Julian has a great way of breaking down a complex subject that makes it easier to understand. While also retaining the relevant details one should be considering.
I use an F6 and an F3 and the convenience is fantastic! That said - I learned about 32 bit float recording's limitations the hard way. I was working with a microphone that had a max SPL of 110 (cough cough Videomic Go 2). No matter what happened, whether I used the Zoom F3, Zoom F6 or Zoom F2, the audio would distort when I got really really really loud. On the meter it showed normally but the audio was distorted. (For those wondering it wasn't a DAW issue, it's an SPL thing). You are limited to the SPL of the microphone. This also explains the few times I actually managed to clip even on an F6 with a Neat King Bee version 1. A lot of the marketing around 32 bit float recorders is a bit hogwash :P You'll often see the ever-popular diagram of the 24 bit graph with a small tiny little grey arrow that goes upward and the 32 bit float recording diagram with a grey arrow up to 600 DB. First off, you'd literally die if you recorded at 600 DB :P. Second of all, most mics are capped off at around 145 DB SPL so regardless of how awesome 32 bit float recorders are - you are capped off at the max SPL of the mic. That being said, what I can say - and Julian touched on this - is convenience. With a 32 bit float recorder I can do things 4-5 times faster than before. YOU SKIP STEPS. I no longer have to do things anymore. It's easy. No more fiddling with things. No more hiss-penalty, no clip-penalty, it's like the cheat-code in audio. No quarters needed, just infinite lives.
You are right about SPL of the Mic being a limiting factor, but about 32bit float Mic, a USB one, like Rode NT1 5th gen? With this mic you are not capped, right?
@@diktor_ilyakhaiko The 5th Gen NT1 has a max SPL of 142 DB. In other words, it's not as high as stated. That said - no one really gets that loud. But even 24 bit caps at around 144 DB. You'll clip out the mic before even touching the capability of 32 bit float. That said, in like 99% of cases you won't clip the mic as 142 DB is quite loud.
@@Dracomies I see, it is just physics, yeah? If you want to have more SPL, you must do something specific with a capsule, right? And not just add digital 32 bit float function. I thought 5th gen has 142 dB SPL as a XLR mic, but as USB it has more, like 32 bit.
Yeah, that mic is pretty limited. If you go with the VideoMic Pro it at least gives you three different options for gain, -10dB, 0dB, and +20dB. That can come in very handy depending on the environment in which you're recording.
Assumptions are being made here that are not correct. You say “no longer have to do things. SKIP STEPS. No more hiss penalty. No more clip penalty.” That is for most situations completely incorrect. You still have to fit the microphone’s analog input signal into a window the bottom of which is the electronic noise floor of that microphone and top of which is distortion generated by that same microphone. If you don’t, you will get a digital signal with infinite range playing back the same hiss and distortion you fed it. This is not cheat-code. It is not magic.
Started seeing 4th-Gen Focusrite Scarlett reviews popping up on TH-cam... But i'm personally holding out for Julian Krause's review, & ONLY Julian Krause's review! 😁
Nice coverage, Julian, very well explained. 32-bit float is useful if you wish to push decision-making and work to post production, assuming that your microphone can handle the SPLs you record. I'm with you, 32-bit float is very rarely useful for my work. In fact, if I'm using a channel strip or interface or recorder with a compressor, I can get closer to where I want my final audio to sit with 24-bit and save myself some work in post.
@curtisjudd so limitations are given by the mic’s SPL. But how would you control your gain, and know you have enough, not being too quiet, when using zoom f3, for instance? Also, does f3 have direct monitoring?
A great explanation of the differences between/advantages of 24bit vs 32bit recording. I just got my first 32-bit float capable recorder. My reason is that as a solo performer, when I'm recording my gigs, I have to put the recorder out front, set the level, and hope for the best that I don't overload the signal. I've had some potentially great recordings ruined by clipping because things got too hot for the actual recorder. With 32-bit float, I still try to set things as best as possible, but I have the sort of added insurance that I won't clip the recording. But I agree that it's not a magic bullit, as you can still clip the signal before it gets to the recorder. Good recording technique is still needed.
Yes if the audio is so loud that it clips the mics or inputs but with speech or unamplified audio this is unlikely and recording in 32 bit float is a great safety feature
Julian, you are the no nonsense, straight up facts & figures guy with the penchant for explaining concepts so nicely !! Kudos man..i always look for your reviews on a unit to actually gauge a product !! Keep being awesome 😎
I use a Zoom F3 all the time for video production. The part where you don't have to make a critical decision about levels until post is a game changer for me. The shots I do are improvised one takes and I don't have the ability to manage levels while I'm rolling. There is just too much going on to get levels right without leaving excessive amounts of headroom. Knowing I am going to get clean audio without having to set the gain just right is fantastic.
Granted you have a great experience with Zoom F3 that's great. But there are three things to notice. 1, Everything you have experienced can be done in 24bit. It's the technique that combines two(or more) ADCs' signal makes the difference. 2, The technique itself is not magic either, it's not necessarily different from using two input channels, one at lowest gain, one at 30db or max gain and choose the proper one for proper sections after the recording. 3, It's also similar to have a limited gain settings. 2 adc solution is essentially the same as having on 2 gain settings which means if you record at just over the threshold of the higher gain setting (low noise setting) you would have pretty bad noise performance. Although it's combining two adc signals is better than what I just described. It benefits a lot from using more ADCs, so better solutions use 4 5 or more adcs to accomplish a excellent result. That adds to the cost and price. So you probably would get away with getting a better adc and set the gain conservatively or use two adcs for two tracks with different gains. Also still not outside the capabilities of 24bit format.
@@johnyang799 ti feels kinda pedantic to describe everything as "yeah you could do it even with a rock" -that's basically the feeling it gives- when the person above is clearly saying that the flexibility of having 32 bit is what makes a game changer. what makes you think that they do not adjust the best as possible the gains and everything? the relief is that when you are field recording, or just recording for videos outside or just post production you are just running (and how we say it out here, everything is for yesterday), and you can miss easily a little bit of headroom. it happens even to the best. so, just knowing that even if you didn't manage to get those 3 or 6 db of headroom that you are missing because you couldn't do a proper gain staging and would have been really nice to have now is finally a thing, wouldn't you be happy for that?. sometimes we need to humble ourselves. not everyone can afford the expensive equipment, more if you are just starting. you wouldn't recommend to someone that lives with their parents to buy a neve console right?, that's my problem with your comment. lol.
I think this sort of live production scenario is really the primary reason for 32 bit to exist. It makes a ton of sense on field recorders, but it really is not that useful (except as a bit of idiot proofing) on interfaces for home/studio recording where you can prep everything and leave yourself plenty of headroom. Maybe it's useful for streamers who get a little too excited sometimes, but then again, I think some of them like the "aesthetic" of having things clip when they get too loud, so whatever.
@@nickglover In realistic sense unless you don't want to compress the signal in anyway, a compressor or an AGC(automatic gain control) circuit would work just fine.
Typically I don't think people do want to do that, especially for, say, a movie. You can't undo compression that's baked into the file, so you don't want anything affecting it. Similarly, you can't get an actor to perfectly redo that one perfect take that you messed up because it clipped a bit, but if you have 32 bits, you won't run into that problem.
24 bit converters gave us a solution we’d all been awaiting. When recording with 24bit converters, a -20dBFS peak level is totally usable, which was not the case in 16bit. 24bit recording gave us flexibility at the input, flexibility setting record levels. A 32bit recorder does not add to that input level flexibility as the limits of physics have been met already at the input. We can’t lower the source noise by 48 db and we can’t send 250Vac from our preamps to a DAW. For those of us who have equipment with multi converter and preamp tech such as a few field recorders or the Neumann Digital microphone,the safety extension IS available for 32bit recording. For the other 99.999%, it is not. This distinction is important and often misunderstood. Any situation in which the recording engineer has control (even a second of soundcheck) will return perfect results recording to 24 bits, using 3 or 4 of those bits (18 to 26db) for safety, much like we have always set safe levels, but with far greater room for error. Recording with the same front end (mic, preamp) to a 32 bit file accomplishes absolutely nothing. A 32bit float(ing point) digital file gives infinite range for the most unimaginably clumsy file handling such as very large gain and eq processing IN THE DIGITAL REALM without check the result AND simultaneously erasing the original signal. If these exceptional conditions match your workflow, then 32bit ‘Float’ is manna from heaven. Such situations might arise in field capture for film or TV. In the studio or live concert recording, these actions would disqualify any engineer. In 50 years, I’ve never encountered them.
Interesting reading the comments after they all saw the same thing so well explained, yet many still misunderstand. I doubt we'll ever really get this across, but you tried. Great work Julian!!
Very good explanation Julian, I think 32 bit float will come in handy in converters/interfaces for people using consoles and hardware which can exceed the clipping point in a DAW or driving them hard without needing to use an pad adapter on the direct outs which can lower impedance and muck up the transients. It's also great for less experienced engineers recording drums or recording live shows where volume can get way too high. Cheers!!!
Hey Julian, I just wanted to let you know that I really really appreciate your videos! You are very knowledgeable and the format you make your videos makes it easy to find the information I'm looking for. I spent days looking all over TH-cam and the internet to decide on what mic, interface and fethead. Your videos as well as others from creators like podcastage really made my life easier. I was able to see comparisons with real life scenarios and technical data that helped me make a choice. I personally chose the rode podmic, the motu m2 interface and a Triton audio fethead for my "" budget"" recording set up. This ended up being an extremely good combination and with a little bit of EQ I was almost ready. Now I'm picky and need the AC on when recording so sound treatment is necessary. I am setting up a ""studio"" using moving blankets from harbor freight and a garage hanging storage rack from Amazon to have extra storage in my room as well as something to hang the blankets from over my gaming setup. My room is upstairs in a attic like space that has short knee walls so it's a difficult space to use but it's all I got 😅. Anyways thanks again for all the hard work!
I think 32-bit might as well be, now that it's available, a must-have feature for a field recorder, as used for example in journalistic capacity. When you may be recording something that happens exactly once and that you won't be able to re-record. So not having the ability to screw up the gain setting during capture, with gain being done exclusively in post production, is one substantial fewer thing that could possibly go wrong. On the other hand, i don't think it's necessarily a desirable feature for a computer audio interface.
don't clip your 24 bit recordings in the first place. 24 bits is enough for audio recording, imo. 32 is fine if you want to add 25% more file size to your project for negligible returns. some people even record at 96 kHz sampling rate instead of 48. Hey, remember when 16 bit 44.1 was all the rage when EVERY compact disc used that for encoding? Math was used to determine that 16 bits and 44.1 kHz sampling rate was sufficient to capture 99.9% of everything we as humans can possibly hear. So 24 does give you more headroom and lower noise floor and 48 a few more samples of accuracy. So that, to me,, makes good sense. But if you have the $ and think it's worth the extra disk space and transfer times. go for it. I have a 4K TV but I don't need 8K. I won't be using a size big enough to warrant that much detail. To each their own but if your ears can't hear the difference, why pay for something that does nothing beneficial? Lots of people probably have deficient monitors anyway or noisy components/signal chains. They definitely aren't going to benefit. Weakest link concept.
@@SameeraWeerasinghe On POINT! :) Another point... I doubt any pro studio is using 2" tape anymore for multitracking. It makes no sense anymore. The .0001% smoothness of analogue isn't worth sacrificing all the pros/benefits of recording/mixing all in the digital domain. Only analogue stuff you need is just stuff you want. You can literally model just about anything now but I know some engineers and mixers prefer to still use some hardware like compressors and EQs and such. At this point in time, the tech has surpassed our ability to discern the difference so using more than necessary seems like just a waste to me. Someone smart once said use only as much as you need - no more. or something similar. No one waters their lawn with a fire hose and no one cuts their front lawn with scissors. BAM!
Just because the colors represented in a jpg doesn't mean you should shoot in jpg quality. No, I'll shoot in raw and convert to jpg for delivery. Saying "Well you don't need that many colors, because you can't see them anyway." is faulty thinking. I want the extra bit depth for archival purposes, and the high frequency of 96khz, or even more because of filtering foldback. The higher it is, the more transparent the conversion filtering is, again for having the most pristine version for archiving. There's pre and post echo, or the Gibbs phenomenon that higher frequencies actually help with in the time domain capturing at higher frequencies.
@@aaronmarshall That's a good point. As math teachers once said, "Never round off in the middle of a problem." You want to have the highest resolution during the processing, THEN downsample/round. But if you can hear the difference between 48K and 96K, you are one bad mofo!
Thanks. I’ve been deciding between the SSL 12 and Motu M6 interfaces. I’ve decided on the Motu, but I had an unresolved question about whether I was giving up an “advantage” offered by the SSL: 32-bit float. Now I feel better about my decision.
There's another little detail that is not known by most audio people (but very well known by electronic engineers and the like), and that is the limit of thermal noise, or Johnson-Nyquist noise. Audio devices will output at most, let's say, 10V (usually a lot less, more like 2.5v), so that's our 0dBfs, the maximum value we can ever get. At room temperature, the thermal noise for human hearing range bandwidth is about -140dB at most, being quite generous. No matter how perfect and quiet your setup is, the electrons themselves shaking inside your electronic circuits will make MORE noise than the resolution 24bit provides for you. Even in scientific applications, 24dB is more than enough almost all of the time. Even theoretically, you CAN'T EVEN USE the whole 24bits, there's at least 1 or 2 bits of extra headroom.
As further clarification: I'm not considering cryogenic measurement equipment, nor devices specifically created to measure abnormally loud sounds, like field explosions. Those will usually have either measures to reduce thermal nosie, or output way over 10V, or other special measures that most humans haven't even heard about.
Getting to the limits of 24 bits is not as outrageously difficult as you might think. There are various commercial headphone amplifiers posting DR in the 140s of dB these days. The practical limit is about 152 dB at unity gain, though obviously a unity gain amplifier is not the most useful thing in the world. That would be a maximum amplitude of 9 Vrms (a typical value on +/-15 V) vs. an input noise density of 1.6 nV/√(Hz). If you insisted you could squeeze out another 6-10 dB on the bottom end for sure. The limiting factor is converter analog stages, which at best have a +5 V supply available, often less, and being CMOS parts don't sport voltage noise as low as bipolar circuitry would allow. Commercial audio DACs will just about make it to 140 dB with all 8 channels in parallel, and their ADC cousins top out around 130 dB. The combination of both should illustrate why variable input and output gain as well as hybrid ADC concepts are useful to have. Since instantaneous dynamic range comes with a cost in power consumption, modern DACs more on the mobile side of things are pulling some tricks like DRE (automatic output gain switching) or even Class-G (various steps of output stage power supply voltage chosen depending on signal) as well.
For just one example, there's the Topping L50, posting a 145 dB(A) spec. And that seems to be actual instantaneous DR, as the difference between noise floor and maximum output level at low/mid gains actually comes out to over 152 dB(A). Of course nobody needs that much audio DR all at once, it's more a thing of providing ample DR across a wide range of headphone sensitivities, from BA IEMs to HE-6s (a range approaching 40 dB).
the thing is the actual dynamic range of 24 bit or 32 bit float is only 120 dB. Lower than that is phase noise, or simply said intermodulation distortion due to discretization. With test signals it's rather dynamic, but never with complex signal.
The point is not to just record with great dynamic range and resolution. The point is to be able to do a lot of processing without accumulating errors from such sources as round off, overflow during multiplications, etc.
Well explained. I do live multi-channel capture recordings, while I rarely have any quality problems with 24 bit recordings an interface that had 32 bit float preamps would have advantages.
The whole issue really comes down to the microphone. Barely any microphones has the dynamic-range capture capabilities of 16 bit, before the mic will distort/clip the signal. maximum SPL is often combined with the pad, and you have to deduct the self noise as well, so if the gain is set correctly, the dynamic range that the mic is capable of, is typically less than the 96dB of 16 bit, that the maximum loudness is higher than the dynamic range doesn't matter, since you will only capture sounds starting that the lowest possible sound the mic can capture, and that has to be louder when the pad is active. So in most cases 16 bit is enough, with the right gain, since the mic will be the limiting factor. But then you have 24 bit, that I think surpasses the dynamic range of pretty much any microphone, if the A/D converter can even handle that range. So at the end of the day, what you end up with, is at best some very specific situations where 32 bit recorders will be able to rescue really quiet sounds if the gain was set slightly wrong for 24 bit in combination with that mic. But then you will have quiet sound near the noise of the mic, pre-amp and possibly the A/D converter so you can't simply boost it in the way it may look like in the example earlier in this video, as in that case it will not be a clean signal. It will be a waveform barely distinguishable from the noise recorded. You will not be able to really save the recording, for anything but scientific use. And you will be able to clip and distort the signal, as the mic and the recording path will either introduce an upper limit, before you hit the any dynamic range benefits of 32 bit vs 24 bit. And a dual A/D converter setup, could be used to record to 24 bit, as it is only about optimizing for the available dynamic range of the recording equipment, not to extend the dynamic range beyond 24 bit, really as there will be other limiting factors. But I get it, if you are using a field recorder and can't easily adopt the recording for all various uses as you will not actively monitor the sound at all times, it makes sense. For recording planned music, it really only creates unnecessarily big files, where you probably would have benefited from adjusting the gain anyhow in terms of quality of the actual recording. For recording music: Pretty much only marketing hype. For field recordings: possibly useful in some cases. So if you are looking to record music, it makes more sense to consider other aspects. If it is an recording interface, the latency performance, the feature-set, the pre-amp gain, and so on. And if it is more expensive compared to the competition, the money might be better spent on the actual microphone, as that may be limiting the dynamic range quite a lot already, much more than the 24 bit dynamic range limitation.
"Barely any microphone"? How about a large number of LDCs for one? Any self-respecting one these days will have self noise of at most 18 dB(A) (and less than 5 dB(A) at times), while handling 130+ dB SPL or even more with pad. That is well beyond 16 bits. Also keep in mind that microphone self-noise tends to be heavily 1/f in spectrum, so you have to be very careful when comparing it to the generally much flatter noise floors of electronics as numbers can be misleading.
@@PileOfEmptyTapes the max spl is typically the number for when the pad is engaged, an then typically with 20db of pad. You then have to deduct the noise and or resistance of the membrande. Some microphones thus actually have postings for their actual dynamic range, so we can tell the dynamic range they can record, and you will find that number typically is below 100dB.
I am also SO GRATEFUL when I have an audio question and you have a video for it. I know I’m gonna get a real answer to my question and leave feeling a little smarter. Keep up the amazing work!
Hi Julian, i guess the more polemic topic in whole music gear world is if you really **need** a dedicated preamp or not, how different they are from the ones that comes in sound interfaces. Thanks for all the knowledge
I feel like the most useful place for 32bit float is in recorders and run and gun situations. One man crews for instance. Where you might not always have the ability to control levels in real time and have to set-and-forget. Or in situations where there's big differences in levels (think of meeting or whatever with speakers at all sorts of different levels) where you want to normalize the levels.
The biggest advantage for processing audio at 32bit is that you don't need to dither whenever you bounce anything, so you don't increase the noise floor every time you print anything.
@@aceyage teaching sound design I stay with 24 bit, partly because it's more useful to know about setting optimal levels etc. than abandoning control.. same way I don't properly know in new Polo GTI how the headlights work because they take care of themselves. I'm still not sure in relation to 32 bit, if decisions about levels are relinquished, how it affects the long-held understanding of ideal levels at point of recording for mixing. Similarly, with noise, low but existent noise seems ideal, and 32-bit floating point seems to upset acquiring understanding of relevant ratios.
Awesome video, learned lots! So if one records in 32 bit float of talent that has a really deep and bassy voice, would 32 bit float be able to eq that said voice much easier than 24 bit? Or would 24 bit stand the same chance as 32 bit float. Had an interview recently where talent was a very deep and bassy voice, lav on tshirt and recorded in 24 bit. I'm seeing that I can clean it up ok but not great in post. I'm wondering if I had a 32 bit float recorded if I would have faired better? Or is it just a different mic technique?
Thank you for this video 🙏🏽 for a non-audio expert like me, this actually confirms the need for 32bit f considering that in a run-&-gun situation i don’t have to be so meticulous about the audio recording and still have flexibility in post to make audio sound great even though i might have not set gain levels properly or placed the mic too far from my talent. In a perfect world I’d have an audio guy on-set making sure that audio is being captured properly, but in real life and with impatient clients 32bit F comes in handy 😎🙌🏽
It’s worth remembering that 24-bit integer precision gives you over 16 million (2²⁴=16,777,216) possible amplitude levels for every sample. That’s a heap of resolution, and as others have commented, you’ll be limited in practice by other things, such as the noise floor of the source and the analog front-end. Setting input levels is fairly uncritical when you have 24 bits of precision to play with: you can be pretty conservative (leaving plenty of headroom to avoid clipping) with negligible effect on the digital noise floor. One should be more worried about the analog noise going into the converters. As you say, using 32-bit float makes it wonderfully easy to process a signal. There’s an exact mapping from every signed 24-bit integer value to its corresponding 32-bit float-it’s a lossless conversion-and once you have the float, you can scale it up and down wildly without clipping, so it’s great for DSP (though you can lose precision). So, floating-point is great to have “in the box”, but hardly necessary for capture.
Nobody would record with an 8 bit ADC unless they're looking to get some of that sweet sweet quantization distortion. If you're recording a person who whispers and also screams - where the scream is occupying digital full scale (24 bits) and the whisper is ≈ 90dB lower... The whisper is effectively using around 9 bits. A 32 bit recorder would encode that whisper using ≈ 17 bits. These kind of finger wagging summations of dynamic range also fail to consider factors like "headroom" which basically assumes that any recording of a sound is both unpredictable and fraught with technical constraints. So yeah, it's technically true that 24 bits can record from the threshold of pain down to the threshold of hearing, but the lower part of that range will be sitting at or very near a noise floor. Ultimately 32 bit is another tool in the toolbox that when cleverly deployed gives the recordist a bit more leeway to capture something with the highest possible quality
perfectly summed up thx! i've seen a few videos on it here on yt and some guys somehow seemingly misinterprete 32bitfloat. the convenience, as you're calling it, is simply a benefit that's worth something, at least for me. it's probably not a revolution for recording in general, but at least i won't need to level and set up a limiter, plus i can maintain the otherwise clipped signal, it's simply one less thing to keep in mind. i guess there are also still a few recording enthusiasts around, who overestimate the importance for the high art of levelling an audiosignal a little bit, over what they actually record... anyway, greets from outer space!
I always wondered about it since the 32-bit float format has exactly the same precision as fixed point 24-bit PCM: 1 bit of sign and 23 bits of mantissa. Those extra 8 bits for the exponent can't do much lifting, even in a multi-stage recording setup. They're just handy to have as an intermediate during processing.
One thing to note as well: You can't actually get a full 24 bits of info from recordings. At about -123 dB, cosmic background radiation and thermal noise come into play, and with current circuity technology, there is no way to avoid it. So really, the maximum amount of dynamic range you can record is only 21 bits or so. That said, you can synthetically create a wave in a file that exceeds this, but the moment it enters the analog domain, this limit comes into play. 24 bit is plenty. That said, the convenience of using two ADCs to avoid clipping can be pretty nice, but that's about it.
Hi Julian, Thanks for your very instructive videos. The CD industry still uses 16bit/44.1kHz. I produced several CDs in classical music, and the tradition was to record also with 16bit/44.1kHz, by fear of truncation. What do you think/advise about ? Should one continue to record 16b/44.1kHz as CD requires, or increase the rates ? Greetings
On my Daw I record 24bit 48.1 Hz but when I record chamber Orchestra and small acoustic outfits live. I use a Roland VS 100 that records 32bit FP 96k to SDHC format via 2 microphones Recording in stereo. I never clip But I still have to go through with spectral layers or RX 10 and remove Audience cell phones Ringers etc. Thank you for the video with explanation
A very good and voodoo-free explanation as usual. What could have been stressed out in this video though is that actually also integer based 24 bit PCM is some kind of "convenience" to some extend when it comes to pure recording and playback / distribution as here, the SNR of linear 16 bit PCM is already extremely high and virtually never utilized. Further processing approaches excluded of course. It would be nice to see another "unagitated" video about the floating point format itself and the - as far as I understand it - inherent decoupling of signal to noise ratio and dynamic range which usually go hand in hand with linearily used integer variants.
Excellent presentation; very helpful to me,and wonderful English! It's probably the best I have heard for some time from a native speaker of Deutsch! Lass es Dir gut gehen!
Very good video, Julian! Yes, I agree, 32-bit float is not a magic, but I don't agree it is not a "game changer". As sound designer and reporter working in the field, I love the freedom which I achieve with Zoom F3 or Tascam X8 recorders. I don'r care if levels temporarily peak above 0dB - I still have a good recording and I'm able to restore the proper levels in post-production. In the field you cannot fully prepare for changes in the environment - some of changes are too quickly to re-set the gain to "safe" value, and as you mentioned in the material, setting this level too low in 24-bit (to be safe "in the case") decreases the dynamic range. From this point of view, 32-bit IS very convenient and makes the life easier :)
Why everyone seems to tie 32bit float to the technique that combines multiple adcs signal together? All you have experienced can be done in 24bit format. It's the combining two adcs' signal that makes the difference.
@@johnyang799 Of course. But multiple adcs are available only in 32-bit recorders. And some of these recorders has ONLY 32-bit format available. That is why :) On the other hand, if you have two or more adcs, you can process more than 144dB (just in case) and for that you have to use 32-bit. Do you need more than 144dB? Not really, but some mics have larger dynamic range, so you are able to catch whole the signal from them (e.g. Shure Beta 52A can handle 174dB SPL)
@@GadesChannel @GadesChannel Yea. I'm not saying the actual products are not good nor the fact you shouldn't use it. It's just that either good or bad marketing people tie the two together tightly. btw there's also 32bit fixed point which is different from 32bit float.
@@GadesChannel well also being able to handel 174db spl is not remotely the same as 174 dynamic range. And over 140dbspl in real world is extremely unrealistic. Beyond that you can combine two 24bit tracks to create more than 144db dynamic range. That's also part of the reason daw runs much higher precision underneath.
@@johnyang799 Yes, there is 32-bit fixed, I'm aware of that, but for now I never used any recorder using this format :) And again, in my opinion, using multiple adcs without 32-bit float is not practical. Together they are very convenient even if you in theory can "fit" combined audio from multiple adcs in 24-bit dynamic range. BTW. Do you know ANY recorder/interface which has multiple adcs per channel/track and has only 24-bit? I don't.
I'll pass and wait for the 64-bit double recorders. I know you are going to say nobody will ever need it. Neither were 1080p, 4K, and 8K video formats needed, and yet those are needed now. 🙃
Hi Julian, Thank you for your great videos. Between the solid-state logic SSL 12 and the Motu Ultralite mk5, which one should I buy if my primary use case is recording with the Shure Sm7B? Can you suggest the best audio interfaces under $700 for the SM7B?
@@joesalyers That's not quite right. The SSL12 is only similar in appearance to the smaller models. It has a different converter, can do 32bit, has a higher dynamic range, etc. I'd be curious to see how it performs in a test.
For me the main selling point of 32-bit float comes when combined with DSP's, there are a number of mics and interfaces that within the DSP are able to process the audio in 32-bit float. If I have my mic setup such that it has a good output level close to where I wish to record to, if I record in 24 bit float, and suddenly get much louder I'd risk clipping. However within the compressor of the DSP it has access to more information and when it detects levels that would normally clip it can automatically lower the gain digitally without having lost data. In this way I can record into a 24-bit float format on my computer, and also not worry about clipping because prior to reaching my computer the processing on the DSP has already made sure the signal is within a good range. This is why a lot of mics claim to not clip, because they have built-in processing that takes advantage of 32-bit float to keep the levels in a safe range. As was stated in the video the main benefit of 32-bit float comes in audio processing, but audio processing isn't just post-processing, there's also pre-processing, and many mics and interfaces that advertise anti-clipping do preprocessing with their 32-bit float before it even reaches the device.
bro, imagine a camera operator with a lot of stuffs to control, snario, direction photography, actors/models, and still have to think about the audio, I dont want to check if the file on the set if going well, I just want get the audio and make it works on the pos production, so, it's a game changer for run & gunprofessionals for sure
32 bit float in field recording makes a lot of sense when you don't know how loud things are going to get. If you're recording audio for a movie or game, and you're going to blow things up, you're not going to get many chances to set the levels.
All good. I use Logic Pro, which uses the 32-bit float domain for built-in samples and its built-in plug-ins, and a lot of my tracks are sample-based, so I think it actually may also have an advantage over 24-bit. But see if you can set me straight on this: While the dynamic range of 24-bit is fine if used properly for a single recorded track, what if you have 100 tracks, and you set a lot of headroom, just to be careful to not exceed 0 DBFS? Does that not raise the cumulative noise floor? It's the accumulation of 100 noise floors. That means the effective aggregate noise floor is a lot higher than -144 dB. So is that significant, or not? Would this give 32-bit float an advantage over 24-bit in mixing/mastering? Also, the digital noise floor for 24-bit is made of quantization noise, which if amplified, sounds like a buzzsaw. The analog noise floor is essentially hiss from thermal noise, which is far less annoying to human ears. Logic also uses something called '64-bit summing', which I can't seem to find much info on. I'd love to see a video from you on that.
Thanks Julian. I think the convenience is good for me as an amateur producer honing my tracking skills. Using Cubase in 32float recently was fun. I felt/held the full signal and used the 'clip gain' feature to easily eliminate the clipped sound. Previously, I would have had to re- record. I agree its not a need but a huge WANT.
Hi, Julian, I`ve seen a few of your videos about interfaces, can you pls advise me, do I need to change NI KA6 mk1 interface? Are modern interfaces of the same level better in preamp and AD/DA quality?
I think 32 bit is the great Equalizer. It makes Audio less intimidating for new entrants in the Audio space and gives them the opportunity to save time going through the learning curve to be able to get the same results with 24bit as some audio guy who had to learn 3 years to get it right. With 32 bit, they can focus on others aspects of their productions more efficiently as well
I think 32 bit float is extremely useful for a personal recorder like my Zoom F2, when you need to set up a lapel mic on someone and you have no idea what the sound levels will be.
Agreed. I have 32bit float on my Tentacle Track E bodypack recorder, and also on my Sound Devices MixPre 6II recorder. While I always do a sound check... it's nice to know that I am somewhat protected from clipping if the subject gets too loud. I often hook my MixPre recorder to a sound board backstage at an event, while I'm in the audience shooting video. Therefore I'm not able to monitor the audio. But the files I record can be adjusted later in post if things are too loud or too soft. It's a nice safety-net for us one-man-band shooters. 😎
I work almost exclusively in 24-bit. The only time this changes is if I'm delivering multitracks or stems to someone; I prefer to render to 32-bit float so I don't have to dither. I always assume the other party _will_ dither, so I don't want to add noise twice. This may not even be sound logic, but I get peace of mind and I've never had any complaints.
Please correct me if I'm wrong. I see a difference between bits required to represent the full dynamic range of audio and bits required to capture the dynamic range of audio with our devices. While 24 bits can represent all the dynamic range our ears can tolerate, in order to record it we have to capture audio above the noise floor of our signal chain. This is why instead of recording low level input with lots of headroom, we try to record at levels closer to the top and risk clipping. Ideally we could record with an average signal of -12 dBFS and never go over 0, but in practice this can be risky. We can record at a lower input level, but then raising the signal in post production we will also be raising the noise floor of our signal chain. A benefit to recording in 32 bits is setting your recording level so your average signal is higher above the noise floor. The extra headroom of 32 bit float may save the day when there would otherwise be clipping.
First, you're the best!! I understand that one can get the advantages of 32bit Float, if at all, regardless of type of analogue signal including Mic, Instrument, or Line level as long as the signal has not been through A/D conversion in the signal path when it reaches the 32bit Float interface . Is this correct?
And thus begins a new generation of hardware pushed and advertised to end users by the manufacturers, advertised as the new ‘industry standard’ and ‘must have’. All companies will follow suit for fear of being left behind. They will seek big named artists and engineers to help sell their stuff. And before long, the art of gainstaging is all but a distant memory. Why do we fall for this time and time again. Just because something is technically an improvement on ‘paper’, doesn’t mean ANYTHING for the majority of amateur users who actually need to spend their time and energy on their craft - not new gear. The most important point in this video ‘shit in, shit out’. Don’t fall for the hype kids. Learn your trade as a musician, recording engineer, mix engineer… learn a bit of physics theory behind the electronics… even if you stay firmly in your DAW, understanding WHY your vintage plugins work the way they do will help you apply solid PRINCIPLES to all your future work. Remember, we’re just shoving electrons around until eventually we shove some air around at the end of it!
Hi, many thanks for this Video! I am new to Audio recording. I have only an old Rode Videomic pro and am intrested to buy a 32bit float audio recorder from Tascam (DR 10L -PRO) I always read they are used with Lav-Mic's.. But in theory i could attach the shotgun mic with it's preamp set to 20dB, to the Tascam DR 10L-PRO to record and bring it down in post? Or would the preamp from Rode Mic fry the preamp from Tascam? While the 32Bit Container/Format could handle it, the electronics may be the bottleneck here?
Developer here: Bitdepth just influences the noisefloor's depth. You don't hear a difference between 16 and 24bit and definitely none of the higher ones as well. but a lot of your plugins might internally convert 24bit inputs to higher ones to prevent mathematical errors on extremely small values, which matters for example in filter design or extremely slow LFOs, so when your project is already at a higher bitdepth those plugins don't have to convert anymore. it's not that this conversion is super cpu-inefficient though. i suppose almost no difference. so whatever you decide, it's alright.
Been following your videos for a long time and i just want to say you're the best. Also i wanted to know of the possibility of doing a video of interfaces with great RTL times
Thank you for explaining this. I am still curious about whether 32 bit float creates more safety around sudden spikes in db, like a sudden loud laugh. That type of sound makes me wonder whether 32 bit would be helpful.
I feel like the best use case scenario for 32 Bit Float is field recordings in compromised situations. Run and gun or novice setups. Using it for mobile video production is kind of a game changer.
There is one difference mentioned here: dual amp means than at some point the amp can switch frol one to the other. this is audible with ultrasonics content where you can have sudden noise burst in ultrasonics content, which is why I hear people recording with ultrasonics still seems to prefer 24 bits. So thing is recording with dual amp do have some unexpected consequence.
Let´s say its the other way arround more important so with 32Bit, you can increase very silent mixes without yielding with unpleasent artifacts. If you do that with 16Bit, you destroy the mix, if you do with 24Bit, it can be awkward, if you do with 32Bit, most of the sound textures stay clean. So yes, then you can send the mastering engineer, he can down dither to 24bit or 16Bit then at his mastering. By the way the precision is much much higher and I see no point, why people question it, as it is just a much more sharper knife in your kitchen. Of course yes I love it and I take it. Thanks and best Regards
thanks.. great video.. so: do I need a 32 bit interface: no do I need 32 bit wav storage in my DAW :well no, but if you can you should enable it because it can save you in some circumstances, and you dont need to worry about recording too hot in your DAW. Got it.
Hello Mr Krause, Your helpful addvice helped with me getting a MOTU M4 which i love to bits ❤. It was to replace a Behringer UMC204, (becuase no new drivers, it wouldn't talk to my PC) the Behringer UMC204 was powered through the USB i took a chance and plugged it into a phone charger and it looks like it works well in all aspects other than interfacing with computers. For example can run a SM57 mic through it, out the signal trough a guitar pedal on the insert, and hear the result through the headphone socket. Not sure if this wold work for most audio interfaces (scared to try with my MOTU M4). Please say if it is. Is very useful for super cheep preamp with extra features.
When everything was 16 bit, gain settings were crucial because of the dynamic range limit. With 24 bit it was widened. The remaining problem with 24 bit was that if you recorded a low signal, you may have to boost the gain digitally in post (say, if it's a single source like a voice). Boosting or normalizing brings with it the inherent noise in the signal. So it's not that input gain level doesn't matter in 24 bit. You still want to capture a signal as strong as you can get while maintaining plenty of headroom for clipping. With 32 bit, this problem seems to be eliminated.
@@diktor_ilyakhaiko As far I know many if not most 32 bit devices have fixed gain and input sensitivity anyway. What one is changing with the knob, especially in ones that have dual DACs (this 32bit float), is simply the structure of the file. And how can it record low signal without noise? Well the more bits You have the more of them are usable aka not post to noise. And simply the amount of possible values is waaaay greater. 24 bit is 16 777 216 possibilities 32 bit is 4 294 967 296 possibilities You can fit entire 24 bit range 256 times inside of 32 bit range. Just think about it, even if You have 100 gain/volume settings on 24bit ADC that still less than how many times 32 bit can overshadow it.
For the end product you convert the audio file to 44.1 kHz 16bit 256 kbps AAC or OPUS (← 48 bit). The very little quality loss is worth the tremendous amount of saved space. But it's always convenient to keep lossless high quality audio as a source file for editing purposes.
I'm wondering based on your older video on preamp gain effecting the level of the noise floor in a recording that if you intentionally place the gain on a 32bit recording device (i have the Zoom F6) between the two preamps and lower it in post, do you get less preamp noise? Is it just two halfs equal 1 situation?
I have recorded album with 32bit float on F6 in 2022 drums, piano, bass, guitars, vocals. And this year 2023 i did another one with my old Zoom R16 in 24bit but i think the most difference there is are the more noisy preamp's Zoom R16 has. With F6 i can boost the top end over 10k high's of a piano without bringing up a bright noise like with the R16 preamps. On this years record i went more mellow, natural sounding overal feel and did not boost those. Overdubbing with F6 is time consuming process couse you have to record with two recorders, one to capture the sync + F6 to get the sound of the preamps and sync those tracks in post but the result sounds really good and there was some depth i could hear in the sound
One thing I do not understand. 32 bit audio interface (steinberg ur22c for example) will use those 32 bits for recording only? If I won't record anything and just want to make music in Ableton in 32 bits - do I need a 32 bit audio interface for listening to what I do? Or I can have 24 bit interface and still produce in 32bit (i.e. interface bits and ableton bits are not related)?
It sounds like a good idea when recording audio along with video because it's hard to take care of audio levels when you are trying to record video as well so it's easy for distortion to happen that can't be fixed later. With 32 bit float you can fix the problem later and reduce the signal to below the distortion.
Yes, and 24 bit is much harder to clip than 16 bit, so 24 bit would be a preferred recording format. 32 Bit float only has an advantage if you record something with very large volume fluctuations, make it hard to set the gain. If a volume is consistent, like with recording a person speaking during an interview, you can set the right gain with 16 or 24 bit, and there will be no advantage using 32 bit. You would be hard-pressed to find a 16 bit recorder these days, almost all have 24 bit at minimum.
For field recordings 32bit float has all advantages as 24bit and when you blow a recording u just pull it down to 0 or less Or if the signal is very weak u can pull it up50-60 db without adding noise
I’ll never buy a 24 bit recorder again! The easiness knowing that u do not blow out recordings when using lots of money and energy getting places like the arctic or Antarctica is good enough. Our zoom f4, h5 and H6 are all in the shelf!! F3, F6 F2! and mixpre3 ll are the tools and they work all well with em272 and Sennheiser em66 and 62 Load up, normalize snd start editing….
The issue I've been trying to understand, but haven't really seen discussed is how, if it all, 32 bit float is applicable in a live format like streaming. For example, I have the Tascam Portacapture X8, but haven't used or explored it much yet, but if I was using it as an audio interface to my computer, and then through OSB or something similar to live stream, would there be any advantage to 32 bit float at all?
No real benefits with 32 Bit float there. Most streams dont use 32bit float. The audio Stream has to be encoded into a compressed format such as aac. Even if it would be possible to stream in 32 bit float the people would still have to convert it back into an analog signal to send it to speakers or headphones. Most of which dont have the dynamic range of even 16 bit. Also from a listening point of view most people dont seem to like it if you yell at them with 140dB SPL right after you have whispered into there ears 😅. The only benefit could be if you have a 32 bit digital signal processor(hard or software) before the stream in which you could apply effects such as compressor and limiter: you dont have to worry as much about the settings and internal levels exceeding 0dBFS as long as the output doesnt.
Live streaming tends to require a tightly managed output dynamic range (since your bottleneck tends to be viewers' playback equipment), and the kind of dynamic range you can actually capture on the input side tends to be small enough for 24 bits either way. The biggest difference between int24 and float32 would be the input levels that the compressor sees.
With 32 bit float, say the audio was recorded, quite low, so -20 DB and you wanted to bring it up to 0DB is there. Also a noise floor that is brought up when you increase the audio volume as well?
So, 32bit fixed point audio interfaces vs 32bit floating point interfaces? Is that a thing? The hardware has to be 32bit floating point for the recorded file in the DAW to have this dynamic range...is that correct @JulianKrause ?
For me, 32-bit float recording is very helpful because I record live events where the audio level can drastically change at any moment. One second we may need to pick up someone whispering, and the next there could be thunderous applause. I don't have the luxury of changing the record volume on a moment's notice. I think this is the kind of situation where it comes in extremely handy.
Definitely a good example of a situation in which I would think 32 bit FR would be a blessing to have. As well as ambient environments like city streets in which many varieties of sound are coming at the mic from varying distances and decibels.
Hey Julian. So i have a wish in case others find it interesting too and most of all if you feel like makinga video about it. It is about D/A converters. I frequently hear people for years now, argue how their 1000+ dollars interfaces (RME, Universal you name it) sound „not good enough“ and that a much more expensive „super“ converter like a Lynx or Crane song etc will make a „huge" difference. I somehow do not believe that in converters of our time in history there can even be a huge differences but i cant back it up on a technical level as you would be able to do. So i would be very interested in a comparisson of a high class interface to a high end „professional“ grade converter with actual measurements of „What goes in, What comes out and What the actual difference is between them beyond Esoteric interpretations“
Float means "floating (moving) decimal point". Thinking decimal: imagine a 7 digit number with floating decimal point - we can write numbers from 0.000001 to 1,000,000. Traditionally we use integer numbers for 16 and 24 bit depths and, once again thinking decimal and using a 7 digit number we can go from 0000001 to 1,000,000 - that's because integer numbers don't allow for fractions. Let's compare how much larger the second number is compared to the first one in both cases: 7 digit decimal float: 1,000,000 divided by 0.000001 equals 1,000,000,000,000 7 digit decimal integer: 1,000,000 divided by 0000001 equals 1,000,000 Comparing how wider is the float range compared to the integer range (same number of decimal digits¹): 1,000,000 divided by 1,000,000,000,000 equals 1,000,000 - using float allows for a range (from smallest or quietest to largest or loudest) a million times greater. 1) I used decimal digits instead of binary digits (a.k.a. bits) because it's easier to understand. Remember that floating point is a bit more complicated than what I did but I think keeping it simple helps to understand why floating point gives a larger dynamic range than using integer. I did not talk about dBs because I know very little about it. Thanks, Julian! I finally understood how 32bit float works and how it can be implemented using less than 2^16 16bit ADCs!
I’m gonna give you two use cases where 32 bit float reigns supreme on location - shooting ranges and loud concerts front row trying to get that shot. Not having to worry about levels on the fly is the best. Buying a larger SF card is the smallest price to pay.
Question no one has answered for me yet. On windows in the bottom right sound properties next to clock there is a tab called recording. If I right click my mixer in there hit properties there is a recording level setting. Should I ever touch this? Reason: If I set my gain on mixer just right where the clip light rarely comes on and keep my mixer volume at +-0 but not adjust system recording level. Then I end up with extremely loud clipping audio. So I have nice clean audio going into computer (monitoring with headphone) but unless I lower one of the following I have issues. Gain/Mixer volume slider/system recording level. Which is better to lower or am I doing something completely wrong?
As an Adobe Audition user your aware that its multitrack sessions default to 32 bit. I have never questioned that default but now I wonder if I even need it? Most of my mixes are very long - around an hour - and they can take several minutes to mixdown. If 32 bit is doing nothing for me in the mix yet is slowing down my mixdowns, would it make sense to mix in a lower bit rate?
Hello Julian! I'm currently trying to use my Zoom F3 as 32 bit audio interface, but audition is still clipping my recording on 0db. I downloaded specific driver from zoom website, updated firmware on the recorder which is adding 32 bit AI mode, chose ASIO in audition, but it didn't help.
I often increase or decrease the speed of videos that I watch, and this introduces various forms of audible signal processing anomalies in the playback. Would using 32 Bit Float improve or affect the playback quality in such a scenario?
It wouldn't affect in that sense. The artifacts you hear are most likely caused by the algorithm used to keep the original pitch of the sounds. If you just increase the speed by brute-force (playing every sample at twice the speed), everything would sound high pitched (Nightcore music takes advantage of this fact). But you don't want chipmunk voices in Julian Krause's new video when watchign at 2x, so an algorithm is introduced that keeps the pitch and the pronuncioation of the letters, but increases the speed. This is not perfect and introduces a lot of artifacts (and it's usually optimized to mostly keep human voices and words, so other sounds might get more distorted)
Julian has a great way of breaking down a complex subject that makes it easier to understand. While also retaining the relevant details one should be considering.
Sehr gerne, Julian. Gute Arbeit!
Always set my DAW to 32bit float & never have to think about it. Storage is so inexpensive nowadays. Great video!
Whats your reason?
I use an F6 and an F3 and the convenience is fantastic! That said - I learned about 32 bit float recording's limitations the hard way. I was working with a microphone that had a max SPL of 110 (cough cough Videomic Go 2). No matter what happened, whether I used the Zoom F3, Zoom F6 or Zoom F2, the audio would distort when I got really really really loud. On the meter it showed normally but the audio was distorted. (For those wondering it wasn't a DAW issue, it's an SPL thing). You are limited to the SPL of the microphone. This also explains the few times I actually managed to clip even on an F6 with a Neat King Bee version 1. A lot of the marketing around 32 bit float recorders is a bit hogwash :P You'll often see the ever-popular diagram of the 24 bit graph with a small tiny little grey arrow that goes upward and the 32 bit float recording diagram with a grey arrow up to 600 DB. First off, you'd literally die if you recorded at 600 DB :P. Second of all, most mics are capped off at around 145 DB SPL so regardless of how awesome 32 bit float recorders are - you are capped off at the max SPL of the mic. That being said, what I can say - and Julian touched on this - is convenience. With a 32 bit float recorder I can do things 4-5 times faster than before. YOU SKIP STEPS. I no longer have to do things anymore. It's easy. No more fiddling with things. No more hiss-penalty, no clip-penalty, it's like the cheat-code in audio. No quarters needed, just infinite lives.
You are right about SPL of the Mic being a limiting factor, but about 32bit float Mic, a USB one, like Rode NT1 5th gen? With this mic you are not capped, right?
@@diktor_ilyakhaiko The 5th Gen NT1 has a max SPL of 142 DB. In other words, it's not as high as stated. That said - no one really gets that loud. But even 24 bit caps at around 144 DB. You'll clip out the mic before even touching the capability of 32 bit float. That said, in like 99% of cases you won't clip the mic as 142 DB is quite loud.
@@Dracomies I see, it is just physics, yeah? If you want to have more SPL, you must do something specific with a capsule, right? And not just add digital 32 bit float function.
I thought 5th gen has 142 dB SPL as a XLR mic, but as USB it has more, like 32 bit.
Yeah, that mic is pretty limited. If you go with the VideoMic Pro it at least gives you three different options for gain, -10dB, 0dB, and +20dB. That can come in very handy depending on the environment in which you're recording.
Assumptions are being made here that are not correct. You say “no longer have to do things. SKIP STEPS. No more hiss penalty. No more clip penalty.” That is for most situations completely incorrect. You still have to fit the microphone’s analog input signal into a window the bottom of which is the electronic noise floor of that microphone and top of which is distortion generated by that same microphone. If you don’t, you will get a digital signal with infinite range playing back the same hiss and distortion you fed it. This is not cheat-code. It is not magic.
Started seeing 4th-Gen Focusrite Scarlett reviews popping up on TH-cam... But i'm personally holding out for Julian Krause's review, & ONLY Julian Krause's review! 😁
Nice coverage, Julian, very well explained. 32-bit float is useful if you wish to push decision-making and work to post production, assuming that your microphone can handle the SPLs you record. I'm with you, 32-bit float is very rarely useful for my work. In fact, if I'm using a channel strip or interface or recorder with a compressor, I can get closer to where I want my final audio to sit with 24-bit and save myself some work in post.
@curtisjudd so limitations are given by the mic’s SPL. But how would you control your gain, and know you have enough, not being too quiet, when using zoom f3, for instance? Also, does f3 have direct monitoring?
@@thesilencer6736the F3 does not have an analogue gain setting. Yes, the F3 has a headphone jack for monitoring.
A great explanation of the differences between/advantages of 24bit vs 32bit recording. I just got my first 32-bit float capable recorder. My reason is that as a solo performer, when I'm recording my gigs, I have to put the recorder out front, set the level, and hope for the best that I don't overload the signal. I've had some potentially great recordings ruined by clipping because things got too hot for the actual recorder. With 32-bit float, I still try to set things as best as possible, but I have the sort of added insurance that I won't clip the recording. But I agree that it's not a magic bullit, as you can still clip the signal before it gets to the recorder. Good recording technique is still needed.
Yes if the audio is so loud that it clips the mics or inputs but with speech or unamplified audio this is unlikely and recording in 32 bit float is a great safety feature
Julian, you are the no nonsense, straight up facts & figures guy with the penchant for explaining concepts so nicely !! Kudos man..i always look for your reviews on a unit to actually gauge a product !!
Keep being awesome 😎
I use a Zoom F3 all the time for video production. The part where you don't have to make a critical decision about levels until post is a game changer for me. The shots I do are improvised one takes and I don't have the ability to manage levels while I'm rolling. There is just too much going on to get levels right without leaving excessive amounts of headroom. Knowing I am going to get clean audio without having to set the gain just right is fantastic.
Granted you have a great experience with Zoom F3 that's great. But there are three things to notice.
1, Everything you have experienced can be done in 24bit. It's the technique that combines two(or more) ADCs' signal makes the difference.
2, The technique itself is not magic either, it's not necessarily different from using two input channels, one at lowest gain, one at 30db or max gain and choose the proper one for proper sections after the recording.
3, It's also similar to have a limited gain settings. 2 adc solution is essentially the same as having on 2 gain settings which means if you record at just over the threshold of the higher gain setting (low noise setting) you would have pretty bad noise performance. Although it's combining two adc signals is better than what I just described. It benefits a lot from using more ADCs, so better solutions use 4 5 or more adcs to accomplish a excellent result. That adds to the cost and price. So you probably would get away with getting a better adc and set the gain conservatively or use two adcs for two tracks with different gains. Also still not outside the capabilities of 24bit format.
@@johnyang799 ti feels kinda pedantic to describe everything as "yeah you could do it even with a rock" -that's basically the feeling it gives- when the person above is clearly saying that the flexibility of having 32 bit is what makes a game changer. what makes you think that they do not adjust the best as possible the gains and everything? the relief is that when you are field recording, or just recording for videos outside or just post production you are just running (and how we say it out here, everything is for yesterday), and you can miss easily a little bit of headroom. it happens even to the best. so, just knowing that even if you didn't manage to get those 3 or 6 db of headroom that you are missing because you couldn't do a proper gain staging and would have been really nice to have now is finally a thing, wouldn't you be happy for that?.
sometimes we need to humble ourselves. not everyone can afford the expensive equipment, more if you are just starting. you wouldn't recommend to someone that lives with their parents to buy a neve console right?, that's my problem with your comment. lol.
I think this sort of live production scenario is really the primary reason for 32 bit to exist. It makes a ton of sense on field recorders, but it really is not that useful (except as a bit of idiot proofing) on interfaces for home/studio recording where you can prep everything and leave yourself plenty of headroom. Maybe it's useful for streamers who get a little too excited sometimes, but then again, I think some of them like the "aesthetic" of having things clip when they get too loud, so whatever.
@@nickglover In realistic sense unless you don't want to compress the signal in anyway, a compressor or an AGC(automatic gain control) circuit would work just fine.
Typically I don't think people do want to do that, especially for, say, a movie. You can't undo compression that's baked into the file, so you don't want anything affecting it. Similarly, you can't get an actor to perfectly redo that one perfect take that you messed up because it clipped a bit, but if you have 32 bits, you won't run into that problem.
For perspective, a freshly aligned 24 track 2” machine with Ampex 996 is around 11-12 bits of dynamic range.
Thank you for calmly explaining what I've been screaming about in comment sections for years.
24 bit converters gave us a solution we’d all been awaiting. When recording with 24bit converters, a -20dBFS peak level is totally usable, which was not the case in 16bit. 24bit recording gave us flexibility at the input, flexibility setting record levels. A 32bit recorder does not add to that input level flexibility as the limits of physics have been met already at the input. We can’t lower the source noise by 48 db and we can’t send 250Vac from our preamps to a DAW.
For those of us who have equipment with multi converter and preamp tech such as a few field recorders or the Neumann Digital microphone,the safety extension IS available for 32bit recording. For the other 99.999%, it is not. This distinction is important and often misunderstood. Any situation in which the recording engineer has control (even a second of soundcheck) will return perfect results recording to 24 bits, using 3 or 4 of those bits (18 to 26db) for safety, much like we have always set safe levels, but with far greater room for error. Recording with the same front end (mic, preamp) to a 32 bit file accomplishes absolutely nothing.
A 32bit float(ing point) digital file gives infinite range for the most unimaginably clumsy file handling such as very large gain and eq processing IN THE DIGITAL REALM without check the result AND simultaneously erasing the original signal. If these exceptional conditions match your workflow, then 32bit ‘Float’ is manna from heaven. Such situations might arise in field capture for film or TV. In the studio or live concert recording, these actions would disqualify any engineer. In 50 years, I’ve never encountered them.
The very first video about 32Bit float i can fully agree with! And yet simple and short!
Yeah
Interesting reading the comments after they all saw the same thing so well explained, yet many still misunderstand. I doubt we'll ever really get this across, but you tried. Great work Julian!!
Very good explanation Julian, I think 32 bit float will come in handy in converters/interfaces for people using consoles and hardware which can exceed the clipping point in a DAW or driving them hard without needing to use an pad adapter on the direct outs which can lower impedance and muck up the transients. It's also great for less experienced engineers recording drums or recording live shows where volume can get way too high. Cheers!!!
Hey Julian, I just wanted to let you know that I really really appreciate your videos! You are very knowledgeable and the format you make your videos makes it easy to find the information I'm looking for. I spent days looking all over TH-cam and the internet to decide on what mic, interface and fethead. Your videos as well as others from creators like podcastage really made my life easier. I was able to see comparisons with real life scenarios and technical data that helped me make a choice. I personally chose the rode podmic, the motu m2 interface and a Triton audio fethead for my "" budget"" recording set up. This ended up being an extremely good combination and with a little bit of EQ I was almost ready. Now I'm picky and need the AC on when recording so sound treatment is necessary. I am setting up a ""studio"" using moving blankets from harbor freight and a garage hanging storage rack from Amazon to have extra storage in my room as well as something to hang the blankets from over my gaming setup. My room is upstairs in a attic like space that has short knee walls so it's a difficult space to use but it's all I got 😅. Anyways thanks again for all the hard work!
I think 32-bit might as well be, now that it's available, a must-have feature for a field recorder, as used for example in journalistic capacity. When you may be recording something that happens exactly once and that you won't be able to re-record. So not having the ability to screw up the gain setting during capture, with gain being done exclusively in post production, is one substantial fewer thing that could possibly go wrong.
On the other hand, i don't think it's necessarily a desirable feature for a computer audio interface.
Great videos, they have helped me make wiser purchases!
Thanks for the support Joe, glad the videos help you out :)
@@JulianKrause Just ordered a MOTU Ultralight to replace my redundant RME Fireface. Your video gave me confidence that it does everything I need!
don't clip your 24 bit recordings in the first place. 24 bits is enough for audio recording, imo. 32 is fine if you want to add 25% more file size to your project for negligible returns. some people even record at 96 kHz sampling rate instead of 48. Hey, remember when 16 bit 44.1 was all the rage when EVERY compact disc used that for encoding? Math was used to determine that 16 bits and 44.1 kHz sampling rate was sufficient to capture 99.9% of everything we as humans can possibly hear. So 24 does give you more headroom and lower noise floor and 48 a few more samples of accuracy. So that, to me,, makes good sense. But if you have the $ and think it's worth the extra disk space and transfer times. go for it. I have a 4K TV but I don't need 8K. I won't be using a size big enough to warrant that much detail. To each their own but if your ears can't hear the difference, why pay for something that does nothing beneficial? Lots of people probably have deficient monitors anyway or noisy components/signal chains. They definitely aren't going to benefit. Weakest link concept.
On spot! Cheers
@@SameeraWeerasinghe On POINT! :) Another point... I doubt any pro studio is using 2" tape anymore for multitracking. It makes no sense anymore. The .0001% smoothness of analogue isn't worth sacrificing all the pros/benefits of recording/mixing all in the digital domain. Only analogue stuff you need is just stuff you want. You can literally model just about anything now but I know some engineers and mixers prefer to still use some hardware like compressors and EQs and such. At this point in time, the tech has surpassed our ability to discern the difference so using more than necessary seems like just a waste to me. Someone smart once said use only as much as you need - no more. or something similar. No one waters their lawn with a fire hose and no one cuts their front lawn with scissors. BAM!
Just because the colors represented in a jpg doesn't mean you should shoot in jpg quality. No, I'll shoot in raw and convert to jpg for delivery. Saying "Well you don't need that many colors, because you can't see them anyway." is faulty thinking.
I want the extra bit depth for archival purposes, and the high frequency of 96khz, or even more because of filtering foldback. The higher it is, the more transparent the conversion filtering is, again for having the most pristine version for archiving. There's pre and post echo, or the Gibbs phenomenon that higher frequencies actually help with in the time domain capturing at higher frequencies.
@@aaronmarshall That's a good point. As math teachers once said, "Never round off in the middle of a problem." You want to have the highest resolution during the processing, THEN downsample/round. But if you can hear the difference between 48K and 96K, you are one bad mofo!
New focusrite 4th gen Scarlett’s out, looking forward for review
Thanks. I’ve been deciding between the SSL 12 and Motu M6 interfaces. I’ve decided on the Motu, but I had an unresolved question about whether I was giving up an “advantage” offered by the SSL: 32-bit float. Now I feel better about my decision.
There's another little detail that is not known by most audio people (but very well known by electronic engineers and the like), and that is the limit of thermal noise, or Johnson-Nyquist noise.
Audio devices will output at most, let's say, 10V (usually a lot less, more like 2.5v), so that's our 0dBfs, the maximum value we can ever get. At room temperature, the thermal noise for human hearing range bandwidth is about -140dB at most, being quite generous. No matter how perfect and quiet your setup is, the electrons themselves shaking inside your electronic circuits will make MORE noise than the resolution 24bit provides for you. Even in scientific applications, 24dB is more than enough almost all of the time. Even theoretically, you CAN'T EVEN USE the whole 24bits, there's at least 1 or 2 bits of extra headroom.
As further clarification: I'm not considering cryogenic measurement equipment, nor devices specifically created to measure abnormally loud sounds, like field explosions. Those will usually have either measures to reduce thermal nosie, or output way over 10V, or other special measures that most humans haven't even heard about.
Getting to the limits of 24 bits is not as outrageously difficult as you might think. There are various commercial headphone amplifiers posting DR in the 140s of dB these days. The practical limit is about 152 dB at unity gain, though obviously a unity gain amplifier is not the most useful thing in the world. That would be a maximum amplitude of 9 Vrms (a typical value on +/-15 V) vs. an input noise density of 1.6 nV/√(Hz). If you insisted you could squeeze out another 6-10 dB on the bottom end for sure.
The limiting factor is converter analog stages, which at best have a +5 V supply available, often less, and being CMOS parts don't sport voltage noise as low as bipolar circuitry would allow. Commercial audio DACs will just about make it to 140 dB with all 8 channels in parallel, and their ADC cousins top out around 130 dB.
The combination of both should illustrate why variable input and output gain as well as hybrid ADC concepts are useful to have. Since instantaneous dynamic range comes with a cost in power consumption, modern DACs more on the mobile side of things are pulling some tricks like DRE (automatic output gain switching) or even Class-G (various steps of output stage power supply voltage chosen depending on signal) as well.
For just one example, there's the Topping L50, posting a 145 dB(A) spec. And that seems to be actual instantaneous DR, as the difference between noise floor and maximum output level at low/mid gains actually comes out to over 152 dB(A). Of course nobody needs that much audio DR all at once, it's more a thing of providing ample DR across a wide range of headphone sensitivities, from BA IEMs to HE-6s (a range approaching 40 dB).
the thing is the actual dynamic range of 24 bit or 32 bit float is only 120 dB. Lower than that is phase noise, or simply said intermodulation distortion due to discretization. With test signals it's rather dynamic, but never with complex signal.
The point is not to just record with great dynamic range and resolution. The point is to be able to do a lot of processing without accumulating errors from such sources as round off, overflow during multiplications, etc.
Well explained. I do live multi-channel capture recordings, while I rarely have any quality problems with 24 bit recordings an interface that had 32 bit float preamps would have advantages.
The whole issue really comes down to the microphone. Barely any microphones has the dynamic-range capture capabilities of 16 bit, before the mic will distort/clip the signal. maximum SPL is often combined with the pad, and you have to deduct the self noise as well, so if the gain is set correctly, the dynamic range that the mic is capable of, is typically less than the 96dB of 16 bit, that the maximum loudness is higher than the dynamic range doesn't matter, since you will only capture sounds starting that the lowest possible sound the mic can capture, and that has to be louder when the pad is active. So in most cases 16 bit is enough, with the right gain, since the mic will be the limiting factor. But then you have 24 bit, that I think surpasses the dynamic range of pretty much any microphone, if the A/D converter can even handle that range. So at the end of the day, what you end up with, is at best some very specific situations where 32 bit recorders will be able to rescue really quiet sounds if the gain was set slightly wrong for 24 bit in combination with that mic. But then you will have quiet sound near the noise of the mic, pre-amp and possibly the A/D converter so you can't simply boost it in the way it may look like in the example earlier in this video, as in that case it will not be a clean signal. It will be a waveform barely distinguishable from the noise recorded. You will not be able to really save the recording, for anything but scientific use.
And you will be able to clip and distort the signal, as the mic and the recording path will either introduce an upper limit, before you hit the any dynamic range benefits of 32 bit vs 24 bit. And a dual A/D converter setup, could be used to record to 24 bit, as it is only about optimizing for the available dynamic range of the recording equipment, not to extend the dynamic range beyond 24 bit, really as there will be other limiting factors.
But I get it, if you are using a field recorder and can't easily adopt the recording for all various uses as you will not actively monitor the sound at all times, it makes sense. For recording planned music, it really only creates unnecessarily big files, where you probably would have benefited from adjusting the gain anyhow in terms of quality of the actual recording.
For recording music: Pretty much only marketing hype.
For field recordings: possibly useful in some cases.
So if you are looking to record music, it makes more sense to consider other aspects. If it is an recording interface, the latency performance, the feature-set, the pre-amp gain, and so on.
And if it is more expensive compared to the competition, the money might be better spent on the actual microphone, as that may be limiting the dynamic range quite a lot already, much more than the 24 bit dynamic range limitation.
"Barely any microphone"? How about a large number of LDCs for one? Any self-respecting one these days will have self noise of at most 18 dB(A) (and less than 5 dB(A) at times), while handling 130+ dB SPL or even more with pad. That is well beyond 16 bits. Also keep in mind that microphone self-noise tends to be heavily 1/f in spectrum, so you have to be very careful when comparing it to the generally much flatter noise floors of electronics as numbers can be misleading.
@@PileOfEmptyTapes the max spl is typically the number for when the pad is engaged, an then typically with 20db of pad. You then have to deduct the noise and or resistance of the membrande. Some microphones thus actually have postings for their actual dynamic range, so we can tell the dynamic range they can record, and you will find that number typically is below 100dB.
I am also SO GRATEFUL when I have an audio question and you have a video for it. I know I’m gonna get a real answer to my question and leave feeling a little smarter. Keep up the amazing work!
Hi Julian, i guess the more polemic topic in whole music gear world is if you really **need** a dedicated preamp or not, how different they are from the ones that comes in sound interfaces. Thanks for all the knowledge
I feel like the most useful place for 32bit float is in recorders and run and gun situations. One man crews for instance. Where you might not always have the ability to control levels in real time and have to set-and-forget. Or in situations where there's big differences in levels (think of meeting or whatever with speakers at all sorts of different levels) where you want to normalize the levels.
The biggest advantage for processing audio at 32bit is that you don't need to dither whenever you bounce anything, so you don't increase the noise floor every time you print anything.
Dithering noise is very low. Nothing to worry about. In music it's actually cool to have a bit of intentional noise.
@@aceyage teaching sound design I stay with 24 bit, partly because it's more useful to know about setting optimal levels etc. than abandoning control.. same way I don't properly know in new Polo GTI how the headlights work because they take care of themselves. I'm still not sure in relation to 32 bit, if decisions about levels are relinquished, how it affects the long-held understanding of ideal levels at point of recording for mixing. Similarly, with noise, low but existent noise seems ideal, and 32-bit floating point seems to upset acquiring understanding of relevant ratios.
Awesome video, learned lots! So if one records in 32 bit float of talent that has a really deep and bassy voice, would 32 bit float be able to eq that said voice much easier than 24 bit? Or would 24 bit stand the same chance as 32 bit float. Had an interview recently where talent was a very deep and bassy voice, lav on tshirt and recorded in 24 bit. I'm seeing that I can clean it up ok but not great in post. I'm wondering if I had a 32 bit float recorded if I would have faired better? Or is it just a different mic technique?
Thank you for this video 🙏🏽 for a non-audio expert like me, this actually confirms the need for 32bit f considering that in a run-&-gun situation i don’t have to be so meticulous about the audio recording and still have flexibility in post to make audio sound great even though i might have not set gain levels properly or placed the mic too far from my talent. In a perfect world I’d have an audio guy on-set making sure that audio is being captured properly, but in real life and with impatient clients 32bit F comes in handy 😎🙌🏽
It’s worth remembering that 24-bit integer precision gives you over 16 million (2²⁴=16,777,216) possible amplitude levels for every sample. That’s a heap of resolution, and as others have commented, you’ll be limited in practice by other things, such as the noise floor of the source and the analog front-end. Setting input levels is fairly uncritical when you have 24 bits of precision to play with: you can be pretty conservative (leaving plenty of headroom to avoid clipping) with negligible effect on the digital noise floor. One should be more worried about the analog noise going into the converters.
As you say, using 32-bit float makes it wonderfully easy to process a signal. There’s an exact mapping from every signed 24-bit integer value to its corresponding 32-bit float-it’s a lossless conversion-and once you have the float, you can scale it up and down wildly without clipping, so it’s great for DSP (though you can lose precision).
So, floating-point is great to have “in the box”, but hardly necessary for capture.
Nobody would record with an 8 bit ADC unless they're looking to get some of that sweet sweet quantization distortion. If you're recording a person who whispers and also screams - where the scream is occupying digital full scale (24 bits) and the whisper is ≈ 90dB lower... The whisper is effectively using around 9 bits. A 32 bit recorder would encode that whisper using ≈ 17 bits. These kind of finger wagging summations of dynamic range also fail to consider factors like "headroom" which basically assumes that any recording of a sound is both unpredictable and fraught with technical constraints.
So yeah, it's technically true that 24 bits can record from the threshold of pain down to the threshold of hearing, but the lower part of that range will be sitting at or very near a noise floor. Ultimately 32 bit is another tool in the toolbox that when cleverly deployed gives the recordist a bit more leeway to capture something with the highest possible quality
perfectly summed up thx!
i've seen a few videos on it here on yt and some guys somehow seemingly misinterprete 32bitfloat.
the convenience, as you're calling it, is simply a benefit that's worth something, at least for me.
it's probably not a revolution for recording in general, but at least i won't need to level and set up a limiter, plus i can maintain the otherwise clipped signal, it's simply one less thing to keep in mind.
i guess there are also still a few recording enthusiasts around, who overestimate the importance for the high art of levelling an audiosignal a little bit, over what they actually record...
anyway, greets from outer space!
I always wondered about it since the 32-bit float format has exactly the same precision as fixed point 24-bit PCM: 1 bit of sign and 23 bits of mantissa. Those extra 8 bits for the exponent can't do much lifting, even in a multi-stage recording setup. They're just handy to have as an intermediate during processing.
One thing to note as well: You can't actually get a full 24 bits of info from recordings.
At about -123 dB, cosmic background radiation and thermal noise come into play, and with current circuity technology, there is no way to avoid it.
So really, the maximum amount of dynamic range you can record is only 21 bits or so. That said, you can synthetically create a wave in a file that exceeds this, but the moment it enters the analog domain, this limit comes into play.
24 bit is plenty. That said, the convenience of using two ADCs to avoid clipping can be pretty nice, but that's about it.
Hi Julian, Thanks for your very instructive videos. The CD industry still uses 16bit/44.1kHz. I produced several CDs in classical music, and the tradition was to record also with 16bit/44.1kHz, by fear of truncation. What do you think/advise about ? Should one continue to record 16b/44.1kHz as CD requires, or increase the rates ? Greetings
On my Daw I record 24bit 48.1 Hz but when I record chamber Orchestra and small acoustic outfits live. I use a Roland VS 100 that records 32bit FP 96k to SDHC format via 2 microphones Recording in stereo. I never clip But I still have to go through with spectral layers or RX 10 and remove Audience cell phones Ringers etc. Thank you for the video with explanation
A very good and voodoo-free explanation as usual. What could have been stressed out in this video though is that actually also integer based 24 bit PCM is some kind of "convenience" to some extend when it comes to pure recording and playback / distribution as here, the SNR of linear 16 bit PCM is already extremely high and virtually never utilized. Further processing approaches excluded of course.
It would be nice to see another "unagitated" video about the floating point format itself and the - as far as I understand it - inherent decoupling of signal to noise ratio and dynamic range which usually go hand in hand with linearily used integer variants.
Thanks Julian, long live and prosper. 🖖
Excellent presentation; very helpful to me,and wonderful English!
It's probably the best I have heard for some time from a native speaker of Deutsch!
Lass es Dir gut gehen!
Very good video, Julian!
Yes, I agree, 32-bit float is not a magic, but I don't agree it is not a "game changer". As sound designer and reporter working in the field, I love the freedom which I achieve with Zoom F3 or Tascam X8 recorders. I don'r care if levels temporarily peak above 0dB - I still have a good recording and I'm able to restore the proper levels in post-production. In the field you cannot fully prepare for changes in the environment - some of changes are too quickly to re-set the gain to "safe" value, and as you mentioned in the material, setting this level too low in 24-bit (to be safe "in the case") decreases the dynamic range. From this point of view, 32-bit IS very convenient and makes the life easier :)
Why everyone seems to tie 32bit float to the technique that combines multiple adcs signal together?
All you have experienced can be done in 24bit format. It's the combining two adcs' signal that makes the difference.
@@johnyang799 Of course. But multiple adcs are available only in 32-bit recorders. And some of these recorders has ONLY 32-bit format available. That is why :) On the other hand, if you have two or more adcs, you can process more than 144dB (just in case) and for that you have to use 32-bit. Do you need more than 144dB? Not really, but some mics have larger dynamic range, so you are able to catch whole the signal from them (e.g. Shure Beta 52A can handle 174dB SPL)
@@GadesChannel @GadesChannel Yea. I'm not saying the actual products are not good nor the fact you shouldn't use it. It's just that either good or bad marketing people tie the two together tightly.
btw there's also 32bit fixed point which is different from 32bit float.
@@GadesChannel well also being able to handel 174db spl is not remotely the same as 174 dynamic range. And over 140dbspl in real world is extremely unrealistic.
Beyond that you can combine two 24bit tracks to create more than 144db dynamic range. That's also part of the reason daw runs much higher precision underneath.
@@johnyang799 Yes, there is 32-bit fixed, I'm aware of that, but for now I never used any recorder using this format :)
And again, in my opinion, using multiple adcs without 32-bit float is not practical. Together they are very convenient even if you in theory can "fit" combined audio from multiple adcs in 24-bit dynamic range.
BTW. Do you know ANY recorder/interface which has multiple adcs per channel/track and has only 24-bit? I don't.
I'll pass and wait for the 64-bit double recorders. I know you are going to say nobody will ever need it. Neither were 1080p, 4K, and 8K video formats needed, and yet those are needed now. 🙃
Hi Julian, Thank you for your great videos. Between the solid-state logic SSL 12 and the Motu Ultralite mk5, which one should I buy if my primary use case is recording with the Shure Sm7B? Can you suggest the best audio interfaces under $700 for the SM7B?
for live concerts and events having 32bit and device that actually can capture like a dual amp would benefit so much
audio restoration for example from vinyl records, reel to reel tapes (multitrack/stereo masters), shellacs, etc. special cases, of course..
will you ever review the SSL12? Curious how it compares to an SSL2 which I own
Its the same converters as the smaller interfaces with more i/o so if you like the SSL2 or 2 plus you would like the SSL12. Cheers!
@@joesalyers That's not quite right. The SSL12 is only similar in appearance to the smaller models. It has a different converter, can do 32bit, has a higher dynamic range, etc. I'd be curious to see how it performs in a test.
For me the main selling point of 32-bit float comes when combined with DSP's, there are a number of mics and interfaces that within the DSP are able to process the audio in 32-bit float. If I have my mic setup such that it has a good output level close to where I wish to record to, if I record in 24 bit float, and suddenly get much louder I'd risk clipping. However within the compressor of the DSP it has access to more information and when it detects levels that would normally clip it can automatically lower the gain digitally without having lost data. In this way I can record into a 24-bit float format on my computer, and also not worry about clipping because prior to reaching my computer the processing on the DSP has already made sure the signal is within a good range. This is why a lot of mics claim to not clip, because they have built-in processing that takes advantage of 32-bit float to keep the levels in a safe range. As was stated in the video the main benefit of 32-bit float comes in audio processing, but audio processing isn't just post-processing, there's also pre-processing, and many mics and interfaces that advertise anti-clipping do preprocessing with their 32-bit float before it even reaches the device.
bro, imagine a camera operator with a lot of stuffs to control, snario, direction photography, actors/models, and still have to think about the audio, I dont want to check if the file on the set if going well, I just want get the audio and make it works on the pos production, so, it's a game changer for run & gunprofessionals for sure
32 bit float in field recording makes a lot of sense when you don't know how loud things are going to get. If you're recording audio for a movie or game, and you're going to blow things up, you're not going to get many chances to set the levels.
That sounds like a difference between breaking the gear and finishing your work day. 😆
All good.
I use Logic Pro, which uses the 32-bit float domain for built-in samples and its built-in plug-ins, and a lot of my tracks are sample-based, so I think it actually may also have an advantage over 24-bit.
But see if you can set me straight on this:
While the dynamic range of 24-bit is fine if used properly for a single recorded track, what if you have 100 tracks, and you set a lot of headroom, just to be careful to not exceed 0 DBFS? Does that not raise the cumulative noise floor? It's the accumulation of 100 noise floors. That means the effective aggregate noise floor is a lot higher than -144 dB.
So is that significant, or not? Would this give 32-bit float an advantage over 24-bit in mixing/mastering?
Also, the digital noise floor for 24-bit is made of quantization noise, which if amplified, sounds like a buzzsaw. The analog noise floor is essentially hiss from thermal noise, which is far less annoying to human ears.
Logic also uses something called '64-bit summing', which I can't seem to find much info on. I'd love to see a video from you on that.
Thanks Julian. I think the convenience is good for me as an amateur producer honing my tracking skills. Using Cubase in 32float recently was fun. I felt/held the full signal and used the 'clip gain' feature to easily eliminate the clipped sound. Previously, I would have had to re- record. I agree its not a need but a huge WANT.
Hi, Julian, I`ve seen a few of your videos about interfaces, can you pls advise me, do I need to change NI KA6 mk1 interface? Are modern interfaces of the same level better in preamp and AD/DA quality?
Pls rewie Yamaha AG03/06 mkII.
"If you record shty audio it will sound shty in 32 bit float as well" hah. Well put. Great video!
I think 32 bit is the great Equalizer. It makes Audio less intimidating for new entrants in the Audio space and gives them the opportunity to save time going through the learning curve to be able to get the same results with 24bit as some audio guy who had to learn 3 years to get it right.
With 32 bit, they can focus on others aspects of their productions more efficiently as well
I think 32 bit float is extremely useful for a personal recorder like my Zoom F2, when you need to set up a lapel mic on someone and you have no idea what the sound levels will be.
Agreed. I have 32bit float on my Tentacle Track E bodypack recorder, and also on my Sound Devices MixPre 6II recorder. While I always do a sound check... it's nice to know that I am somewhat protected from clipping if the subject gets too loud. I often hook my MixPre recorder to a sound board backstage at an event, while I'm in the audience shooting video. Therefore I'm not able to monitor the audio. But the files I record can be adjusted later in post if things are too loud or too soft. It's a nice safety-net for us one-man-band shooters. 😎
That's what mic checks are for.
I work almost exclusively in 24-bit.
The only time this changes is if I'm delivering multitracks or stems to someone; I prefer to render
to 32-bit float so I don't have to dither. I always assume the other party _will_ dither, so I don't want
to add noise twice.
This may not even be sound logic, but I get peace of mind and I've never had any complaints.
Please correct me if I'm wrong. I see a difference between bits required to represent the full dynamic range of audio and bits required to capture the dynamic range of audio with our devices. While 24 bits can represent all the dynamic range our ears can tolerate, in order to record it we have to capture audio above the noise floor of our signal chain. This is why instead of recording low level input with lots of headroom, we try to record at levels closer to the top and risk clipping. Ideally we could record with an average signal of -12 dBFS and never go over 0, but in practice this can be risky. We can record at a lower input level, but then raising the signal in post production we will also be raising the noise floor of our signal chain. A benefit to recording in 32 bits is setting your recording level so your average signal is higher above the noise floor. The extra headroom of 32 bit float may save the day when there would otherwise be clipping.
Can you do one video on a new apollo interface pls, just cs is missing on the comparisons.
Btw gr8 job man on explanations
First, you're the best!! I understand that one can get the advantages of 32bit Float, if at all, regardless of type of analogue signal including Mic, Instrument, or Line level as long as the signal has not been through A/D conversion in the signal path when it reaches the 32bit Float interface . Is this correct?
Incredibly clear video that answered all my questions and some questions that I didn't even have!
I just ordered a zoom r4. Im super excited
And thus begins a new generation of hardware pushed and advertised to end users by the manufacturers, advertised as the new ‘industry standard’ and ‘must have’. All companies will follow suit for fear of being left behind. They will seek big named artists and engineers to help sell their stuff. And before long, the art of gainstaging is all but a distant memory. Why do we fall for this time and time again. Just because something is technically an improvement on ‘paper’, doesn’t mean ANYTHING for the majority of amateur users who actually need to spend their time and energy on their craft - not new gear.
The most important point in this video ‘shit in, shit out’. Don’t fall for the hype kids. Learn your trade as a musician, recording engineer, mix engineer… learn a bit of physics theory behind the electronics… even if you stay firmly in your DAW, understanding WHY your vintage plugins work the way they do will help you apply solid PRINCIPLES to all your future work. Remember, we’re just shoving electrons around until eventually we shove some air around at the end of it!
Review on the new Scarlet 2i2 ? 4th gen ?
Yes please!! 🎉
Hi, many thanks for this Video! I am new to Audio recording. I have only an old Rode Videomic pro and am intrested to buy a 32bit float audio recorder from Tascam (DR 10L -PRO) I always read they are used with Lav-Mic's.. But in theory i could attach the shotgun mic with it's preamp set to 20dB, to the Tascam DR 10L-PRO to record and bring it down in post? Or would the preamp from Rode Mic fry the preamp from Tascam? While the 32Bit Container/Format could handle it, the electronics may be the bottleneck here?
I would really love if you test groove box audio interfaces such as MAschineMK3/+ , Ableton Push 3, Akai MPC one/live, Force. etc..
Developer here: Bitdepth just influences the noisefloor's depth. You don't hear a difference between 16 and 24bit and definitely none of the higher ones as well. but a lot of your plugins might internally convert 24bit inputs to higher ones to prevent mathematical errors on extremely small values, which matters for example in filter design or extremely slow LFOs, so when your project is already at a higher bitdepth those plugins don't have to convert anymore. it's not that this conversion is super cpu-inefficient though. i suppose almost no difference. so whatever you decide, it's alright.
Been following your videos for a long time and i just want to say you're the best.
Also i wanted to know of the possibility of doing a video of interfaces with great RTL times
Thank you for explaining this. I am still curious about whether 32 bit float creates more safety around sudden spikes in db, like a sudden loud laugh. That type of sound makes me wonder whether 32 bit would be helpful.
Whats your take on THE AI impact on Music Production?
I feel like the best use case scenario for 32 Bit Float is field recordings in compromised situations. Run and gun or novice setups. Using it for mobile video production is kind of a game changer.
There is one difference mentioned here: dual amp means than at some point the amp can switch frol one to the other. this is audible with ultrasonics content where you can have sudden noise burst in ultrasonics content, which is why I hear people recording with ultrasonics still seems to prefer 24 bits.
So thing is recording with dual amp do have some unexpected consequence.
Let´s say its the other way arround more important so with 32Bit, you can increase very silent mixes without yielding with unpleasent artifacts. If you do that with 16Bit, you destroy the mix, if you do with 24Bit, it can be awkward, if you do with 32Bit, most of the sound textures stay clean. So yes, then you can send the mastering engineer, he can down dither to 24bit or 16Bit then at his mastering. By the way the precision is much much higher and I see no point, why people question it, as it is just a much more sharper knife in your kitchen. Of course yes I love it and I take it. Thanks and best Regards
thanks.. great video.. so:
do I need a 32 bit interface: no
do I need 32 bit wav storage in my DAW :well no, but if you can you should enable it because it can save you in some circumstances, and you dont need to worry about recording too hot in your DAW. Got it.
Awesome! Now I can just send people this instead of going over this again and again... and again! Life much changed for the better!😆
Hello Mr Krause, Your helpful addvice helped with me getting a MOTU M4 which i love to bits ❤.
It was to replace a Behringer UMC204, (becuase no new drivers, it wouldn't talk to my PC) the Behringer UMC204 was powered through the USB i took a chance and plugged it into a phone charger and it looks like it works well in all aspects other than interfacing with computers. For example can run a SM57 mic through it, out the signal trough a guitar pedal on the insert, and hear the result through the headphone socket. Not sure if this wold work for most audio interfaces (scared to try with my MOTU M4). Please say if it is. Is very useful for super cheep preamp with extra features.
When everything was 16 bit, gain settings were crucial because of the dynamic range limit. With 24 bit it was widened. The remaining problem with 24 bit was that if you recorded a low signal, you may have to boost the gain digitally in post (say, if it's a single source like a voice). Boosting or normalizing brings with it the inherent noise in the signal. So it's not that input gain level doesn't matter in 24 bit. You still want to capture a signal as strong as you can get while maintaining plenty of headroom for clipping. With 32 bit, this problem seems to be eliminated.
So, if you have a 32 bit device, record a very low signal, than boost it in post, the noise won't be boosting too? I don't see how is that possible)
@@diktor_ilyakhaiko
As far I know many if not most 32 bit devices have fixed gain and input sensitivity anyway.
What one is changing with the knob, especially in ones that have dual DACs (this 32bit float), is simply the structure of the file.
And how can it record low signal without noise? Well the more bits You have the more of them are usable aka not post to noise. And simply the amount of possible values is waaaay greater.
24 bit is 16 777 216 possibilities
32 bit is 4 294 967 296 possibilities
You can fit entire 24 bit range 256 times inside of 32 bit range. Just think about it, even if You have 100 gain/volume settings on 24bit ADC that still less than how many times 32 bit can overshadow it.
For the end product you convert the audio file to 44.1 kHz 16bit 256 kbps AAC or OPUS (← 48 bit).
The very little quality loss is worth the tremendous amount of saved space.
But it's always convenient to keep lossless high quality audio as a source file for editing purposes.
32 bit float is a genuinely helpful safety net so to speak. I only wish they started making more 32bit float interfaces besides the 2input zoom one.
I'm wondering based on your older video on preamp gain effecting the level of the noise floor in a recording that if you intentionally place the gain on a 32bit recording device (i have the Zoom F6) between the two preamps and lower it in post, do you get less preamp noise? Is it just two halfs equal 1 situation?
I have recorded album with 32bit float on F6 in 2022 drums, piano, bass, guitars, vocals. And this year 2023 i did another one with my old Zoom R16 in 24bit but i think the most difference there is are the more noisy preamp's Zoom R16 has. With F6 i can boost the top end over 10k high's of a piano without bringing up a bright noise like with the R16 preamps. On this years record i went more mellow, natural sounding overal feel and did not boost those. Overdubbing with F6 is time consuming process couse you have to record with two recorders, one to capture the sync + F6 to get the sound of the preamps and sync those tracks in post but the result sounds really good and there was some depth i could hear in the sound
One thing I do not understand. 32 bit audio interface (steinberg ur22c for example) will use those 32 bits for recording only? If I won't record anything and just want to make music in Ableton in 32 bits - do I need a 32 bit audio interface for listening to what I do? Or I can have 24 bit interface and still produce in 32bit (i.e. interface bits and ableton bits are not related)?
It sounds like a good idea when recording audio along with video because it's hard to take care of audio levels when you are trying to record video as well so it's easy for distortion to happen that can't be fixed later. With 32 bit float you can fix the problem later and reduce the signal to below the distortion.
Would it be safe to say 16 bit and 24 bit is fine if you don't clip? Love your vids!
Yes, and 24 bit is much harder to clip than 16 bit, so 24 bit would be a preferred recording format. 32 Bit float only has an advantage if you record something with very large volume fluctuations, make it hard to set the gain. If a volume is consistent, like with recording a person speaking during an interview, you can set the right gain with 16 or 24 bit, and there will be no advantage using 32 bit. You would be hard-pressed to find a 16 bit recorder these days, almost all have 24 bit at minimum.
For field recordings 32bit float has all advantages as 24bit and when you blow a recording u just pull it down to 0 or less
Or if the signal is very weak u can pull it up50-60 db without adding noise
I’ll never buy a 24 bit recorder again! The easiness knowing that u do not blow out recordings when using lots of money and energy getting places like the arctic or Antarctica is good enough. Our zoom f4, h5 and H6 are all in the shelf!! F3, F6 F2! and mixpre3 ll are the tools and they work all well with em272 and Sennheiser em66 and 62
Load up, normalize snd start editing….
The issue I've been trying to understand, but haven't really seen discussed is how, if it all, 32 bit float is applicable in a live format like streaming. For example, I have the Tascam Portacapture X8, but haven't used or explored it much yet, but if I was using it as an audio interface to my computer, and then through OSB or something similar to live stream, would there be any advantage to 32 bit float at all?
No real benefits with 32 Bit float there. Most streams dont use 32bit float. The audio Stream has to be encoded into a compressed format such as aac. Even if it would be possible to stream in 32 bit float the people would still have to convert it back into an analog signal to send it to speakers or headphones. Most of which dont have the dynamic range of even 16 bit.
Also from a listening point of view most people dont seem to like it if you yell at them with 140dB SPL right after you have whispered into there ears 😅.
The only benefit could be if you have a 32 bit digital signal processor(hard or software) before the stream in which you could apply effects such as compressor and limiter: you dont have to worry as much about the settings and internal levels exceeding 0dBFS as long as the output doesnt.
Live streaming tends to require a tightly managed output dynamic range (since your bottleneck tends to be viewers' playback equipment), and the kind of dynamic range you can actually capture on the input side tends to be small enough for 24 bits either way. The biggest difference between int24 and float32 would be the input levels that the compressor sees.
With 32 bit float, say the audio was recorded, quite low, so -20 DB and you wanted to bring it up to 0DB is there. Also a noise floor that is brought up when you increase the audio volume as well?
which is better for recording loud motorcycle exhaust?
So, 32bit fixed point audio interfaces vs 32bit floating point interfaces? Is that a thing? The hardware has to be 32bit floating point for the recorded file in the DAW to have this dynamic range...is that correct @JulianKrause ?
For me, 32-bit float recording is very helpful because I record live events where the audio level can drastically change at any moment. One second we may need to pick up someone whispering, and the next there could be thunderous applause. I don't have the luxury of changing the record volume on a moment's notice. I think this is the kind of situation where it comes in extremely handy.
Definitely a good example of a situation in which I would think 32 bit FR would be a blessing to have. As well as ambient environments like city streets in which many varieties of sound are coming at the mic from varying distances and decibels.
Hey Julian.
So i have a wish in case others find it interesting too and most of all if you feel like makinga video about it.
It is about D/A converters. I frequently hear people for years now, argue how their 1000+ dollars interfaces (RME, Universal you name it) sound „not good enough“ and
that a much more expensive „super“ converter like a Lynx or Crane song etc will make a „huge" difference. I somehow do not believe that in converters of our time in history there can even be a huge differences but i cant back it up on a technical level as you would be able to do.
So i would be very interested in a comparisson of a high class interface to a high end „professional“ grade converter with actual measurements
of „What goes in, What comes out and What the actual difference is between them beyond Esoteric interpretations“
Float means "floating (moving) decimal point". Thinking decimal: imagine a 7 digit number with floating decimal point - we can write numbers from 0.000001 to 1,000,000. Traditionally we use integer numbers for 16 and 24 bit depths and, once again thinking decimal and using a 7 digit number we can go from 0000001 to 1,000,000 - that's because integer numbers don't allow for fractions.
Let's compare how much larger the second number is compared to the first one in both cases:
7 digit decimal float: 1,000,000 divided by 0.000001 equals 1,000,000,000,000
7 digit decimal integer: 1,000,000 divided by 0000001 equals 1,000,000
Comparing how wider is the float range compared to the integer range (same number of decimal digits¹):
1,000,000 divided by 1,000,000,000,000 equals 1,000,000 - using float allows for a range (from smallest or quietest to largest or loudest) a million times greater.
1) I used decimal digits instead of binary digits (a.k.a. bits) because it's easier to understand. Remember that floating point is a bit more complicated than what I did but I think keeping it simple helps to understand why floating point gives a larger dynamic range than using integer. I did not talk about dBs because I know very little about it.
Thanks, Julian! I finally understood how 32bit float works and how it can be implemented using less than 2^16 16bit ADCs!
I’m gonna give you two use cases where 32 bit float reigns supreme on location - shooting ranges and loud concerts front row trying to get that shot. Not having to worry about levels on the fly is the best. Buying a larger SF card is the smallest price to pay.
Wenn es um Audiotechnik geht, dann komme ich manchmal gar nicht drum rum, ein Video von dir zu nehmen.
Mach weiter so!
Question no one has answered for me yet. On windows in the bottom right sound properties next to clock there is a tab called recording. If I right click my mixer in there hit properties there is a recording level setting. Should I ever touch this?
Reason: If I set my gain on mixer just right where the clip light rarely comes on and keep my mixer volume at +-0 but not adjust system recording level. Then I end up with extremely loud clipping audio. So I have nice clean audio going into computer (monitoring with headphone) but unless I lower one of the following I have issues. Gain/Mixer volume slider/system recording level. Which is better to lower or am I doing something completely wrong?
Thank you for your report, this brings to a more understandable and clear concept of the discussion or audio marketing politics. Danke, Julian.
As an Adobe Audition user your aware that its multitrack sessions default to 32 bit. I have never questioned that default but now I wonder if I even need it? Most of my mixes are very long - around an hour - and they can take several minutes to mixdown. If 32 bit is doing nothing for me in the mix yet is slowing down my mixdowns, would it make sense to mix in a lower bit rate?
You can render 20% faster in 24bit
Extremely informative. Great video!
Hallo Julian. Warum nimmst Du Dir nicht mal eines der UA Apollo interfaces vor? Wäre sehr dankbar 🙏
should i use mono for yt audio record or stereo
Hello Julian! I'm currently trying to use my Zoom F3 as 32 bit audio interface, but audition is still clipping my recording on 0db. I downloaded specific driver from zoom website, updated firmware on the recorder which is adding 32 bit AI mode, chose ASIO in audition, but it didn't help.
I often increase or decrease the speed of videos that I watch, and this introduces various forms of audible signal processing anomalies in the playback. Would using 32 Bit Float improve or affect the playback quality in such a scenario?
It wouldn't affect in that sense. The artifacts you hear are most likely caused by the algorithm used to keep the original pitch of the sounds. If you just increase the speed by brute-force (playing every sample at twice the speed), everything would sound high pitched (Nightcore music takes advantage of this fact). But you don't want chipmunk voices in Julian Krause's new video when watchign at 2x, so an algorithm is introduced that keeps the pitch and the pronuncioation of the letters, but increases the speed. This is not perfect and introduces a lot of artifacts (and it's usually optimized to mostly keep human voices and words, so other sounds might get more distorted)
No unless you have clipping in the recording in 24bit. But you can clip using 32bit format too. Hardware is limited.