Debunking peak levels in mastering…
ฝัง
- เผยแพร่เมื่อ 28 เม.ย. 2024
- ⭐ STEM MASTERING MASTERCLASS - www.panoramamastering.com.au/...
📧 The BEST newsletter for working audio professionals who strive for more! Subscribe here - www.panoramamastering.com.au/...
🎹 Proud of your mix? AWESOME! Get a fresh set of ears on board for mastering and let’s work together! - panoramamastering.com.au
🛠️ Mastering Template for ProTools, Logic and Ableton- www.panoramamastering.com.au/...
--
Hello, I'm Nicholas Di Lorenzo, Studio Owner, Mixing and Mastering engineer at Panorama Studios.
I'm an Italian-Australian born and raised in Melbourne. I've been a creative professional for 10 years managing some pretty awesome projects for artists, labels and producers all around the globe.
What motivates and drives me?
My family,
Good food,
Great coffee.
You can find me on many platforms:
Instagram: / panorama_mastering
Facebook: / panoramamastering
Twitter: / panoramamasters
Kit: kit.co/Panorama_Mastering
![MEYOU - กอดหมอน [OFFICIAL MV]](http://i.ytimg.com/vi/RAG4Ee8FPxA/mqdefault.jpg)
Humility is the sign that I'm on the right channel. Thanks man
My pleasure! Thanks for watching
Agree! Love this channel!
Dancing to the delta of hard clipping. Pretty avantguard of you! 😂
Haha interpretative dance
I completely agree, at 96 kbps the codec artefacts massively outweigh the extra clipping distortion. It might be worth doing the same test @128 kbps, but even there it could be tough to hear.
Fwiw I got 4/5 on headphones, but here’s the thing - it could actually be easier to hear on the TH-cam stream because of the extra layer of encoding (!) - clipped material is harder to encode a clean result from.
You also made it super-hard for yourself by blind testing with only a few attempts, though. Blind testing is really difficult (and fatiguing), even with stuff that’s reasonably easy to hear. In fact, maybe I should do a video on how to blind test effectively, hm…
One easy suggestion to try - train yourself sighted, first. Listen to A, listen to B, figure out what you think the difference is, and try and check you really can hear it before going into blind comparisons. Also for distortion, headphones can be easier than speakers, because even tiny head movements can introduce apparent tonal changes (which is how binaural works) which can distract our focus.
Overall though I agree that -1 is a good safe choice, and this is why I suggest -1 dBTP as a great rule of thumb. It’s not foolproof, but for most material at not-insane LUFS levels, you’re unlikely to get any audible problems, even at low data-rates.
Great observations all round!
I retested this morning at 5/6… after some lengthier playbacks!
I’d be interested to hear this same test without true peak enabled. Also love the humility on being able to do an actual blind test and fail! Love that. No BS here. Why you’re one of my favourite people to learn from.
I couldn’t tell either but that’s likely because I was listening through TH-cam’s codec on an iPhone 😂
I personally stick with a -1dB ceiling, because even at that, I still run into pretty noticeable artifacts on things like Soundcloud. The problem is, the way these platforms cause your data to be lossy, it's not just simply hard clipping at 0dB, or intersample peaks getting chopped at lower resolution. Different streaming platforms compress your audio differently. It's not just bit depth crushing. Some platforms start cutting the very high end, people have found that TH-cam will roll off as much as everything over 16000Hz (which some people confuse themselves with after doing TH-cam to MP3 and think the file was actually mastered that way). Some platforms might compress out side information and be more mono, so if you push your mix/master too wide, some of that information may get lost upon compression. So on top of a safe ceiling, I think it's important to make sure your master passes a mono test, and that the high end doesn't have too much buildup.
What I've noticed that's worse than 'distortion' in uploading to some platforms is really bad compression artefacts, that tends to happen in the high end around hats/cymbals. They kind of come off as harsh ticks or squeaks, or the top end just sounds phasey altogether. I think what happens is if a platform applies a steep/brickwall high end cutoff, that's going to introduce a crazy phase shift, which can lead to resonant boosts which magnifies with distortion/clipping.
Another thing to consider for people who do audiovisual content is to not only consider a safe ceiling for audio exports, but also for their video exports. So if you take a song and put it into something like Adobe or Davinci Resolve, to make sure you have a limiter or something in the video export itself to ensure a ceiling there, because few platforms screw your audio up more than TH-cam, Facebook, Reddit, Instagram, etc.
I think ADPTR Streamliner is a great plugin that gives you a breakdown of how every streaming platform compresses audio files, and shows you that kind of 'metered' breakdown if you're on strong desktop connection or poor mobile quality, etc. Good companion plugin to Ozone's codec features.
Hope he reads your comment!
Great observations! Appreciate you sharing
@@panorama_mastering No problem, iZotope wrote up an article on this back in 2020 titled "Optimal Mastering for SoundCloud (and Compressed Audio Formats)". I think it's a good read that kind of highlights some additional issues you could run into with codec translation and lossy compression.
Honestly i could hear a difference in all the A/Bs, but i couldn't really tell if one version sounded better than the other
Yeah same, I could pick out the clipped one 5 out of 6 times, but it was such a tiny difference and i think mostly coming from the codec. It didn't really sound much diferant in totality, there was just a tiny bit off the top of the "shh" sounds, and that's the only diferance I heard and the only way I could tell.
Listening to what the codec took away, that was the predominant sound, so it must of mostly been coming from that I think.
Functionally speaking, I think it’s less important whether unwanted clipping can be heard on studio monitors in a treated environment than it affecting the sound on systems where it will have more of an impact on the overall listening experience. For example, I use my phone and a certain set of earbuds for the “is it clipping too much” test.
Also, like others have stated, listening zoomed in on a short loop may be less useful than zoomed out over a longer segment and listening general “bite”, “brightness”, “crunchiness” etc …b/c that’s how audiences listen.
Appreciate the analytical approach 🙏🏽
I agree. I was listening on my phone speakers and got 4 of 5. But it seemed harder with the shorter loops . Barely noticeable difference though
Good observation!
Got all right. It is subtle but noticeable.
got 4 out of 5 ill go truepeak -0.2
I won the null test 4/5, even after TH-cam CODEC, but it wasn't easy. I specifically identified the differences, which was a very subtle crunchiness when the kick hit. I do have a headphones on though, which helps in many many blind tests. Good video, thank you for sharing!
Good job! My pleasure :)
@@panorama_mastering odd theory, Listening through my stock computer speakers i could hear the harmonies get louder, but i put ATH-M50's on and i couldn't. Wondering if phase relationship can give more context to why some people can and can't hear it. I'd try it again and move your head back and forth.✌
Well done, love these tests and observations!
Glad you like them! My pleasure!
this is preeeeetty interesting!! thanks for the great video 🙌
You're so welcome!
Great video as always!! This is such valuable info, and really worth thinking about. FYI I was able to hear a difference 4 out of 6, from the in built speakers of a mac pro 15"
I found it in the vocal and verb tails. Thanks a lot for your work, it helps us all!
Awesome! Thank you!
Good info. I was able to guess 4/5 correct. The raw file seemed just a touch fuller, though the last one I had right at first and then changed my answer during the loop and got it wrong
Thank you
You're welcome
I always find I can hear harmonics and other sounds coming in at roughly -50dBFS, maybe -55dBFS at a stretch. I'm using KH120A + KH750 + MA-1 correction. I got 40% in the blind test you ran. Really can't hear the difference, and I don't think most people would be able to.
To be honest, we're spoiled with the technology we have compared to what was available throughout the history of music. While we're worrying about a teeny weeny bit of clipping noise, the oldies up in Heaven will be laughing at us haha! Then again, I always try my best to avoid any artifacts during mastering. Always best to err on the side of caution. I think of it like Audio Archivists - they always want the absolute best A/D conversion for future-proofing. Makes perfect sense when you think of it like that. If you have an upper limit of capability with your hardware/software, why not push the absolute limits?
Great observation and closing thoughts!
Thanks, Professor Panorama!😁
My pleasure!
In my studio the clipped one sounded a tiny bit brighter maybe harsher & the none clipped sounded smoother. Surprised but I got them all right :). Thanks again your videos make me so much better!!!!!
Great video! Thanks for addressing directly something I commented on in a previous video. Seems like -1 is best for avoiding artifacts almost completely but I feel safer about going to 0.
My pleasure
Love you video btw. I wear glasses and I've noticed you wear yellow lenses. I just read up on them and the benefits, will be getting myself some new glasses it's good for seeing dark areas and more dynamic range🔥💯
Yeah! Yellow (80% bluelight blocked) from 8am-6pm and red from 6pm onwards
3:17 ... a new genre of dance music is created ! ... "delta clip" they are calling it
I boogie down
The difference is obvious, but hard to say what sounds better.The vocals sound a bit softer and more natural on the raw version. The clipped version has some kind of enhancing effect on the vocals. Can be perceived as a good thing....That Hofa blindtest thing is great, have to get it. Thanks for another great video!
My pleasure! Thanks for watching along. The blind a/b is golden
On headphones it's easier to pick up the clipping harshness introduced. But still very very very subtle.
I got all of them correct, but I focused my attention only on the hihat, because it's the only way to find out, it's the most profound. Damn right, better louder and distorted than quieter and, well, still distorted.
Nick, it is your fault you couldn't hear the difference, we heard all six differences and got a 100% on their A/B tests.
Just kidding, good video.
Haha love it
I got 4 on 5, but I'm on headphone (watching at 1.5 speed, not sure if that help or not), but the difference can be heard only if you look for them, in a "real life" situation, it would not be noticeable (as you said, the codec quality would be the first obvious thing to ear)
Being forced to mix just with headphones on my laptop (without headphones amplifier) many times, and having a tight ear canal, I always need more pressure compared to the average, so I've naturally developed a way of mixing. First, I work on the individual tracks to shape the sound and to balance the various elements between them with eq and comp, while going even +3dbfs on the master bus, mainly paying attention to not get distortion in the lows when the peaks reach around + 3 dbfs. Then when the elements are balanced to my ear and they sound like they should sound, i search for what causes the spikes and I do what is needed to fix the issue. I put 2 compressor in series on the master bus to tame them. Subtle gain reduction, with 2 different parameters settings. At this point I have a clean dry rough mix, which sound already loud and full like it was mastered, ready to be colored and enhanced with any artistic choice it needs or I want.
Nice!
Nice! Thanks for sharing
Great experiment. Did your original master control the true peaks?
Yes
Can you please analyze tr5 one (Push Knob) it's something that sounds very good but I'm a little unaware of what exactly it's doing
In the A/B test the one sounding darker, more mellow in the high mids/highs is the CLIPPED one. RAW one sounds brighter and is more open/breathing in the top end and I can hear that even through youtube compression. I was correct every time. Maybe time to change/upgrade the monitoring or add another pair of monitors? Listening on Eve Audio's 2070s here. The tweeter on these monitors is incredibly revealing in the highs mids, especially for any saturation/clipping. However I'm pretty sure you'll notice the same difference on any ATCs from 25s up. The difference was really obvious. BTW at -60dB differences in 2 files are pretty big and clear to notice, -75 to -80 are really small and more difficult to notice from my experience.
Not sure why, I had to tempt fate and tested again this morning and picked up 5 of 6…
Hey, I've just seen your older video about this same topic where you end up auditioning -0.4 as having the least audible artifacts for the clipping.
There, you make a strong conclusion to use a different celeing for different material.
Could you explain what happened between then and now? Did your research/testing back then not check out?
Just listening on my phone and $40 Sony consumer headphones the clipped version had less transients and was slightly duller. I could hear it every time. Perhaps you'd hear it on cheap headphones?
Possibly so!
You can pick it out on the iPhone 📱 you can hear it in the roundness of the kick it sounds like it shifts where the punch resides
Good observation
I 100% passed the blind test from my phone. The snare's transient gets louder and the mids get tamed and clearer
I made a with 35.6db true peak and 24.6 lufs. is that good enough?
Could you null out the clipping distortion?
general question: How do you capture your screen with audio if there is the interface is used by Pro Tools?
I feel like the clipped version is a bit louder which made it hard to pick out the quality degradation
Interesting test, useful data for mastering, what are you using to monitor codec conversion? Have you tried the Sonnox one.
I was surprised re the Hofa tests, I think I got 4 out of 6. But surely the point is, even though the difference level was 60dB down, and maybe “inaudible” to most people, what is the issue with having a TP of -1dB or even a bit less. All streaming services normalise to a LUFS value of between -13 and 15. Surely the job of a mastering engineer is to translate what the mix engineer intended, but fix any obvious issues that can be done at mastering without lowering the quality. So perhaps a comparison test worth doing would be to look at the difference file between a track mastered to -1dbfs and 0dbfs. As if this is greater than -60dbfs clipping difference then your argument about leaving TP at 0dbfs is valid if on the other hand it is lower then mastering to -1dbfs would be more accurate.
The other point I was going to mention, do your same listening tests on mastered piano music, I think you will find the differences more marked sonically.
Feel free to point out any errors in my logic, as I am always willing to learn.
Really interesting channel am enjoying watching your videos.
I use ozone’s codec auditioner!
Good insights, thanks for sharing. Including trying the test out on different styles!
Test was bizarre. I swear I could hear one as being a bit more 'prickly' and bright on the transients, and the other more smooth, but I too only got about 50% right
It’s wild !
ahhh i noticed after the third one i got it right- i kept thinking the raw was the clipped and the clipped was the raw until i changed then i think it was easy to tell. the raw almost had more trans info which i thought was from the dist from the clip why i got it wrong then when i switched i got it right lol but actually the clip was taking away the trans - ever so slightly
I think I suffered the same bias!
I think the listening environment has a lot to do with the blind test. I was listening on some passive in-ear headphones and it started to get easier. I couldn’t hear the clipping though, just the difference. Try listening to the midrange and test again.
Good point!
To my great surprise I picked the clipped one every time. Since I have been worried about my audio discrimination (being 73! now, 20kHz, bye bye) that has cheered me up. Of course 5 out of 5 proves nothing, 90 out of 100 might. But thanks! I enjoyed that. Dum di dum di dum.... (hint: clipped one is "sharper" to me, with my 11kHz roll-off, listening on Audeze Penrose)
I got 4/5, but i used good headphones, dont think i'd pick them up on my monitors.
Good effort
Maybe I'm missing something :) Why not to follow general recommendation of -1 dB True Peak? From my experience the issue with fast peaks is not when a master file gets uploaded/converted to Spotify, TH-cam, etc. Even if it is lossless. Audible clipping might accrue on consumers D/A converters. Smartphones and laptops is the good example. So -1 dBTP is for downward SRC and to protect customer's D/A.
It’s less about following general recommendations, but doing tests and everyone setting up tests to explore this and determine black and white what works best
@@panorama_mastering Thanks a lot for sharing your experience!
Steve Jobs' adoptive father was a furniture maker and made sure that cabinets looked good on the back, even if you couldn't see them. This already influenced Steve Jobs when he designed the interior of the first computers. Would Jobs have lived with clipping that hardly anyone can hear?
I completely agree, this is why I want to get granular on the details.
IN MOST GENRES... if your master doesn't read over -0 on true peak . you aren't competing . I have a limiter set to -1 but im usually hitting +3 +6 on export . and compared to top 40 pop music that's LOW
Good observations
Yeah ! 5/5 for me on an iPhone speaker dude !!! 😅🎉
Good job!
Sometimes is easier to hear the distortion/clipping when listening to the side signal only.
Yes!!! But, no one listens to side signal only on the platforms ;)
Doing the test using high bitrates would have demonstrated fewer artifacts in general but would have been a better test in terms of worst case clipping artifacts vs compression artifacts. High bitrates definitely reduce compression artifacts but I'm not sure how much they would reduce clipping artifacts because the signal still has to be converted into the frequency domain and back again regardless of bitrate.
I think the question is: Are the artifacts introduced by lossy compression of music true peak limited to 0dBFS worth preventing by sacrificing some loudness or are they overshadowed by compression artifacts at even the highest bitrates?
The higher bit rates had less overs/lower peak levels.
@@panorama_mastering But are both clipping and compression artifacts reduced by equal amounts? I thought that at bitrates that are considered "transparent" it might make even reduced clipping artifacts more audible.
This is all probably heavily dependent on the particularities of the codec anyway... there has to be at least a dozen ways for codec developers to deal with the issue.
Thanks for running these sort of tests. Interesting content.
I don‘t get you test setup with the hard clipper. I am not sure you can replicate the distortion that‘s introduced in a certain DA-Converter this way. On my IPad i could hear the difference because of the slightly added harshness (interestingly it was harder to hear on short loops). After all … if you have a song that‘s already brighter and with added distortion it‘s not good to introduce even more in an unexpected way on certain hearing environments. I normally go with -0,3db and i doubt many can hear that difference to 0,0db 😊
I've been told the results in a real D to A could actually sound worse than pure hard clipping in some cases.
But the real issues is clipping at the decoder stage (before normalisation) which happens in several mainstream players (eg. TH-cam running under Chrome) and in this case Nick's test setup is accurate
Serban Mixes are peaking up to 4db above 0dbfs and still his mixes sound phenomenal! I don’t know if they do it on purpose to keep some of dynamics after very loud mixes!?
So many peoples mixes have peaks over 0 that are negligible.
But how do we know what it would have sounded like BEFORE that? If the ceiling were to have been brought down and free of that?
Raw sounded harder and punchier, clipped sounded a tiny bit more squashed. That's how I could tell (with headphones)
The difference is open high mids and highs in a raw track, clipped one sounds darker and more mellow
I hear the opposite. The clipped version has peaks folded over into higher harmonics and gives a saturated punch, making it brighter and punchier sounding
All descriptors mean different things to different people. So much fun reading through these comments
There's some inaccuracies in here. You seem to be conflating codec clipping and DAC clipping.
Codec clipping comes from overs created by a lot of filters that cause hard clips on their own. All of the lossy algorithms are floating point, and it actually happens during the translation back to fixed point.
DAC clipping happens mostly in the analog circuitry post-conversion but within the DAC, and that's where the pad (or just a lot of headroom) built into some DACs comes into play.
The test itself is still valid and shows some points, but the explanation is a bit more complex than what you've given.
Good insight! I did want to get someone with an osciloscope hardware to run tests on the analog outs of some daca based on output levels, but no luck so far
@@panorama_mastering yeah...the stuff to analyze DACs isn't cheap.
I think there’s something wrong with your routing and it makes it hard to pick out the differences.
You send to NULL the signal that already is lossy to both raw and clipped aux (it seems to me that you’re only trying to see how bad converters would react, but we must remember that a pro grade studio has good converters which will make this as smooth as possible)
I think raw aux should receive the studio quality (wav) and the clipped to have ozone with 96kbps and the standard clip.
Please reply if I’m wrong.
You observed correctly! But I am only doing that (trying to see how bad the converters would react :))
Why leave in a distortion when it can be easily avoided by slightly lowering your ceiling? I doubt anyone could pick out a undithered master in a blind test either, but we do it to achieve the best possible result.
Also, running at 96kbps maybe a worse case scenario with the most clipping, but it’s also drastically increasing your noisefloor, making it harder to hear. Doing it at 128 or 256 would have been a fairer assessment I think?
Possibly, more tests to be done!
I got 4 out of 5 correct in the blind test. What's interesting is that I've been a drummer in loud bands for over 50 years and I have hearing loss. My ears are fried. Normally I use hearing aids, but I wasn't using them for this test. And yet I could actually hear the difference. I didn't have to struggle. I could tell right away. The mp3 sounded ever-so-slightly distorted to me. Curious, isn't it? I did get one wrong, so, not perfect, but still... it's fascinating how the brain can compensate for the loss. I've also been mixing for 50 years so my ears are very finely tuned to minute changes in tonal balance and distortion, so maybe that's it? I can't hear over 8k, yet I'm acutely aware of my limitations and experience has taught me what to avoid in the top end, and I always have a younger assistant double check my mixes for any top end problems, but that seems to work for me. Mixing with hearing loss is a dirty little secret for a lot of people in this biz, and yet somehow we can pull it off. Love to hear (wait, I need to go get my hearing aids first!) more discussion about this from the pros.
I thought Spotify coverts everything to -1db headroom. Am I wrong?
I think you missed them because you’re listening critically. I can’t hear past 16k but combine feeling and listening you’ll feel relax or a sense of dull compared to a slight bite on the top end, even sharp or slightly pointy top. It’s actually useful for attention. If you can’t hear something move on, you will eventually and then you can’t undo, it’s a curse.
Good observation! Thanks for sharing
It's one thousandth
1 = 100%, it isn't showing percentile it is showing decimal equivalent
A tenth of a percent is one thousandth :)
Do u ever just digitally clip songs and let a bunch of stuff above 0?
Good to define what is other stuff?
@@panorama_mastering instruments bass vocals
I’m can’t hear a got damn thing. lol
I could be wrong, but I think you actually could hear the difference, because you consistently got it wrong, so I think your expectation/perception were not in sync and you were perhaps falling prey to your own bias. 🤷♀ I could be wrong, only you know for sure.
I think this might have been the case!
Cool tune? No, rubbish. Not worth doing a great mastering on it. 🙂🙃
Solid tune! I enjoy it :)