long as that hardware is compatible with your audio format the difference in energy usage will be negligible. Though to be fair when you stream the stuff it'll be more noticeable. Otherwise I agree though. Usually I try to prefer FLAC simply because memory is cheap(unless it's from Apple lol) and it often doesn't make a difference in price. Even if it's just to be able to losslessly convert the format if needed.
Isn't a higher sample rate with more values resolved by a high bit depth and lower step sizes represent a better reproduction of the original continuous signal when optimized for high accuracy? I don't see an issue with people knowing for sure they are listening to the most faithful reproduction of music especially if they are willing to pay for it.
Nope, that’s what I thought originally as well. But that’s not how it works. In reality, there are no “steps” when you play back digital audio. There is only a continuous analog waveform, identical to the original, except that you have to eliminate frequencies above a certain point, and there is some additional noise 96 dB or more down, where no one will here it. It’s a common misconception, and one that this video hopes to correct. Counterintuitive I know, but that’s how it works. Very best, Justin
I'm 55 years old, and greatful that I can hear 14 kHz
I’m 55 too and glad that I hear around 15kHz ( left ear 14+kHz Rt Ear) 🙏👍
I'm 43 and I'm glad I'm fine at 14 khz as well
At this point, we just have to put on a high shelf and hope for air that isn't killing the young 'uns :)
Yeah I think mine cuts out around 13khz
Roughly same age, 57, and about 14kHz too (actualy 14khz for the right hear and 13,5 for the left one).
It's so refreshing to watch this.. logical, sensible, reasoned advice. Thank you!
Audiophile nonsense has entered the chat. If your amp didn't cost more than my house, and if the knobs aren't made of bubinga, are you really trying hard enough?
Audiophiles are like flat-earthists, they don't believe in phase cancelation and when you talk about a null test they say it was filmed on a set by Stanley Kubrick 😁
So sony is just a fraud? Apple music also? Ok im gonna throw all of them in the garbage can. Thanks to you.
@@iwancarnifex No and Yes. First, with Apple Music you get now Lossless Music in CD Quality. Before Apple music only offered AAC. Yes You dont need to listen to Highres Music but CD Quality is an improvement. Second just because a High-Res Headphone makes no sense doesnt makes it worse for listening to CD quality music.
All my music is “mixed for bats 🦇”... “can I have a dollar now?” 😎
I noticed when I converted some of my masters to mp3 the mp3 file sounded more scratchy bright than the same track with a 24 bit wav. It was noticeably different in the high frequencies.
After some trial and error I eventually fixed the problem by applying a high cut filter to the inaudible high frequencies in the track. From this I surmised that these inaudible frequencies were being mapped to audible frequencies by the mp3 codec.
I’m not saying that high definition audio is superior because it handles frequencies, but rather curious why this happens? Why does the mp3 dithering not handle the mapping more gracefully?
Probably a bug in the mp3 encoder, maybe they are resampling the audio using a naive averaging algorithm
MP3 uses some psychoacoustic tricks, it essentially always loses details not only bc. of the high cut filter. First of all, the Version of the Converter matters! MP3 Converter Software has been improved dramatically over the years, but also the Bitrate is extremely critical. When you really want to use MP3, and i personally am completely have abandoned this format more then a decade ago, except when i want to share a song and don't have either enough bandwidth or Storage Capacity. Beside that, there's no point of using MP3 any longer in 2024! It was invented when the Internet still was on 64 Kbit DSL and you had to pay a ton of money for a subscription plan. Simply switch to FLAC. It has all the comfort of Tagging whilst being completely transparent and it saves you 40 to 60% of Storage Capacity. Beside that, you can ALWAYS retrieve tthe Original WAV. File back and burn a new CD for example.
If you're a music listener, you're not going to hear an improvement going from 320 kbps MP3 to lossless FLAC. Save the hard drive space and invest in better equipment (headphones, amps, speakers, etc.)
Exactly correct. As a digital signal processing expert, I deal with this situation every day. With modern antialiasing filtering, reconstruction algorithms etc higher sample rates does not matter. 16 bit vs 24 bit resolution only changes the effective signal to noise ratio the systems can support. 16 bit supports 65535 slicing levels. 24 bit supports 16,777,216 slicing levels. Hence 16 bit supports 96 dB SNR where 24 bit supports a SNR OF 144 dB.
And then add noise shaping to that dither and the effective SNR of 16bit goes up to 120dB.
Get yourself better listening devices. stop using cone and dome drivers with spindel spring coil and rubber etc suspention = high moving mass slow stop and start. .Go for stax lambda best or their top model 009 as far i know. i have stax lamda signature mk2 from 92 No dynamic earphone has been able to beat it. still working fine. real earpads and go for a prize that is more than when i got them new still.
tweeters air motion transformers are the way(AMT) and planar magnetic mids.with sensitivity at 96 db pr1w 1 M smashes most old school type drivers
Modern antialiasing filters still cause audible phase shift at 44.1/48 kHz, that's why 24/96 is considered minimum for archiving - phase shift from the low pass filter is less audible.
I will say though, one advantage of using higher bit depths is that it can save a take if you realize (too late) that one of the mics wasn't amp'd enough, and you can't take another take, having that lower noise floor, from what I've seen anyways, allows you to really amplify the signal in the mix without getting any noticeable noise.
But admittedly, that is a pretty specific usage, that happens because of user error to begin with.
(edit) the whole thing about audio being better when treated less preciously is precisely why people seem to prefer vinyls & tape, even in today's day and age, I work in digital graphic arts, and it's really the same, often in digital artworks you want to add 'noise' and irregularities into artwork, that usually always makes it pop more, and gives it character
Yeah, if you were to record a sensitive source at something crazy low, like -60dBFS or lower in 16 bit you could MAYBE run into noise floor issues.
That’s an insanely low record level, so it’s unlikely to be a problem very often. But because of this unlikely possibility it is definitely a best practice to do 24bit when recording instead. That seems reasonable to me.
But even 24 bit is actually overkill for that, because none of your analog front end can use 24bit’s 144dB of dynamic range.
20 bit probably would have worked just as well for this purpose, because that’s about where really nice analog gear is going to max out for keeping the noise floor down, but it’s no big deal either way.
For consumer audio though? No way! :-) Even 16bit is probably overkill for all practical purposes.
Hope that makes sense!
-Justin
Exactly. When you use 24 Bit your A/D Converter has a S/N Ratio of 146 dB instead of 'only' 96 dB with 16 Bit and that certainly gets rid of the noise floor.
I always record in 48 kHz/24 for two simple reasons.
1. All my gear supports 48 kHz, so whatever interface I choose to use it’s working with 48 kHz.
2. If the recordings will be used in a video I will have a perfect sample match. If you use 44.1 kHz with a 24 FPS video each frame will have 1837,5 samples, so if I cut both the video & audio when the grid in PT is set to frame the cut will be between two samples. This is avoided with 48 kHz.
So, that’s my approach, I use 48 kHz for practical reasons, not audible.
Valid reasons for sure!
Same for me, I work in 48 for practical reasons, the TV/media business works with this sampling rate and expect you deliver your tracks/stems in 48k/24b.
If you guys are that kind of person whose doesn't hear any difference between "Apple Digital Master" versus CD, you may be definitely deaf. Even from the former Mastered for iTunes quality, from the same AAC quality it DOES A BIG DIFFERENCE over CD quality. Why the people doesn't notice any difference? The MASTER it does matter at all.
I record in 47.924khz at 23 bit, send it to you and dont tell you about it
@@JnL_SSBM Yeah. That's the masstering, not the sampling rate or bit depth.
The fact that you had no uptake on your listening challenge from 10 years ago is very telling.
Excellent, excellent video explaining the trivial quest for high res. I've played with various rates in my DAW and have not heard differences. Now I understand why. Thanks, Justin; my hard drives appreciate you....
It's worth repeating, as you mention, that capture and delivery are very different things. For example, Zoom just released a sound recorder that records 32-bit float (the F2 lavalier recorder). The huge benefit of that is that it's impossible to blow out the recording as it doesn't peak at 0dB, and levels don't need to be set. You just hit record and you're golden. Once the sound is captured, that 32-bit float just becomes 32-bit bloat (you heard me). Recording headroom is just recording headroom. The end listener doesn't need to hear the empty headroom, and won't benefit one iota from doing so.
You are right. There's not much a point for super Hi res other than selling really expensive gear to people who don't know any better. The one advantage I'm seeing is you have a lot of very affordable devices coming into the market which offer tremendous quality. I'm speaking of desktop dacs costing less than 200$. And current audio is getting way better than it was 10 years or 20 years back
Neil Young and Robert Fripp are suiting up for battle.
Now FRIPP on the other hand.....
They're delusional.
Yeah, and here are two conceptual things that are often misunderstood:
1. There are no stair steps. Digital audio is entirely smooth, continuous, and analog once it goes through the low-pass filter. That’s the whole point of the filter, is it turns everything at the upper frequencies into pure, smooth sine waves, filtering out all the squareness (which are just the upper partials).
2. It only takes two samples to accurately reconstruct any given sine wave at or below that frequency, perfectly in terms of frequency, amplitude, and phase. Sine waves have a particular shape such that you can mathematically reconstruct the whole thing just from the two samples.
Bingo! You are exactly right.
Either a given frequency can be reproduced perfectly at a given sample rate (minus some noise, which is determined by the bit depth) or it can’t be reproduced at all. It’s a bit of an all-or-nothing affair.
Thanks for the comment,
Justin
@@SonicScoop Thanks for the video! Clearly explained, and your 10-year challenge is exactly the kind of real-world experiment I keep using to point out how people really can't hear the difference (though I wasn't aware of yours until now).
@ReaktorLeak Great example!
@ReaktorLeak Thank you for this analogy👍
What's misunderstood is anti aliasing/mirror effect requiring low pass filters which he mentions in this video, but doesn't go through much detail. If you pay attention he is staying that the mastering process could benefit for higher frequency rate on the recoding session in terms of having a easier time filtering high frequo based on equipment used and recording in higher bit rate gives you much more headroom so there is less likely clipping while recording. I think the majority of people missed the boat on this. In other words, if you actually record in a higher bit rate with higher frequency rates where higher frequency noise with high enough amplitudes could enter the mix and a higher chance of clipping, it would make a difference.
I'd like to point out a big missconception here: quality matters if it's higher or lower sample rate, the lower sample rate the lower the quality that is a scientific fact judging by how the quantization works. The big point that all seem to miss is that beyond 44.1kHz one would not be able to tell the difference and basically is pointless to have so much storage space spent for nothing.
Exactly! 22K is pipe dream for most people; most adults hear in the 10 - 16K range, dependant on age, and the music doesn't sound 'different' just because we've lost a few KHz here and there! If cymbals still zing, and ting like they should, then excessive high frequency recording is a waste.... I know in some circumstances that there are higher frequencies that work with others to give a certain order of distortion that resonate well (certain harmonics) that if missing would destroy the feel of the music and maybe even notes being played, but these are all recorded within the 'lowly' 16/44.1 redbook solution, so it's not an issue!
As you say, storage space for needlessly high 'resolution' files is just pointless!
I love good old analogue, but I also love my (well recorded/mastered) CD's... :-)
Thank you for interjecting some reason and science into this area. On a somewhat related note, I've noticed that many people equate "pristine" audio with detail. But audio that is very revealing of detail is not necessarily perceived as musical or all that pleasing to the ear. (I came across at least one blind study that arrived at the same conclusion).
True. Witness the love of the sound of vinyl - which has a smaller signal to noise ratio than 16bit digital audio, has more stereo crosstalk, and changes its frequency response and harmonic distortion characteristics according to how far towards the centre of the disc the needle happens to be.
Not pristine at all by comparison to digital audio, but people love it.
@@latheofheaven1017 Great points.
When he played the 15k sound at 8:23 I thought there was something wrong with my headphones
Same lol
Very good summary, I totally agree!👍 Especially concerning our golden time of audio quality! Unfortunately I can't imagine what we can expect from future developments🤔. The great leaps in improving audio quality are behind us. But luckily we are now able to concentrate fully in our creativity without be concerned about quality - even at home! I remember recording on noisy tapes in the 80ies. Everything today was science fiction at those times (a DAW running on an tablet!!!)
I really appreciate you taking the time to explain this in a way a layman like myself can understand. I was finding myself obsessing a bit on trying to get these higher bitrates and it was frustrating. I am at peace now with how things truly work. Especially someone like me who listens to music on bluetooth!
I was running this video in background, and that 15kHz startled me, I thought something was wrong with my head. Anyways, it was mathematically proven that you only need double sample rate of what your maximum recorded frequency is, and that would give you accurate reproduction. Any errors in audio are not likely and negligible. Search for Nyquist theorem. 44.1 sample rate can be used to reproduce frequencies up to 22.05kHz. Only a few people, if any at all, can perhaps hear that. Noise for dogs.
Man, I'm 34 and I still use a tascam 4 track cassette recorder and a Alesis 3630 compressor for my music, and people that hear it don't tell me, nobody will like your music because it was recorded on a tascam cassette recorder and a Alesis 3630 compressor
Becuz: the resolution of good quality tape and agood cassette machine and analog outboard gear is better than a lot of digital stuff. Especially when you start using plugins.
Great point, also if the music is great all the techy things don't matter as much. LIke, what is the best mic? The one with a real artist in front of it. WHIle i've heard some crappy recordings done in garageband. I've heard some great stuff as well. Its more so the music, and the person turning the knobs.
Thanks so much for clearing this up. A while ago I did the 320kbps mp3 vs wav test and could not hear a difference at all... With high grade converters and headphones! I thought my ears were to blame.
You are not alone! Being human is to blame, not your ears :-)
Good on you for actually doing the blind listening test. A lot of people with surprisingly strong opinions on this never actually do that.
Listening to Spotify premium 320kbps Vs CD quality music (Tidal) is a big difference in musicality. I can EASILY hear the difference between the two. With lower end equipment though that detail that you gain may not be as obvious to your ears. Going back from Tidal CD quality and higher to Spotify premium actually sucks it's so noticeable
@@zachunter2357 Hey Zac, you must have excellent ears! Maybe it also depends on the mix. I will redo a comparison to double check :)
@@georgearrows7701 also remember that the sound coming out from your headphones/speakers is going to be affected by everything that it goes through between the source and the end point. So if your pc has a shitty DAC it will lower the quality, a worser power source will also affect it. And also, the mastering of the track will affect it too. Start a free trial on a hi-fi service and listen to some different tracks. Do a side by side comparison
@@zachunter2357 That's the thing. I am using a pretty expensive semi-pro interface (RME Babyface Pro) + pretty good headphones (Slate VSX) and don't really hear it. Anyway, I will do the comparison again because this has been bugging me for years.
I think there is a better low bitrate formats than the mp3 like m4a aac, ogg, because the spectral analysis sometimes shows that some mp3 converters screw up the songs by cutting too much frequencies. So you still get 320kbps but in reality it's less.
Another well-reasoned and delightfully delivered info download! Thanks, Justin!
Try as I might I have yet to be able to hear the difference between cd quality and the high res audio, which could explain why sa-cds never took off. I’ll have to try the MP3 320k to see, but I know that at 256k I can tell but it’s really difficult.
Absolutely BRILLIANT tutorial! Thank you so much for sharing this knowledge!
Very impressive presentation. Really well done!
The other thing is that although we do have the best audio formats available to consumers now, they don't listen to music in any way to take advantage of them. Phone speakers? Laptop speakers? Alexa speakers? People aren't really even listening in stereo.
It’s true, people can listen on some pretty garbage speakers these days! But on the other hand, if they wanted to, really great sounding speakers and headphones are available at a much better fidelity and a much lower price than ever before in history. That’s kind of awesome.
Let’s not romanticize the past too much either though. Back in the day, most consumers also listened on total garbage speakers and hardware, often as bad or worse than the laptop speakers of today :-)
But people have very fond memories of listening to music on their terrible tiny old transistor radios, or even today on their phones and laptops or even a single shared earbud with friends.
As cool as good audio is and can be, music is ultimately way cooler and way more powerful.
Hope that makes sense,
Justin
@@SonicScoop Hey Justin. You're right of course. Transistor radios were shrill and distorted. I inherited a basic little reel-to-reel machine when I was about ten, which only had one speaker (even though it was a stereo tape head). But then we also had a radiogram! The turntable in it ran at 33rpm (or less), not 33/3rd. I only realised when I took my albums over to a friend's house, and at first thought their machine was running fast. But everyone else's was the same, so it was our radiogram that was wrong.
I would somewhat argue against that, I would say that a lot more people have headphones that sound good, compared to how many people had good sounding headphones or stereo systems in the past. So while some listening, does take place on sub par playback equipment, a lot of people actually also listens to equipment that is far superior to the typical listening equipment just a decade or more ago.
However, that is an issue, that mixing and mastering focuses so much on speakers rather than headphones, as most listening with quality equipment these days is actually with headphones, compared to the past when stereo speakers were the dominant listening environment of quality. This causes a lot of issues with the stereo representation, where speakers provide a channel crosstalk that doesn't exist in headphones, and the fact that we can't tell the direction of low frequency sounds in a room, but if if is only sent to one headphone we can.
Brilliant. Anybody interested in music should watch this.. and understand it.
I'm going to be getting theoretical here but what the heck; I get aggravated, too. Much of this presentation is based on a common misunderstanding of Nyquist's sampling theorem regarding the reconstruction of a signal that has been sampled at anything strictly more than twice the highest frequency present (the "Nyquist rate"), and a common misunderstanding of the frequency content of audio as based on Fourier analysis. Nyquist reconstruction only works if the signal has a limited frequency spectrum but, according to Fourier, any signal that is of finite duration has an infinitely wide frequency spectrum, thus Nyquist reconstruction does not apply to music. Even a pure sinusoid of limited duration has an infinite frequency spectrum (I suspect many audio folks don't realize this); only a sinsuid that started at -infinity and runs to +infinity has a frequency spectrum that looks like a spike. Just imagine any three evenly spaced samples of a single cycle of a sinusoid, around the Nyquist rate. According to the video, you can reconstruct it but there's not a chance. Further, Nyquist's reconstruction requires the availability of every sample to reconstruct any point of the original waveform. So, even if the sampling rate wasn't an issue for a signal that somehow had a limited frequency spectrum, you couldn't reconstruct anything in real time from a digital input stream. You'd have to wait for the entire song to download before you could even start playback. Our digital audio realities are fine but I get aggravated when I hear theories that don't apply being crushed to explain how things work. And while I agree that our hearing just isn't good enough to benefit from high resolution audio, I think a distinction has to be made between (1) what difference high resolution processing makes when it's used throughout the entire recording chain, and (2) what difference high resolution audio makes just for the final output. I suspect high res makes a significant difference in the first case especially if a lot of processing is used, if for no other reason than cumulative errors (resulting in unwanted artifacts) are minimized, but the evidence is in regarding the second. We just cant hear the difference.
My dog has been upsampling my music. I am pissed.
Justin, loved it.
Three questions:
. A famed producer (I believe it was Aerosmith's producer said that excessively high resolution audio was allowing us to hear details in the original recording that weren't there and therefore were a bad thing. True/false?
. Jimmy Page said the Zeppelin catalog was recording at such fidelity it's future-proofed for years (or more to come). Page specifically said for future "higher audio standards," which as they're in the future don't yet exist. True/false?
. I thought the sound on VHS was/is pretty groovy. Isn't VHS tape a NON digital format? Hence, the sound is 16 bit which as you noted no one can really distinguish from 24 bit. Hence, analog can reproduce in the instance of VHS sound as close to the source as digital? Not a trick question.
P.S. I think one big misconception you blew-up in this video, is that if you record say, a giraffe making a sound at a zoo - whatever sound they make at 16 bits and you record it at 32 bits - most folks, including many professionals, believe that at 32 bits of 'resolution' you'll hear greater fidelity with 32 bits. You'll hear a better sounding giraffe, as well as the sound of garbage rustling on the ground. A better wind sound. Maybe human speech, better, from visitors to the giraffe enclosure.
Yes, I think those first two examples come from a misconception about how this stuff works.
There are a lot of people who are great musicians, producers-and in some cases, audio people-who don’t really understand the situation fully.
They usually have never done double blind trials on any of this, and often don’t exactly understand what’s going on under the hood. But that’s ok. They can be tremendously great at what they do and be mistaken about this. I was once.
Jimmy Page is an amazing guitar player. That doesn’t mean he knows how digital audio files work, or that he’s done blind listening tests at various sample rates or bit depths.
That’s not really a criticism. You don’t need to know any of that to be a great musician, or even a great producer or mixer.
If you had to pick between being amazing at guitar or production and knowing how digital audio works, the first too are way more badass :-)
VHS CAN be digital (see ADAT) but is usually analog yes. (Technically you could record digital signals on a reel to reel take machine, and some companies made such machines early in.) But yeah, generally analog in both cases without getting too nerdy about it :-)
Analog formats actually have lower dynamic range and higher noise floor than 16 bit. Good vinyl would be the equivalent of about 11 bit in that regard if I remember correctly. I’d imagine VJS would be around there or lower in effect, but I don’t know right offhand.
Generally speaking, any analog formats are going to sound less like the source than a 16/44.1 file. The only thing that comes close is an absolutely excellent tape machine running at 30ips. That might come close to sounding as close to the source as any reasonably well built 16/44.1 system, and would be orders of magnitude more expensive.
But analog can sometimes sound damn COOL, even if it doesn’t sound exactly like the source. A less than stellar tape machine running at 15ips could sound more badass than the original source. That’s subjective. But it probably won’t sound as close to it as even a basic stock AD converter built into a laptop. That’s objective.
Hope that helps,
-Justin
@@SonicScoop Thanks for such a thoughtful answer!
You tube sound seems to cut off at 15kHz. I tried using sound tones n the same way as you did from an internet source in class only to find that those higher frequencies are not available.
Higher sample rates introduce less latency when monitoring natively via the DAW. Once tracked at high sample rate for low latency, there is no need to introduce sample rate conversion, which will cause degradation.
Thank you. Thank you. I’ve been dealing with these toks my entire life.
One benefit of using 96 instead of 48 at recording is the AD converters allow me better headroom.
Not sure how to explain this in technical terms other than what I see when I clip the sound at 48 compared to 96.
Another benefit of higher sample rates is signal latency.
Although I've never considered 96 being “super high resolution.”
Headroom and dynamic range is a function of a bit depth, and of the analog components of your converter. Sample rate just doesn’t play into it, in theory or practice.
If your converter works differently at the two sample rates for some reason, that could be possible I guess. I’d like to know why. That seems like an odd design. But that seems unlikely. If anything, it could be slightly less easy to clip a lower sampling rate because you are dealing with less signal overall because the reduced supersonics and lower anti-aliasing filter.
Have you done any properly controlled tests of this where you can absolutely confirm this is the case? Please share if you can! I’d be curious to know more.
As I’m sure you sneaky know, we all think we hear things that we aren’t really hearing. It’s happened to all of us. I’m sure you’ve had the common experience of fiddling around with an EQ knob, thinking you were changing the sound subtly, until you realized it wasn’t engaged, or you weren’t on the right channel. That’s just part of being human.
As powerful as we can develop our listening skills, it is true that our minds will always be more powerful still.
If you have a test you can share that confirms this, I’d be very interested to see it. Thanks much!
Hope that helps,
Justin
@@SonicScoop Many thanks Justin!
I noticed it when running outboard reverbs both at 48 and 96.
At 96 I had more headroom before the signal clipped in the converter, RME ADI-8 DS Mk3 via ADAT to/from Apollo 8.
So it was purely by eyes I noticed about signal not clipping at a certain level.
Nothing else was engaged or changed so I was a bit surprised myself.
Probably more to it than headroom being changed then : )
I'm like you-being aware of the sound while fiddling and thinking yup that's a tad better-and realising the plugin wasn't active : )
I have that perspective when listening to anything these days but trying to explain how the brain fools us is a difficult task but I'd be happy if you talked more about that, blindtesting, do we actually hear the difference we think we hear etc etc.
Outstanding info and delivery!
I listen to MP3s on an almost 20 year old Zune with Beyerdynamic studio monitor headphones and the sound quality is flat out amazing. 44.100/16bit is more than enough to capture and reproduce sound that humans can hear and a good ripped Mp3 sounds nearly identical.
My argument for using 24 or 32 bit float is just headroom, I can record really quiet sources and then gain them up and compress the living shit out of them and the noise doesn't come into that audible range. For consumer side I go 48/24 because it is standard for video.
I wish I knew half of what you talking about. Really just wanted to know if I wasting money with tidal Master plan. If hi - fi would have been adequate. Or if I'm just imaging Spotify not sounding good from my system.. Last two mins most useful for me. Listen to what works for you. Sound advice
Years ago, I was tracking thru 18bit motorola converters. I was dropping the teac four Trac drum tracks down to digital. I noticed it sounded much more like the tape in 24bit than sixteen bit my sampling rate was 44.1 in both cases. I learned 24bit is better for resolution. However I had to mix down to 16 bit digital in this case. My hardware could only handle 16 in the multitrack window.
I use to bounce my music from Cubase (6) to a 44.1kHz/16Bit wave file, then drag it onto my iPod mini, go out to smoke a cigarette and gaze into the distance. It occured more than once to me, that I was bouncing out an 320kBps MP3 and was not aware of it. The funny thing is: That iPod makes it a blind test without my provision - it doesn't show or tell you in any way which format is being played back. BUT: My nearly 40 years old ears seem to reveal it - whenever I listen to one of my productions and I have the feeling it's not uncompressed sound... guess what: In fact it would be compressed, really. I would yet have to fail on that one to conclude an MP3 sounds just as good as a PCM. But don't get me wrong: I agree wholehartedly to what you say in this video.
First time watching your videos. So relatable. I’ll watch all today and all my family members and neighbours will too 😂. Thanks a lot
Here is a blind test that will change your mind, Justin. Listen to a 320k vs. a wav or flac (create both from the best same source) of The War on Drugs' "Lost in the Dream." I have, and for a fact, I can tell you there is a specific character that shows the difference. I challenge you to notice it, it can be deciphered in the first minute of the song. Another stunning mix was Morning Phase, equally decipherable. The sense of detail and SPACE that is destroyed by 320k is obvious to me, at least.
Most won't notice, so don't stress, that is true. But when you personally want the best, why not go ahead and mix for that. Try to hear it, try to make it.
Interesting to hear! What are you using to listen to the these properly double blind?
I can tell you that I have had similar experiences in sighted listening tests.
For instance, when I created a similar listening test for our readers, I could have sworn up-and-down that I could hear the difference when listening sighted, and that so many people were going to get it right.
But as soon as I properly jumbled them up so I didn’t know which was which, those differences seemed to evaporate like a mirage.
Ultimately, the results of thousands of responses to our online test were no better than chance:
www.trustmeimascientist.com/2012/04/02/take-our-audio-poll-do-we-need-higher-definition-sound/
If you can consistently distinguish this, that would be amazing! No one in the world as of yet has been able to show that they can consistently distinguish between 320kbps and any higher resolution format in a proper double blind listening test.
I’m not saying that it can’t be done. Just that it never has been yet on record. If you can do it, that would be amazing. I’d happily write a glowing full length article about you and how you are able to do this so we can share it with the world.
I’ve had this challenge out for almost a decade now, and so far no one has done it yet. It would be awesome to finally put a bow on it.
What software or hardware solution are you using to do proper double blind listening tests on your end? Is it an ABX tester? That could do the trick.
Thanks,
Justin
PS: Here’s the original challenge on my old blog. It’s been seen tens or hundreds of thousands of times, shared all around forums and all that, but no luck yet! Be our champion :-)
www.trustmeimascientist.com/2013/09/03/think-you-have-golden-ears-take-the-scientist-challenge/
www.trustmeimascientist.com/2013/10/07/update-on-the-golden-ear-challenge-who-will-conquer-audio-mount-everest/
@@SonicScoop Well I went, prepared to be humbled, but the links for Line 'em Up and Kite of Love are broken.
I did find in another of the articles the foobar can have an ABX tester plugin, I'll do that. My previous tests had my wife switching the tracks; a program that does that will be fun, so thanks very much for that.
Also, please know I meant no disrespect, I love your channel and learn what I can from it :)
320k MP3 isnt a "hi res" format - its still a compressed (althogh half decent) file. Im pretty sure Justin is referring to higher sample rates and bit depths.
i agree about your main point, but i can definitely hear the quality difference between full cd and 320kb mp3.
Great video. I’ve had the same concerns myself and you properly articulated why. The best ever hifi upgrade I did was buy a Nord One Up amp. This was a game changer for my PMC OB1s speakers. I have heard differences between formats but not in ABX tests. What I mean is the Linn 96/24 for example sound great but think the real reason is the the audience for these are people who care about music and the producers bother to mix the tracks well in the studio. i.e. don’t dynamically compress the music. Place instrument well etc. It’s not the format it’s the production. Madonna immaculate conception on cd sound just as “interesting” and alive as any 96/24 recording do.
Love your content bro. I'm 28 years old and can hear clearly from 30hrz to 22khz. I tried many app from play store and that's my hearing rage. However that higher frequency I perceive them as noise.
@StringerNews1 scientists lie too.
Ok… sound engineer here. While the maths and physiology is correct, you’re missing a couple of critical details:
Due to a physical limitation, DACs don’t perform optimally at 44.1KHz, or even 48KHz. Clock a DAC at double that rate and it will perform a higher fidelity conversion from digital to analogue. The result is clearer more transparent sound.
That said merely possessing a 96KHz audio file is not sufficient to unlock that performance… for starters the audio has to be read in to your DAC at 96KHz. Now, even if your streaming providers offer 96KHz audio (and some do), if you send that to your transport over Bluetooth or AirPlay, then it’ll be downsampled to 44.1KHz… does your transport upsample it on the fly up to 88.2KHz or better before talking to your DAC?
Unless it does, then you need a source outputting audio at 88.2KHz or better, with a physical cable running between that and your transport... or better yet just skip the whole transport stage and plug your source device directly into your DAC!
Here’s the cool thing though, the original recording doesn’t even have to be at 88.2KHz or better - you can load any old 44.1KHz wave file into a DAW, then re-render to 88.2KHz or better, and that’ll work just fine. If you’re fortunate enough to be aged 15 or less, with pristine hearing, then starting with a 48KHz source file will sound even better, but for the rest of us that’s completely overkill.
So, the first point being that that any improvement isn’t baked into the audio itself, but purely due to the DAC operating in its sweet spot. To built a DAC that could do this at 44.1KHz would be prohibitively expensive, if that were even technologically feasible, and I’m don’t think it is at present.
Second point… bit depth plays a much more important role during the recording, processing, mixing, and mastering processes. I could write a whole book on this, but the long and short of it is that the audio is digitally amplified and de-amplified at various stages then layered on top of many other audio tracks, so all those noise floors quickly stack up, to the point they become very audible… therefore you want the original audio recorded, processed, mixed, and mastered at least at 24bit PCM, but there are arguments for going beyond that. As someone who predominantly works on live concert albums, I favour 32bit FP (floating point), since that completely offsets the issue of digital clipping. But… once the mastering is done, that audio can happily be downsampled and rendered to 44.1KHz 16bit, and the information within will theoretically be indistinguishable to most adults, no matter how good they are at listening!
So, in most circumstances, the advantages gleaned from playing back true 96K 32bit PCM audio at that rate, are only really useful to mixing and mastering engineers. Sure, you’d be able to hear the music more clearly (often to its detriment), but you’d really have to concentrate! We only need this level of clarity to tune the audio to what we think is perfection. For playing music I’m perfectly happy with a 44.1KHz 16-bit… unless I’m only there to listen, wherein I prefer that DAC to be running in its sweet spot… but mainly just so I can get a hard-on from hearing the processing artefacts ;)
Lastly, the biggest influences on audio quality are, in order of importance:
The artist and their performance
Their instruments
The recording environment
The mics and mic placement
The mic preamps
The ADCs
The sample rate and bit depth used to record
The skill of the engineers involved, and the techniques they employ
Your preamp, amp, and speakers
Your speaker placement
Your room
Your DAC
The audio will however, only be as good as the weakest link in that chain.
Things that don’t matter, at least not for digital audio quality:
Your cables
Your power supply
Your transport (provided it only transports… most don’t)
And if there’s any issue with digital audio, then you’ll hear it as a significant disruption of the signal, and never as some barely audible distortion.
Lastly… don’t be blowing hundreds of bucks on an audio cable unless it’s for the looks - you should see the crap we use in the studio. If cable was a problem then the 128 or so cables that said audio passed through before it even left our studio, not to mention the thousands coms cables it ran through as it was piped across the cloud before it finally entered the ones leading to your speakers, would have utterly mangled it! As for digital signals, well the audio either works or it craps out. That’s the whole point of digital!
If you copy a text document from digital device to device, or expose it to vibrations or EMF… does it ever corrupt or degrade in any subtle way? Does the text become blurry? No. You can copy it a bazillion times across every cable in the cloud and it’ll remain forever a perfect replica. That’s the whole killer use case of digital audio!
The real benefit of high sample rate is moving (most of) the phase shift from the low pass filter out of the audible range. You can hear the phase shift, right? If not, maybe mastering isn't your best career option. 24 bits gives you more s/n for mixing, that extra 8 bits is ~48dB you can throw away and still have CD quality s/n.
Oversampling fixed phase issues over 30 years ago. You don't need higher sampling rates to get that benefit.
Yes, that's one of the main points of the video! That any difference heard in sampling rate is really a difference in the anti-aliasing filter :-)
What Viklas says is also correct: Oversampling fixes this issue.
But even if that weren't the case, you may also misunderstand "phase shift".
EQ doesn't "cause phase shift" so much as *phase shift causes EQ*.
For practical purposes, you have it backwards.
It's not that you EQ something and you get EQ, plus this side effect called "phase shift". No, it's quite the opposite:
A conventional EQ allows you to apply phase shift, and the result of that phase shift is the EQ that's being added. The EQ is the side effect of the phase shift! Not the other way around.
Now, this is complicated further when you are EQing two parallel signals that have some variation between them.
If that case, you will end up with some additional adding or summing around the corner frequency of the EQ, and this is indeed more significant on EQ hi and low pass filters.
But the "phase shift" is not some crazy comb filtering effect that our minds may conjure up when we hear the term "phase". Rather, it's generally, a greater reduction than anticipated in the extreme lows or highs.
So really, the "phase shift" just creates a slightly greater amount of EQ than anticipated :-)
The additional s/n ratio for mixing from higher bit depths is 100% irrelevant, because the s/n ratio of 16 bit is already 96 dB+ X-D
Seeing that most commercial mixes are well under 14dB of dynamic range, and even in old school classical and jazz they are rarely greater than 20 or 30 dB (at MOST), this is just not a problem at the mix stage that needs to be solved.
I hope that helps! Very common misconceptions here, and they are all addressed in the video if you get through the whole thing :-)
Totally agree. In fact I could perfectly record at 16 bit and be really happy.
Most of the difference to music quality appears to be down to the recording and mastering process, in my personal experience.
While 320 kbps MP3 might be hard to distinguish from 16/44.1 FLAC... I wonder whether that's not really dependent on the hardware to a good degree. The better setup, the easier the perception of transients and hence space, distance etc., might be. I can for sure tell that the quality difference between streamin of something lossy like through Spotify versus lossless like through Tidal, is night and day, in clear favor of the latter. That is though probably also mostly due to Tidal sourcing the tracks much better.
I've been custom-upsampling (like 64M taps sinc filter using SoX) my FLACs to 24/96 (for my phone) and 24/768 (for my home setup) because it gives me a bit more refined, clearer and smoother highs, which I actually hear the difference in (although the differences are kind of subtle, they won't really hit one in the face), and that's comparing FLACs, and even that on mid-fi equipment, nothing that breaks the bank. MP3 is really out of the picture there. Sorry for not having done the double blind test experiments! Will for sure try that though.
Also, today's delta-sigma DACs and even some other DAC tech like R2R tend to oversample themselves, unless you supply them a high-enough input stream. Save for Rob Watt's M-Scaler or something in that league, the small chips of these DACs that need to do their resampling in real time can't really compete with an offline process that takes a few minutes per track on a modern CPU core and a good 6 GB of RAM to finish...
And another very clear benefit of oversampling to mention is that today's music that gets released is most often overly loud and already digitally clipped, which means informations is already lost in there, mostly impacting the higher frequencies. A proper upsampling process (after digitally lowering the volume beforehand, sometimes by as much as a whole third), can reconstruct the waveform and partially undo the damaging effects of digital clipping. Perhaps it is this that justifies the oversampling hassle to me, since the space storage and the bandwidth between my home NAS and my living room computer doing the audio playback, is not really an issue.
Thank you Justin, this comforts me in my unashamed attitude of "If it sounds good, it's good"! I have a question about the "Air Band" on Maag plugins, I really like the "brightness-not-harshness" it allows my 58 years old ears to perceive and appreciate on instruments and mixes. I usually use 4-7dB @ 20k and only recently used 15k on the EQ2, should we be cautious of unheard frequencies causing aliasing generated by the plugin at 20 or 40k?
Good to hear! Aliasing should’t be a problem and should be filtered out if everything is working properly.
But there’s a chance that to you get ears you might be boosting more HF than you realize :-)
It could be good to double check things on a frequency analyzer to make sure the super high end you have trouble hearing doesn’t look to crazy relative to references.
Either that or visit a high school classroom, draw your nails across the chalkboard to get their attention, and then ask them for feedback on your mix ;-)
Hope that helps!
-Justin
So, Justin, I see you've changed back to another JZ Mic which looks like the V67! Nice choice!
As for the topic, I really enjoyed listening to this and your perspective. I have to admit that I was somewhat of a numbers-fan and thought that bigger meant better. However, after listening to you it made sense to me in that you really won't hear the difference on the listening end of the sound. And I'm convinced that you need some really high-quality headphones to even attempt it....if you can. I've tried and didn't hear a difference. I am going to have to do some really close comparisons when I do get a chance to see if I can capture the differences between the two resolutions. Again, thanks for this video Justin!
I remember converting the black album cd in the '00 to various kbps mp3, starting from 32kbps to a variable which at that time was up to 220 kbps I think. And the sound depended a lot of the speakers. In the PC starting from 128 kbps sounded good, but in the hifi equipment to a properly bass sound must had be the variable, not less.
Even if actual 320 kbps mp3 is said to be enough, I consider quality cd, be WAV, be FLAC, etc, the most honest format to sell music, just in case, the original, nothing removed.
Sounds reasonable to me! Though I understand why people don’t want 4x larger CD file sizes on their devices or streaming sites for practical reasons.
-Justin
I use Wave 48K at 32 float because I believe if I yell too close to the mic, and it goes way into the red by accident, it won't distort. When I export to MP3, it still sounds good. Am I wrong?
Good talk. As a matter of fact, nothing touches say, a 2" Studer and a vintage Neve console. Of course also the old outboard gear. There's a reason for all the modeling algorithms of that vintage analog gear. Yes, analog has a "smearing". But, digital lacks some body that analog has.
Meant to mention. There's nothing like the saturation sound of hitting 2" tape really hard.
Some good examples of that "tape hit hard" sound: "Best Friend's Girl" by the Cars, "Bicycle Race" by Queen, Both produced by Roy Thomas Baker. He would have the engineer hit the 2" 30ips master really hard and said 'if the recording head glows red, we have a hit.'
@@RadioCamp 👍
Well done on this video! I have been singing this same song for years on forums and with students. As a delivery medium, why do you need more than 100dB of dynamic range? (dithered 16 bit). The part I think many still struggle with, and I can see it in some of the comments here, is not understanding the bit depth of a session. (Speaking about Pro Tools here, but I believe other DAWs are similar). If set my session to 16 bit and fill it with 16 bit files, we are not operating in a 16 bit world. Every DAW these days has at least , a 32 bit floating mix engine. You are mixing and outputting a 32 bit signal IN SPITE OF THE FACT THAT IT MIGHT BE MADE UP OF 16 OR 24 BITS FILES. The bit depth of the session only relates to the bit depth of any files you record into the session or any files you import that require conversion.
I started as an engineer in the 80s. 60db headroom was about what we had. When the 16 bits came in no one could believe how low the noise floor was. All DAWs now have 32 bit internal buses and some even have 64 bit summing. No one seems to understand. I still deliver projects at 24 bits as otherwise, people don't think they are getting their money's worth. This is to make vinyl from lol
I find the only benefit of hi-res releases are that the albums normally receive a better mix, Metallica's Death Magnetic is a great example... the retail CD is a brick walled mess victim of the loudness war and due to the compression it's actually tiring to listen to it, the 24bit/88.2khz hi-res release just sounds better due to the mix and lack of compression, so I just converted the flac files to 16bit/44.1khz and burnt myself a CD that sounds better than the one I parted with cash for..
100% this. Of course it’s not a benefit of the hi res format, but having a better mix is a huge reason I buy hi res.
Yes, that is a valid reason to repurchase etc. It's a shame however that people are being convinced they should buy because of the numbers and not because of the fact you get a better mastering or better recording.
If you convince everyone it's the numbers that matter, they will rebut every CD ever owned by them and it also attacks the second hand market.
Imagine reprinting a book and encouraging everyone to rebuy because the new print has a HD font that will improve your immersion in the text.
I swear that CD's sound slightly fuller in the low end to me in my car vs. me playing the same album on Spotify on my phone using an AUX cord. I have no idea how much of this is just placebo or if there's something else at play, such as the DAC in my phone or whatever.
Spotify is objectively worse imo, I listened to my CD version of an album then the Spotify stream and the spotify one is so muddy compared to the CD, like there was a blanket over it. I'm pretty sure I have "high quality" enabled on my Spotify too, and was listening to a downloaded album so it should've been optimal quality
That is because of the Aux cord , you will get some armonic distorsion from the plugs themselves and also if you are using a 3.5mm jack the lower freq range is not as acurate ,, /( correct me if I am wrong but from 120Hz down / so yes, you are rigth, you can feel and hear the diference , but it is not because of the file audio qualitty ;)
@@fernandoferreromusic I thought it might be the aux cord, but then I listened to the same album on my MP3 Walkman (same cord) and it sounded much closer to the CD quality than the Spotify version :o I think this might be specific to Spotify, they must do something to the audio files even if they're technically at 256/320kbps I think they do some extra compressing to save space.. although it could also just be my phone's aux jack vs my Walkman jack, the phone jack being inferior?
There are a few factors here. One is that you could be listening to lower resolution mp3s, such as 160kbps instead of 320kbps, depending on what version of Spotify you’re on. Trained listeners may be able to discern subtle differences in that case. (Though most people can’t.)
Another is the the DAC, sure, but often more importantly, the analog components of both your phone and your car stereo that aren’t in the equation when playing the CD.
Another even bigger factor could be volume differences between the two inputs, including Spotify turning down louder material, but also the audio input (or phone output) being quieter than the CD.
Level differences explain a LOT in audio preferences. And Spotify does indeed turn down the loudest material for consistency, which your CD player and your MP3 player likely doesn’t do.
Even if none of that was at play, and you had identical levels and audio (which you probably don’t) sighted listening makes a difference as well.
I could go on. But I probably shouldn’t! :-)
I hope that helps,
Justin
The same album, but how do you know they are the same MASTERING?
I have 100s of classical cds at 16 41k. The biggest difference I can hear is not in the quality of the amplifier, the DA converter, the headphones, the full tube amp, the toroidal transformer... It's in the microphones, mic placement, instruments, studio preamps and mastering.
I do think a small difference is gained however in SACD.
Will I need the lower noise floor when they play my trap beats in stadiums?
This is a very fin topic.
Someone should have a discussion with the hi-fi (consumer) industry about this and then with the audiophile listeners using converters at 768 sample rates when listening to lower sample rate files.
I do a lot of sound design and multiple renders of audio. Stretching, bouncing etc and have recorded at 96k for the last 12 years for that reason. Also my clients want 48k almost 100% of the time so the conversion is seamless this way. I think some of the “myths” of higher SR are leftovers from when converters weren’t as good (maybe), like early aughts. But if you really wanna hear a marked improvement, get a master clock. That was an eye-opener. I run 96/24 with Lynx Auroras and a Big Ben-that’s enough! But this video gets into consumer habits, and listening-and I agree. CD is fine and I really prefer vinyl. But it’s apples and oranges comparing processing digital audio with delivery specs vs listening to music. My .02.
Absolutely, for sound design or sample libraries where you’ll be time stretching audio or pitch shifting down significantly, higher sample rates make sense. That’s one of the special cases mentioned near the end of the video. And if you’re working primarily with video, 48k there is the norm, so may as well deliver it that way instead of sample rate converting a 44.1 file.
Thanks for the comment,
-Justin
Ah I wrote this literally 1-2 minutes before that caveat, haha. I also agree with the many other attributes mentioned as to creating good recordings, or rather, the Things That Make SR And Bitrate Moot. Like bad preamps, oppressive, stark, reflective rooms, poor levels, and questionable choices-the list goes on!
I've been saying for quite a long time, anything beyond 24-bit 48kHz (chose this really only because it's what you'd see on DVD's and Blu-Rays) is useless outside of production environments.
In the early 2000’s, I remember a salesman at a music store tell me I should use 24 bit over 16 bit to avoid audible “stair stepping” in a fade out to digital black. And yeah, I THINK I heard that once, in the last second of a 16-bit file where sound was attenuating between -60 to -90 dB while my interface and monitors were turned up to maximum volume. *I think.*
I think we should all consider that when listening to pure tones as hearing tests it doesn’t reflect the real world where tones are combined. It’s the simultaneous interaction between frequencies, even the frequencies that are considered outside of the range of human hearing because of single frequency hearing tests primarily, that can get very interesting so to speak in this conversation.
That's not how it works. Yes, there's intermodulation distortion - at least in theory - but that's too low in level to be audible over the actual playback material, and probably not present at all in actual recordings.
I'm not a fan of test tones, either. But this often cited euphonic interaction between frequencies is a myth. There are interactions, but they're either not audible, or just as audible at normal sample rates.
@@michaelanderwald4179 It also sounds more like something that would result in amplitude peaks rather than frequency distortions. If we can't hear a certain frequency, it's not going to matter much if two sounds at that frequency interact a little and boost in volume, as we can't hear it. Perhaps I'm mistaken.
@@FloatingOnAZephyr Depends on the type of interaction of the frequencies with each other. Intermodulation distortion can go much lower than the frequencies that cause it, but it's low in level and probably not something that would sound good. It can actually sound a bit like aliasing, so very much not harmonic at all.
I've sometimes wondered if some people prefer DSD recordings because the energy around 100kHz due to noise shaping adds some kind of biasing effect on the playback system. But that could be added in other ways to a "low" resolution playback system. Also it's pure speculation.
Almost all hifi claims are silly unless they make you happy; the very purpose of a hobby
Thank you I just discovered your channel, this was a great video!!
Hold on... if I'm understanding this correctly: I can record a VO using a sample rate of 44.1 kHz, and I can do the same reading in another recording at 48 kHz, and the only real difference is that some of the higher frequencies (which annoy me anyway) won't be there in the 44.1 kHz?? Or have I misunderstood?
Human voice doesn't have much information beyond 18Khz, I mean there is something beyond that in noisy consonants like K, T, S sounds, and in the sound of breathing and other things like mouth clicks.
But honestly the part of those sounds that you will loose going form 48KHz to 44.1KHz is not something that you would want to hear, if your ears can actually hear it and your sound system can reproduce it.
@@kelainefes Thank you very much! I mean with VO, most of what you mentioned is something that narrators etc. work hard to eliminate anyway, or rely on audio engineers to remove for us! Again, thanks! 😀
@@HungryForTastyFoodAndComicArt Yes, for a voiceover an engineer would want to remove mouth clicks and greatly reduce the amplitude of breaths to the point you fell them more than hearing them, and the consonants would surely not be receiving boosts in very high end, and depending on the microphone used would be processed to sound a bit "darker" or "smoother" and certainly not brighter.
@@kelainefes Exactly, again, thank you 😊. (edit) - many of us have found a huge difference between a Røde NT1A, an AKG C214 and a Lewitt Subzero.
Pretty much! There is some chance that a given converter could sound slightly better at 48k than at 44.1k if it is using an analog anti-aliasing filter, so you might get a teeny bit flatter response around 20k or so with the 48k mode in that converter (if you can even hear that high) but that’s about the strongest reasonable case you can make, I think.
-Justin
Apparently you’ve never heard of noise shaping filters as it applies to bitmapping. Dsd sounds remarkably different than an mp3. Maybe your speakers aren’t able to show you what you’re missing.
condescender.
Thank you very much. Now I can stop searching for "better sound" that I cannot hear and stick with my humble music streamer and Cambridge Audio.
FYI, TH-cam processing adds an incredibly steep low pass filter at 15kHz. Some older vids may have one at 18kHz, but it seems like they lowered it a few years ago.
TH-cam does a low pass, yes, but as you say that it is very steep I will speculate that you are uploading MP4 videos with audio in AAC format.
Many encoders do a low pass at around 15KHz just like YT does, so if your encoder does not give you the option to disable that filter you can assume it is enabled, and that your audio is being lowpassed twice.
The solution to that is to upload .MKV files with MP4 video and wav audio, which are supported by YT.
What’s interesting to note here is that most people not only can’t hear super high frequencies, but they generally ALSO can’t hear a fairly steep roll off beginning at 15k :-)
I can hear the difference between 128kbps and higher resolutions double blind, all day long. And this is exactly what I listen for these days. But even I have to admit it’s gotten fairly subtle these days.
Someday, I’ll be old enough that I probably won’t be able to. Super high frequency hearing tends to decline with age. At that point I’ll probably be approaching the age of the average person who swears that high res is better! X-D
128kbps does benefit from a fairly aggressive high frequency roll off for sure. 256 and 320 less so.
Back in the day it was easier to distinguish 128kbps, back when the codecs were suckier and they didn’t do the high roll off and when I had the ultra high frequency hearing of a teenager! But take any one of those factors away and it becomes more subtle when listening blind, even when you’re trained and can do it reliably.
@@SonicScoop Indeed. Today the only artifact I can consistently hear at lower bitrates is the "digital jingling" at the top-end of hi-hats and cymbals. Even that goes away for me starting at 256kbps and my blind test results 256kbps+ were no better than guessing.
My high pass filter note was really more a warning against using those high-pitched "hearing test" videos on TH-cam...they don't work too well :-)
Just ran this video's audio through a spectrum analyzer and it's all there, right up to the 19k using 144p video quality. I don't think audio quality changes with picture tho. Every other day youtube changes. Thanks for giving me something to do.
I enjoy your explanation. BRAVO
But two things I didn’t quiet get til now, should I use a low pass filter at 22.1 kHz in my master or my mix to get rid of that stuff as soon as I do not apply over sampling with my limiter? I thought over sampling would just make the chops off more rounded like a analogue clipping does? 🤔 and if I apply more over sampling at high rates, the sound really smooth outs more which is not always wanted, I really tested that with the pro l and layed 5-6 different versions from no over sampling I think up to 16x over sampling and stuff, so should I cut everything off higher than 22.1 kHz if I’m not gonna over sample with my limiter? And the second thing is about dithering... I thought dithering should be used as the final mastering process, when converting 32 bit files to 16 bit to avoid artifacts isn’t that so? Because if I’m not hearing any noise floor and I have to admit that was never the case because I’m producing mostly hip hop and rap where are just no quiet passages in my song, does it makes any sense to apply this? Thanks yours sincere Mathew 🙏🏼
And if oversampling is acting like a brick wall I wonder why this is affecting the higher frequencies because mostly the low end gets limited the most or am I wrong in this case? You threw up a huge question mark right here and now lol, but I heard that oversampling is reducing the aliasing but I didn’t know it has to do with these high frequencies, I thought it would be about artifacts in general
Great discussion/explanation...... thanks for putting these concepts into their correct context.....
You're sorta right and sorta wrong. When testing my fave synth at 2441 and 2496, I was shocked at the difference in clarity. Always good to start at the highest possible quality and then Downsville, if needed ... that's my 2 1/2 cents.
I think I mention in the video that synthesis is one of the reasons you might reasonably use a higher sampling rate :-)
Thanks for this post! I’d love to hear your perspective on apple digital masters, I feel like they sound better but I don’t know the engineering behind it.
Thank you. Your video helped me staying with Spotify for the UI, multi-platforms (Connect) and international songs selections. Honestly tried Apple Music and Tidal and just not getting any additional benefits.
Yeah, 320kbps is a pretty amazing consumer format. Improvements from there have routinely been shown to be existent in double blind listening tests, so whichever service you prefer for other reasons is probably the more important variable. Glad you found a choice that makes you happy!
Very best,
Justin
Justin - What bit depth/sample rate will you be using for new projects from now on? Or what would you be using for a tracking session?
Great question. As a mastering engineer, I prefer to receive whatever bit depth and sample rate my clients were working at. But in the overwhelming majority of cases, the final deliverable resolution of the master is going to be 16/44.1k. That's what most digital distributors accept, and for good reason.
If I was still recording and mixing today, I'd probably work at either 24/44.1 or 48. For my tastes, I'd probably go for 44.1k.
In the cases where I can hear the difference between the two rates, it's usually because the converter in question uses an old schools anti-aliasing filter that has slightly more super high frequency roll-off at 44.1k than at 48k, and this can actually sound a little more pleasant and "tape like" to my ear in most cases. But there are probably converters where I might prefer the opposite.
In other cases, say where the converter is designed with a proper digital anti-aliasing filter, the two are indistinguishable, so in those cases, why not use the smaller one, if there's no benefit to going larger?
In either case, I'd probably test my chosen converter or interface in a double blind way to see if I actually got any perceptible benefits either way.
Even in cases where there ARE perceptible differences in a given converter, totally blind, the are likely to be extremely subtle, amounting to a very tiny EQ difference in the extreme highs. If the difference is more significant than this, that's a problem with the converter design IMHO.
If I was working mostly in video, maybe I'd stay at 48k because that's more the norm there to cut down on the number of conversions needed back and forth where errors can happen some of the time in fairly rare cases.
Hope that helps!
-Justin
@@SonicScoop Yeah I did a brief project with Producer Greg Wells and he was using 24/44.1 and so I started using it after that LOL other engineers argued with me but I couldn’t hear the difference. I wanna do a session at 16/44.1 but I’m scared lol
So the streaming services prefer 16/44.1 submissions? Imagine how much less hard drive space I’d use and how much faster my transfers and bounces will be as well!
Looks like 15k is my cutoff at 38yrs old and A LOT of metal concerts in my teens/20s.I barely heard 16k.
End consumers getting the file in the same format it was recorded/mixed avoids a Sample Rate Conversion, which can mean less degradation. Higher sample rates give a more accurate trace of the analog/electrical wave form.
If you want to avoid sample rate conversion, don’t use the unnecessarily high sample rate to begin with :-)
As far higher sample rates “giving a more accurate trace of the waveform”, I’m sorry but this is incorrect.
That’s the exact misconception that we start off the video with, on purpose.
There is just no mechanism in physics by which increasing the sample rate can do anything except for increase the highest frequency you can reproduce.
It’s a bit of an all or nothing affair: Either a given frequency can be reproduced perfectly at a given sampling rate, minus some noise (which is determined by the bit depth) or it can’t be.
This can be confirmed for yourself with proper testing.
Similarly, there is no mechanism by which increasing bit depth can do anything but lower the noise floor.
If you believe that there is, that would be an amazing breakthrough discovery and science!
I mean, could you explain the mechanism by which anything else could occur?
I hope that helps,
Justin
@@SonicScoop when you have more samples per second, there is less rounding going on as the digital waveform tries to approximate the analog voltage (input).
A hypothetical infinitely sample rate would track the voltage perfectly, to the atomic level or beyond. There would be no rounding, at that point. Its calculus. Approximation of a circle using discrete values. More "steps" equals a closer approximation to the continuous curve.
Why not use less noise, noise is not typically desired. Even if not audible, it has interaction with the audio. There is no universal reason to not be technically better. If for no other reason, than archival purposes.
If you null test an mp3 @ 320kbs, and a 24/96 file, there is no residual sound?
Interesting discussion.
-Kyle
@ReaktorLeak im talking about the capture at the ADC stage. Voltage to binary. A "perfect" conversion would track the voltage fluctuations down to the electron. Otherwise your missing information at conversion. ie the information that would otherwise be in between samples, is not recorded, it is ignored, the waveform jumps to the next discrete value, the analog signal is continuously variable. Up to the limits imposed by the hardware.
Trace the outline of a circle with straight lines, once with 1" lines, another with 1/4" lines. The shorter more frequent lines give a "truer" representation of the circle. This is akin to sample rate. More samples, more accurately traces the voltage functions aka the analog audio waveform.
@ReaktorLeak you cannot recreate the circle however, because samples are discrete values (straight lines) in your example you would have a triangle. The samples cannot reflect the "inbetweens". The voltage change happening in between samples. "Reconstruction" of the inbetweens ends up in averaging. You don't "know" the voltage fluctuations, your guessing them based on the two points. A momentary transient occuring between two samples, would be ignored. It would go undetected, and get rounded/averaged away. The samples can't dectect things in between samples. They are akin to steps, not curves.
Calculus shows us that more discrete values (samples) more accurately trace a curve than less.
Its just like frame rate in video. 29fps is fast enough to fool the eye, until you slow it down. Higher frame rate is always truer to continuously varible.
@ReaktorLeak you can perfectly re-create the circle IF know ahead of time its a circle. If you do not know its a circle it could be any possible configuration of lines between those 3 points. A converter doesn't "know" the voltage fluctuations in between samples.
Ultrasonic content in high sampling rate audio can actually be very counterproductive, if the complete audio chain does not support it. Tweeters and amplifiers can generate folding frequencies which can be considered strong distortion components.
I don't know the details of Rubert Neve's listening tests where he claimed people could hear the difference between an 18 kHz sine wave and a square wave with an 18 kHz fundamental, but I suspect this kind of distortion was the culprit.
Last week I switched to high sample rate for my mixing template. Not for that "super high quality" sound (actually when the files were recorded at 48kHz than they CAN NOT be converted to higher samples rates with all that extra information) but for the digital processing I use. Some of them have oversampling but some simply don't. And those plugs which don't have oversampling CAN achieve a more accurate harmonic saturation/distortion. It depends on how well the plug-in was coded. So the sampling rate of my project is high but my exported file is still 16Bit 44.1kHz/48kHz. The only downside is the processing power - but hey with this "trick" you convince yourself using less plug-ins 😉
If it works for you keep on doing it! I would suggest that you eventually do an experiment though, just for fun and certainty. Try this:
Bounce your mix from your 48k session. Then, copy your entire session from 48K to 44.1 K. Then, bounce that otherwise identical mix as well, to the same bit depth and sample rate.
Now, load both up into an ABX tester and see if you can hear the difference!
This would be one way to confirm that you were actually getting the benefit you hope you are getting. Maybe you are, maybe you aren’t. I don’t know!
I have recommended this to a friend who worked at 88.2, and they have found that they couldn’t tell the two files a part. In some cases they stop using 88.2, in some cases they didn’t.
But in no cases so far did they hear anything like the kind of difference that they thought they would.
Maybe your situation will be different. In either case, if you try it, please share the results here! A lot of people would be interested to hear them I think.
Thanks and I hope that helps!
-Justin
i converted a cd wav song with sox to 4 bit 36khz. what shall i say, i dont hear a difference.
The Importance of Bit Depth in Low Dynamic Range
16-Bit and 96 dB Dynamic Range:
In 16-bit audio, this means the entire theoretical dynamic range of 96 dB is available. This is divided into 65,536 steps (2^16). However, in practical use, the entire dynamic range is not always utilized. If the music has a dynamic range of only 5 dB, for example, in practice, only a small number of the available 16-bit steps are used - about 2-3 bits out of the 16 available bits. This means that the remaining bits contain no "useful" information, as the music doesn't get any louder or quieter.
4-Bit and Dynamic Range:
In 4-bit audio, the dynamic range is limited to 24 dB, and the volume levels are divided into 16 steps (2^4). If the music has a dynamic range of 5 dB, the music will be divided into only 2-3 bits, which is more than sufficient to represent the audio content without noticeable quality loss. Since the music has a low dynamic range, the remaining "bits" of the dynamic range are not required and are not used.
Sampling and Bit Depth
Bit depth is primarily a measure of the resolution of the dynamic range, i.e., how finely the volume levels are divided. If the music uses only a small dynamic range, only the corresponding bits are utilized. A 16-bit audio format theoretically provides 96 dB of dynamic range (with 65,536 steps), but if the music only uses a dynamic range of 5 dB, it means that only the finer steps of the lower bits are used - this is 2-3 bits out of the 16 available.
If the dynamic range of the music is very low (e.g., 5 dB), then 4-bit is more than sufficient because the dynamic range of this music lies within the 24 dB of 4-bit. Quantization noise or stepping will not be noticeable because the music simply doesn't have a larger dynamic range.
Summary
16-bit audio provides 96 dB of dynamic range, but with music that has low dynamic range (e.g., 5 dB), only 2-3 bits are used because the volume fluctuations in the music are minimal. The other bits are not utilized.
In 4-bit audio, the dynamic range is reduced to 24 dB, which is more than sufficient for music with a small dynamic range (e.g., 5-20 dB).
So, it's not about higher bit depths offering finer resolution when the music has a low dynamic range. The higher bit depth is only relevant when the music actually requires the full 96 dB dynamic range or more. If the music is recorded with only 5 dB of dynamic range, then the 16-bit resolution is just as unnecessary as the 4-bit resolution because the music only uses a very small portion of the dynamic range.
This is not right. If you record and playback a sound that is 4k on an 8k system you are subject to huge aliasing issues. There is always a filter to roll off the top frequencies so they don't reflect down
Yes, that is correct. You do need to use an anti-aliasing filter, so technically, you need a sample rate ever so slightly higher than double the highest frequency you want to capture.
With modern HD converters that use oversampling for very steep anti-aliasing filters, there doesn’t have to be much of a gap there at all.
I purposely avoided discussing that in great detail so I wouldn’t be throwing in too much technical information for people, but you are absolutely correct about that.
That is why the standards are 44.1k and 48k, not 40k on the nose.
In the early days when analog filters were almost always used, perhaps 44.1k could have been arguably been slightly too low for some filters. But today, that shouldn’t really be the case.
Hope that makes sense,
Justin
@@SonicScoop very much so! Thanks for clarifying
I have to be honest... Im recording in 32bit just because I can, after every is done I export it to 16b anyway... And I really never saw any use in using more than 4x oversampling in some plugins. If someone is using x8, x16 or higher... I'd really like to know in wich cases do u use that much of oversampling?
I use more than 16x only in clippers on my master bus.
Not that I can hear it, but at 16x I can still measure the aliasing distortion from high frequencies at around -90dB, and at 64x it goes down to about -112dB.
It's probably a minor form of OCD on my part more than anything else, but knowing that the aliasing distortion is below the -96dB noise floor helps me sleep better.
Ok scrap probably, it definitely is OCD.
@@kelainefes Yeah to be fair I can see the reasoning behind it. Better safe then sorry 😁
But from my side I never really use clippers on a master, I tend to do it on Bus level... So I can clearly hear the difference switching from native to x2... x4 that depends on the material... Fabfilter recently introduced x32 oversampling... I can't wait to play my tunes to superman so he can hear that buttery smooth top end at 2MHz 😅😁
@@DaveChips My reasoning for a soft clipper before the limiter is this, I use it to lower the highest peaks so that the limiter has to work a bit less.
To my ears, this reduces the loss of punch from kicks and snares compared to achieving the same amount of peak reduction with just a limiter.
Ofocurse I always A/B with and without the clipper and sometimes I don't use it.
This is super interesting Justin. It means I can start a 44khz project in my DAW instead of a 48khz one and use the CPU and RAM saved for distortion, analog emulation, etc to make it all sound more interesting. My main concern would be the Nyquist frequency: when using certain plugins that may not have a resampling option, working at 44khz would be worse than doing so at 48khz as some artifacts would be more noticeable and hard to clean up? Thanks a lot for this episode!
Oversampling is needed only to reduce the aliasing caused by plugins that can introduce harmonics, so saturators, clippers, brickwall limiters, fast compressors etc.
Working at 48KHz doesn't really get rid of aliasing at all, you always need oversampling to do that anyway: if harmonics are introduced to a 15KHz sine wave, in instance, the harmonics produced will be 30KHz, 45KHz, 60KHz, 75KHz, 90KHz, 105KHz etc and that's just up to the 7th harmonic and 192KHz would not be enough anymore so you need to have a sample rate that allows those harmonics to be correctly described.
Whatever harmonics go beyond the Nyquist frequency are mirrored back below the Nyquist frequency.
In instance, if you're working at 48KHz and you use a non oversampling saturator on a 15KHz sine wave, the 2nd harmonic will be 30KHz, and it will be mirrored back and appear at 18KHz (30KHz - 24KHz = 6KHz, and then 24KHz - 6 KHz = 18KHz).
@@kelainefes thanks for the math and clearing up the concept, very well explained.
So, I've done the sine wave test, but one thing that I just don't get is the square wave hearing test. From my understanding, a square wave generator is just adding odd harmonics in ever decreasing amplitudes. So, if that's the case, and with what I can hear from a sine wave, the square wave should start sounding "siney" at (at best) 5.5-5.7 KHz, though when I listen to one, the tone changes quickly starting at 11 KHz or so. Is there someone here smarter than I (not difficult) that can explain what is going on? I mean, from what I understand, it seems like frequencies I can't hear are affecting what I can hear.
I think the one thing I disagree with most is the assertion (not yours, but you mentioned) that certain analog gear can't handle the frequencies of the high sample rates. I would imagine if there was a problem on mixing, then the problem would have been there while monitoring on the way in. I also can't really think of components that would suffer from the higher frequencies other than maybe a cheap and dirty op amp, but even those have been rated 50KHz+ for a very long time now.
As far as the content, I generally agree. Music is dense enough in the audible range that there is just too much going on for us to hear a meaningful difference other than "well mayyybe." I mean, possibly singular sounds higher sample rates would make more sense, I don't know. I thought I've heard differences before, but I figured that was more on (like you said) getting the filter out of the way if it was an old/cheap/crappy converter. That said, I do default myself to 88.2/24 recording because of at least perceived differences I've heard in the past, but have not scoffed when someone asked for 44.1/24 or whatever. And when I buy music, I go CD quality or better (physically, I have switched to buying vinyl whenever possible). Storage is cheap and processing power is ridiculous now.
Good stuff; I've been going back and forth with myself on high resolution audio and audio formats in general since...well, forever. I hope this brings about good discussion and hopefully some other people on the other side of the aisle.
i don't seem to hear anything above 16khz. when i was a boy having equipment that had frequency range up to 12khz was quite good. and everything sounded just fine - percussion etc. for example a sharp boombox. or a turntable with a built in amplifier. they sounded just fine.
Mating sounds for bats is exactly what I do
I have been using yt music for a long time now, and I saw TIDAL commercial about "high res audio" you probably already know this. I took all the audio gear I had, all the headsets and speakers and I compared and tested.... No difference expect that yt music is louder and had bigger bass. No doubt that TIDAL has studio audio quality but I don't hear a difference.
Genuine question. I heard somewhere that 96k is a requirement for film sound/music.
1. Is that true?
2. What benefits (for film) does it give to have audio in such high resolution?
48K tends to be the standard in video production that is used most commonly in my experience. This is mostly because the math works a little bit better between 48K samples and 24 frames per second. That way, you always have an even number of samples for each frame. That’s pretty much why that standard was adopted. Hope that makes sense!
These discussions are always based on the level of human hearing. Just because we can’t hear greater than 20k hz doesn’t mean it isn’t there... and if the sample rate can’t handle it, it will mess with other frequencies and get into the audible range. What’s the frequency response of your tweeter? Probably higher than 20k hz... this then effects the other frequencies that tweeter is producing. I’m new to this though, tell me what I’m missing here?
Hey Cris, I get what you’re thinking, but it’s actually a little bit of the opposite in most cases, surprisingly enough.
Removing excessive supersonics won’t mess with the frequencies down below (you can test and confirm this for yourself with simple null testing). But having excessive supersonics CAN actually mess with the signals below by introducing “intermodulation distortion” to the playback circuits.
Yes, there’s a chance your tweeter and analog chain can reproduce 50k fairly flat if it’s really really expensively designed.
But if it’s not, then pumping a ton of 20k-50k+ through them that they were never designed to handle can actually cause distortion than can fold down into the audible range.
It’s something a lot of people don’t consider, but it’s a true possibility of ultra high sample rates.
Hope that makes sense!
-Justin
@@SonicScoop Yes that makes perfect sense! So it’s not that it can’t handle those frequencies, it’s that it doesn’t pass along those unnecessary high frequencies, which can cause problems themselves. Thanks for the reply and clarification!
There is something I don't get. Take any audio interface that allows you to record at 96khz and TLM 103 mic. How can you actually record 96khz when this mic goes up to 20khz? isn't there any placebo effect?
No, people always use Nyquist-Shannon sampling theorem to say you only need to sample twice the highest frequency you want to record i.e. in Audio you only need 40kHz. But did not understand the whole theorem. The theorem also said you need to multiple every sample to a infinite function call sinc function to be able to recover the same wave form. The important word is "infinite function". It is impossible to calculate infinite function because that would take infinite time. So you always need to stop some where just like when you use Pi. Higher sample frequency will make calculation many times near to "infinite function". So same DAC chip with same computing power suppose to have easier time with higher sample rate in theory.
Well illustrated, on top of all that we should consider the frecuency response of microphones (good old SM57 does not go over 15 kHz), preamps, etc. Just saying,
18:40 Sorry, but that "battery draining 10-20 times faster" is way overblown. Practically every modern multimedia capable device decodes audio and video on dedicated chips, and as long as that hardware is compatible with your audio format the difference in energy usage will be negligible. Though to be fair when you stream the stuff it'll be more noticeable.
Otherwise I agree though. Usually I try to prefer FLAC simply because memory is cheap(unless it's from Apple lol) and it often doesn't make a difference in price. Even if it's just to be able to losslessly convert the format if needed.
Isn't a higher sample rate with more values resolved by a high bit depth and lower step sizes represent a better reproduction of the original continuous signal when optimized for high accuracy?
I don't see an issue with people knowing for sure they are listening to the most faithful reproduction of music especially if they are willing to pay for it.
Nope, that’s what I thought originally as well. But that’s not how it works.
In reality, there are no “steps” when you play back digital audio.
There is only a continuous analog waveform, identical to the original, except that you have to eliminate frequencies above a certain point, and there is some additional noise 96 dB or more down, where no one will here it.
It’s a common misconception, and one that this video hopes to correct.
Counterintuitive I know, but that’s how it works.
Very best,
Justin
Hi
Great video? Does 360 degree reality audio really a good experience for audiophiles?