Why You Don't Need to Record in 32 Bit Float....

[@unityoc]

This reminds me of worrying trends in other areas like photography - fixing things in post via saturation boost, depth of field manipulation, sky replacement etc. is rampant but putting the cart in front of the horse in my opinion! Thanks for this great video Jon.

[@R3fuge]

A correction: 32 bit float and 32 bit fixed have the same number of possible values, but float can represent a larger (and sparser) range of values. There's only ever 2^32 possibilities if you have 32 bits, so you can never represent more distinct values than that with 32 bit.

[@benjamindover4337]

90% of youtubers are unable to handle the responsibility of avoiding audio clipping.

[@FilmmakerIQ]

I feel like for some KZbinrs clipping the mic is a stylistic choice

[@sizzlereel3523]

Great breakdown. People have a hard time differentiating between what's technically possible and practical. Thanks.

[@petercofrancesco9812]

I find 32bit is helpful for theater video because often I can't get a sound check or a good one. So let's say they open the house early or the sound tech guy isn't there, busy, or has a high school student running the board. You say set it at -20db but if you have no audio to base what the peaks will be how are you going to do it? There's classical music like Tchaikovsky Nutcracker that has huge changes in volume. If you're unfamiliar with the music you'll have no idea what is the loudest section. Same for boundary floor mics how loud will a tap performance be, performers are generally not going to come out to do a sound check, even when they do sound checks for musicals often the performs are talking not singing full blast. The point is 32bit takes the worry out unmanned recordings. I can setup my boundary stage mics when I first arrive hit record, then move on to the rest of my setup. I can't tell you how many times I've done a sound check only to find out after the performance that the levels were significantly higher then when I tested and the signal got clipped.

[@FilmmakerIQ]

_I can't tell you how many times I've done a sound check only to find out after the performance that the levels were significantly higher then when I tested and the signal got clipped._ That's because you set your levels too high. You didn't follow the -20db headroom rule. Couple things I need to clear up for you. Giving yourself -20db headroom means you've been 100x MORE signal to go beyond. That's not just 100x more signal than the average... that's 100x more signal than the LOUDEST thing you expect. So a -20db headroom might have have 40db or more headroom for average levels... that's 10,000 times acoustic power. Plenty to work with. You mention boundary floor mics. That's pretty straight forward to set - with a -20db headroom there's virtually no way to clip naturally (unless they're kicking the mics - in that case, who cares if they clip at that point - even 32bit fp won't save you if you overload the mic circuitry). If you can't sound check with performers, just do your own sound check... walk in from and start clapping or making noise. Then dial yourself down and go a little extra conservatively. Now taking signal from a board... that's a bit different. Much easier to overload with electronics - but again the solution is simple - play conservatively. You HAVE to check the signal at SOME POINT (how would you even know it was outputting signal at all?) - aim low so that you're playing in the -30 to -20db range. It will literally take 3 seconds - you can even do this at the start of the show. You're used to see the bars fill up meter - you have to train yourself out of that - the bars should never go higher than -20db (most recorders put that about half way up the scale). 32 bit FP is fine if you have it. It won't be a detriment to use it... but you could do ALL of it with zero downside, if you record 24bit lower than what you think you have to.

[@trav3ls]

As always, absolutely fantastic. Sounds fine. There is also a point that no microphone, no analog-to-digital-converter is able to capture dynamic range corresponding with bit depth higher than 24. Also as a computer scientist, I would like to point out, that 32-bit float actually can represent same number of distinct bit combinations as 32-bit fixed point. Or actually less distinct numbers, because there are multiple ways of representing i.e. 0 (with mantissa of 0, your exponen and sign bits don't carry information about different real numbers). There is wider range of numbers, but 32-bit fixed number can represent ALL the mathematically possible integers within its range, 32-bit float can't. You have only so many bits in your mantissa after all - increasing your exponent does not increase accuracy of your mantissa. Therefore: 32 bit float has actually FEWER possible values. (only: wider range of those values).

[@FilmmakerIQ]

You are correct about 32 bit float! In regards to how no ADC is capable of the range of 24 bit... That's why they use TWO ADCs for 32 bit float :D

[@foljs5858]

"A problem I don't have and you don't have either if you follow these basic techniques". With 32bit float we won't have the problem even if we don't follow these techniques. That's the advantage. One less technical issue to worry about while being creative.

[@FilmmakerIQ]

That's not how it works. That's like saying you don't need to look through the viewfinder or monitor... One less technical thing to worry about... Being creative is about mastering the technical. It isn't one less thing to worry about because I don't worry about it.

[@foljs5858]

@@FilmmakerIQ "That's like saying you don't need to look through the viewfinder or monitor" Nope, because the latter is a creative action (you determinine the framing, lighting, etc). Whereas fiddling to find a suitable audio level because the recorder will clip is just a technical chore because of hardware limitation. Not something creative. By not looking through the viewfinder or monitor means you lose the choice of framing. Whereas by not fiddling with input levels and using 32bit you don't lose anything. "Being creative is about mastering the technical." No, being creative requires mastering the essential technical. The rest is what is called in engineering "accidental complexity" (useless work) as opposed to essential complexity that is required for the task. When something is not required anymore (because technology can do it as good) it stops being essential complexity, and becomes busywork. Note that this is not the same as setting e.g. focus manually, because that's an artistic choice (and you can't undo bad or unwanted autofocus results). Whereas not fiddling with audio levels and using 32bit you don't lose any artistic choice - just the busy work. So, given the availability of 32 FP setting manual levels is more like having to pull a lever to be able to take a new picture - something that made sense in the analog film camera where it rolled the film to the next position, but is useless now.

[@FilmmakerIQ]

Viewfinder isn't just a creative action. You're also checking exposure beyond just creatively adjusting the lighting. Monitoring audio also isn't JUST a technical chore - you're also listening for mic placement and making sure everything is sounding like it should. And in the process, you just set the levels. It takes 10 seconds. Quite literally the easiest thing you do when recording audio. If you consider essential diligence in the job of capturing video and audio "busy work" - then you really don't have enough mastery to be creative. Then I get it - 32 bit FP is for people who don't know what they're doing.

[@dolphinuppercut]

@@FilmmakerIQ I totally hear what you're saying, we would like everyone to be considerate of the technical components so that they can use those techniques in pursuit of quality or creativity, not only moving their own art forward, but pushing the limits of mastery within the community. The main argument of the video is also strong, that no one needs to feel pressured to spend this money if they only need to make small adjustments to their technique. I think we can all agree that we shouldn't be gatekeeping "creativity" with "mastery of technical knowledge." Obviously we don't need to be masters of any skillset to be creative, and the act of creativity itself lends itself the the growth of diverse skillsets. I know you're not trying to gatekeep, just reminding you to be sensitive to that as a leader within our artistic community. Mastery is subjective anyway. Another thing we can all agree on, 32-bit floating point is a tiny level of abstraction in the progression of audio technology and sound art. We can consider it to be one more tool to make sound art more ACCESSIBLE. Technically speaking, no one NEEDS things like non-linear editting (NLE) or DAWs to be creative, everyone is free to go back to reel-to-reel if that's fun for them. As a matter of fact, no one NEEDS recorded media at all. We can perform acoustically, and that's fantastic. Conversely, we agree, no one NEEDS to know how to perform acoustically, or how to do reel-to-reel to excel with NLE and DAWs. Would it be helpful to know those skills? Absolutely. Necessary for creativity? Obviously not. Just to quickly point out, small optimizations can make a considerable impact in cost, resources, and/or time in the big picture. 32-bit floating point can be considered one such small optimization. On a personal level, the more optimized a process becomes, the more time we have within our lifetimes to create ART. We have barely a concept of the future of technology, for all we know, some new technology may come out in the future that practically eliminates gain knobs, noise floors. When new tech is cheap and plentiful, all of our tutorials, all of our textbooks, they'll all be antiquated.

[@dolphinuppercut]

And also quick reminder for everyone, we DO in fact have existing technologies making camera monitoring (viewfinders) more optimized: Motion tracking (Ronin 4d), auto-focus, high dynamic range, 360º cameras, depth detection, and light field cameras (Lytros) are all examples of such tools that visual artists can choose to use. We can also be creative without using a viewfinder at all. One could pursue "the art of photography without using a viewfinder" as their chosen field, if that's fun for them. The statement "being creative is about mastering the technical," I don't think we really mean that. Being creative involves technique, but ultimately only relies on the act of creation.

[@austinlindsay]

Great video! Love the work you out in to your demos here!

[@pgmorrow]

I'm currently doing the dialogue edit on a travel show with audio that was apparently recorded on the cameras. No sound person was involved during the production. The limiters are getting slammed in almost every scene. It would would have been nice to have either an audio professional in the field, or technology that didn't depend on limiters.

[@FilmmakerIQ]

No technology prevents user error.

[@SliceoflifeVR]

@@FilmmakerIQ That makes absolutely no sense. 32 bit float would have prevented this users error (in this case they didnt set gain correctly during filming) The post production limiters wouldnt have even needed to be used. So yes, technology can prevent user error.

[@FilmmakerIQ]

​@@SliceoflifeVR there would just be another user error waiting to happen. Why are you so naive when it comes to technology?

[@SliceoflifeVR]

@@FilmmakerIQ The idea is to prevent the error from being recorded forever. Once it is recorded and clipped, there is no recovery from that error with 24 bit. With 32 bit, you can indeed recover from the initial error, as the information is still there. It is clearly superior to record in 32 bit float, as it prevents this error. As a "Run and gun" producer myself, 32 bit float is an invaluable asset to have in the field. In order to capture genuine interactions with people as they happen, you require the ability to hit record once, and not have to set the gain high or low on the fly depending on how loud the next random person is going to speak. Maybe one day when you are a more experienced filmmaker you will understand the importance of each of the tools technology provides us. Including the importance of 32 bit float.

[@Vlachos91]

In 24 bit when you record at lower levels, the signal over noise ratio isn't as good. This is really easy to test : Record a constant sound that doesn't change (fan, 1000hz from a phone...). Set the gain so it peaks just below 0 dBFS then stop the sound and keep recordind for a couple seconds. Then repeat the process with the gain set much lower (let's say -40dBFS). Finally, boost the second recording so the level of the recorded sound matches the first one, and then compare the "silent" portions. Results : the noise is much louder when the gain is set lower.

[@FilmmakerIQ]

40dB is pretty significant drop but even then you still have about 80dB till you get to the true bit depth noise floor. In my video experiments with the buzz saw I only went up 30dB get the whisper audible. You're suggesting baking peaks 10x softer than my whisper. But you're going to be swimming in ambient noise long before you get to quantization noise. I doubt your experiment would be able to distinguish that... Still the point I make stands, set your gain lower than you think with peaks around -20dB

[@Vlachos91]

@@FilmmakerIQ I ran the test again with 4 different recodings : Just below 0 dBFS, -10 dBFS, -20 and -30. 0 dB file : -71.1 dB RMS Noise floor -10 dB file : -70.6 dB RMS Noise floor -20 dB file : -69.9 dB RMS Noise floor -30 dB file : -67.4 dB RMS Noise floor This may not seem like a huge diffrence, but it is a very audible change. You can mosly hear the difference in the -20 and -30 dB files. Here's a link so you can listen : drive.google.com/drive/folders/1lDZluzcZgseekxGj2GYqH_NcOjXl8AK0?usp=sharing This test was done with a Presonus Quantum 2626. Results may vary depending on your audio interface/recorder. I might run the test again when I receive the Zoom F6 and compare 24 bit with 32 bit float.

[@FilmmakerIQ]

I don't see or hear any audible difference in your files - between -67.4dB and -71.1dB - less than 4dB difference of already nothingness? Plus - there's no way of verifying your precision in measurements. Those minor changes could be anything...

[@Vlachos91]

@@FilmmakerIQ There is a very clear and audible difference. If you can't hear it, this might be where your flawed opinion stems from. I could repeat the test multiple times, the results would be the same, as long as the room tone, mic position and level of the sound source doesn't change. In this test I made sure the only changing parameter was the gain setting. This added noise is pretty easy to deal with in post for spoken word but is still not ideal. Where it is the most detremental I think, is for field recording : A de-noiser will attenuate the noise induced by the recoding device, but will also reduce the ambiant noise (that you actually want to keep). This is why 32 Bit float can be useful. It is true that in most cases it doesn't offer a clear benefit. But there are times where it really does.

[@FilmmakerIQ]

Okay it took me a while get where you're going with this. But you're numbers aren't lying. 3dB down at the -67dB after you've already gain boosted it 30 dB... That's close to rounding errors. And my recommendation of 20dB leaves you calculation at 1 dB loss... Nah that's not credible. I don't hear a difference and given sound's proclivity to bias, I'm better you really could tell if I blinded you and recorded my version of the test. I think you've made more a case for how strong 24 bit already is.

[@Friedeggonheadchan]

This video definitely includes one of the best layman's explanations of what floating point numbers are, but there's one detail incorrect. A single precision (32-bit) IEEE 745 floating point value can *not* represent 3.4028235 × 10^38 possible values, but instead that is the _maximum range_ of a number. Like amount of bits in the name implies, the amount of _possible values_ is the same as 32-bit fixed point, roughly 4 billion (technically a less due to details in the standard but similar amount regardless). This is an important distinction since like the explanation... explains, the values themselves are logarithmically spaced along the number line, with the gaps between subsequent values increasing as the values themselves increase.

[@FilmmakerIQ]

Thank you! Added my error to the description

[@bricsuc]

@@FilmmakerIQ The timing of this error is unfortunate because it's so close to the beginning of the video. It distracted me for a while (a depressing side effect of having a computer science degree). The thing is that your message in the rest of the video is really important---emphasizing stuff like mic placement and testing. I mean, you wouldn't point your camera at the ground and expect it to shoot anything but the ground...

[@DaveTexas]

Man, things have REALLY changed since the days when I was using a Nagra tape deck to record location sound…

[@jonasschloegl]

Honestly... I'm not really interessted in audiorecording but this video was really interesting, and now I'm questioning my self. Thanks for this really cool 24 minutes

[@neonmammals]

Great video. You’ve convinced me. I’ve always been worried about having the internal noise floor wreck recordings that are too low so have had the gain set high to over come that, but making the mistake of never actually testing it it to see what I could get away with. Your demonstration proved that was unfounded. Great stuff as usual.

[@paulisaacs5980]

Another perspective ... Using limiters to prevent sudden extreme louds causing digital clipping has been standard practice for a long time. There has always been an understanding that provided you only use the limiters (good ones) sparingly, their side effects (distortion) will be minimal and acceptable. 32-bit float addresses the same problem that limiters are designed to address but with the added benefit that there is no side effect of distortion - surely that's not a bad thing. That's why limiters are not necessary when using 32-bit float.

[@FilmmakerIQ]

Never said it's a bad thing... However I don't see it as a huge benefit if you are already in the practice of using limiters in a sparing way. The limiters already provide the insurance against clipping which is hard to do if your average peaks are -20dBFS I think the problem is people set their audio levels far too high.

[@Muhbuu]

@@FilmmakerIQ I agree with you, that you could just lower your gain and thus have plenty of headroom for just about anything in 24-bit. However what about target loudness for audio delivery? R128 -23 LUFS or KZbins -16 LUFS, where you need the average dialogue in the audio files to be at a certain level dBFS? Further down the pipeline after the production sound recording, if the audio files are recorded at a much lower gain to secure a solid headroom while minimizing the utilization of input limiters, the editors etc. would have to either gain the audio files in the editing software or raise the calibrated listening level of their sound system, to properly hear the dialogue. The latter creating a false sense of audible loudness range compared to the average consumer playback setting. In 32-bit floating it seems you could record -23 or -16 LUFS "hot" without slamming the limiter or clipping, while still letting the editors avoid any riding of a volume fader? Maybe short of the limiter in the 16/24-bit DAC

[@FilmmakerIQ]

You normalize audio to the target loudness for delivery - it's that simple. This is regardless of what bit depth you record in. LUFs is not your concern when recording - no audio recorder even displays it. You're way way overthinking - 32bit float doesn't negate the need to normalize the audio in post - in fact, it essentially FORCES you to fix it in post because you're much more limited in setting your levels. As for 24bit with the headroom - yes, you will gain it up in post, no false sense of audible loudness needed. And the noise floor of 24 bit is so incredibly low that you can do a lot with it - you'll run into ambient noise long before you hit noise from the system itself. What 32 bit float solves is people NOT following the technique of recording with proper headroom.... which can be negated if you follow the technique.... of recording with proper headroom...

[@DanielHodotcom]

Thank you for this incredible presentation! I've been an audio engineer for decades and have been looking into portable recorders (enticed by the 32-bit float technology). You explain everything so clearly!

[@TiagrajI]

Thank you. This is so clear. I always listened to 32bit depth due to all the marketing you read on audiophile forums. 24 bit is enough

[@Sayphonik]

32 bit float is great because you dont have to worry as much about clipping a signal and allows you to do what you need in post production. This is most useful for really loud sounds, gunshots, sudden signal spikes, etc. When you have high budget shoots going on, it’s a pretty good safety net if location sound is going to be used in post production. Also would like to add that it can record extremely low noises as well so overall dynamics is just much more clear, which you can amplify in post production if needed.

[@FilmmakerIQ]

You really should watch the video... 32 bit isn't more clear than 24 bit. It's actually 24bit with overhead protection. Which is overkill anyways

[@MikhailNikonov-r7o]

​@@FilmmakerIQ , look at it this way: by setting your gain in expectation of very high possible signal, you're borrowing from your signal's resolution; you're sacrificing decibels from full dynamic range of your 24 bit recording to safeguard yourself against clipping, but at expense of dynamic range given to quieter sounds. Or, taking your test with loud and quiet sounds in a same take... Suppose you have a scream, which is, say, 110dB - and then you have a whisper which is, suppose, 20dB, you're recording in 24 bits and you're setting your gain so that your scream would be just shy of clipping. All good and well, and your 110dB scream will map onto 16777216 possible levels. But then, you record a whisper, which, by amplitude, is 90dB quieter, and the maximum amplitude of your whisper's curve will be mapped to sample with value of 16777216 / 31622 = 530. So, you have values 0 through 530 to represent your whole whisper - or, put differently, you're effectively recording your whisper as 9-bit audio (which, if you'll think of amplifying your whisper later, will give you the expected 9-bit quality and quantization noise floor). Granted, 32-bit float isn't a silver bullet, miracles don't exist, and - not even mentioning that it's harder to technically implement - it won't let you record loud and quiet sounds *simultaneously*, can't take recording of whisper overlapped by a kettle's whistling, then filter out the kettle and get your perfect whisper; won't work like that, 32-bit float is a way to represent bigger numbers, but not any more precise numbers compared to 24-bit - so, whisper on top of a huge sine wave will still be represented by the precious few trailing bits of mantissa. But if you record loud and quiet sound one after another, it pretty much lets you record them with same bit resolution, and it's a pretty neat trick in my book - not for all applications, but, say, for voiceover work, for audiobook narration when you could use a high quality whisper right after nice loud scream (without resorting to fake "quiet screaming", which never really feels natural).

[@FilmmakerIQ]

@@MikhailNikonov-r7o It's a mistake to think of sampling depth as "resolution". It doesn't work that way because audio is transformed from digital BACK into analog when you listen to it. So whether you dedicate one bit or 10 bits to describing a sine wave, the analog result is identical. Also a scream is not 110dB.

[@MikhailNikonov-r7o]

@@FilmmakerIQ , I'll agree that scream isn't probably 110dB, but that's beside the point, the numbers were taken just to illustrate the fact that by expecting very loud spikes of signal (and reserving your dynamic range for them) you inevitably sacrifice the amount of bits you use to represent quiet signals. About the resolution part, however, I'll have to disagree. ^^ When you transform audio from digital to analog, your conversion is subject to quantization error, which in essence defines your signal to quantization noise ratio (which is 20*log10(sqr(1.5)*2^b)). You may be thinking of 1-bit ADCs and audio streams, but don't let those fool you, those are entirely different things and convert sampling frequency into increased bit depth during conversion, so, speaking roughly, a PWM signal where amplitude is encoded into fill rate and not extra levels of digital signal. If 8-bit audio was able to produce signals identical to 16-bit (with phase correctness), we would've lived in an entirely different world today. ^^ Quantization noise is not something depending on conversion back to analog signal; quantization noise is created in the process of converting analog to digital representation, once you digitized the signal at given resolution, you're stuck with that SNR limit from that point on.

[@Calculon559]

32-bit float is absolutely essential for audio editing thinks to the basically infinite noise floor, but yeah, for recording purposes it's nice to have but 24-bit is enough as long as you know what you're doing.

[@PostingCringeOnMain]

I just want to help feed the KZbin algorithm with engagement but don’t really have much to say except thanks for laying out what 32bit floating point audio is with just good plain information and no marketing spiel. Keep up the good work :)

[@FilmmakerIQ]

Blessed be those who feed the KZbin Algorithm.

[@PostingCringeOnMain]

@@FilmmakerIQ your videos got me through university, I passed them on to lecturers to show in class and used screen grabs in my academic journals. Now that I’ve graduated, satiating the algorithmic overlords feels like the least I can do in return.

[@Da_Man_1221]

Man, I Love your videos because they work to move past people's instinctual understanding of how filmmaking works and then explain the reality of what's going on. The sale of "never set your levels again!" should really be, "Now you can chose when to set your levels!" The step of level setting hasn't been removed with 32bf; it's just made it more flexible. It also doesn't magically make the talents and knowledge of an audio engineer irrelevant. The properly positioned microphone planted in poinsettias pointed at the piccolo player playing perfectly will always be preferred over Fredrick farting at the far fringes of the foyer with faith that he'll the floating bits will bear a bountiful bouquet of beautiful recordings to be bequeathed to his buddies boldly on Facebook......pineapple. Good technique will carry you farther than any digital commodity.

[@sergiomartinez5946]

I agree. Marketing something as "idiot proof" only breeds more idiots.

[@johncristy6901]

Coming at this as a computer engineer, the magic is mostly the extra 8 bits. That's what gives you more discrete values that can be represented. Anything else is an optimization. That gives you the headroom John Hess demonstrates can be found at 24-bits, but without trying. Floating point vs integer, bit count equal, does change the representation in one way. Where integers stop abruptly (clipping), floating point starts losing precision (distortion). Without excess bits, floating point would require the same tuning and headroom. It's higher precision trailing off to lower precision versus integers constant precision might be better in some use cases worse in others. I see it mentioned elsewhere that this is an advantage for audio and it theoretically is (more precision where it's needed perhaps), but it's hard to say if it's a real world advantage. That said, of all the top to bottom designed non-linear encodings (say the 8-bit encodings used in telephony), I don't think they'd have chosen what IEEE chose for it's general purpose floating point.

[@noisetheorem]

I remember when 24 bit came along and everyone said we didn’t need to worry about setting levels. Now it’s 32 bit float. Guess what? We’ll still need to worry about setting levels properly. If you don’t, you’re hurting every process after recording.

[@FilmmakerIQ]

Well with 32bit float, they make it harder to set levels. On the F6 the dials stop working and you have to go into the menus to adjust things!!

[@lorenzo4708]

I'm all for this argument when talking about recording mainly dialogue, because the human voice, as you have presented with your test, has a very predictable range. Sorry, but that "I haven't clipped in over 20 years" argument does not work when you go more in-depth on sound effect recordings made with very sensitive (or multiple) microphones, and piezos. Especially taking proximity effect into consideration when trying to capture rare and spontaneous performances out of braking glass, metal hits, train pass bys, fireworks, inside of car engines, waiting for a lion decide to roar with two hands on a boom pole and many other tasks where you can't physically manage the levels or predict the peak of the source. In other words, it's not always possible to properly "budget" the dynamic range in those cases. Have all of those items been recorded in 24bit? Yes. Ask any of those engineers and I'll bet even the most experienced ones have lost some valuable and unique takes due to unexpected clipping and they would definitely say that 32bit would be worth not having to go back in time or loosing fidelity with a de-clipper. You might say this is hypothetical but that is the target audience for the Sound Devices advertisement video, IMHO. And KZbinrs that have no idea of what they're doing nor do they care about sound (in Zoom's case). Yes, you should set your levels properly, but my point is, no one is perfect and no situation is set in stone. Using a tool because of the safety net it provides does not mean you don't know what you're doing and does not excuse you from learning the proper techniques as well. Just because a hammer works perfectly fine it doesn't mean you can't use a nail gun if you feel like it as long as you get the job done. I know, bad analogy. In the end, you'll only get one chance at recording that Nuclear explosion, so why not increase your odds of getting it right? Just my two cents... but who am I anyway? Maybe I just don't know how to set my levels, or I'm butthurt because you did not validate my new purchase. lol, it's late.

[@elephantgrass631]

💯 like you said, it’s risk reduction. And if reducing risk means that you have one less thing to worry which will improve your performance of capture, then use it. Using 24 but instead of 32bit float doesn’t make you the better audio recordist, it just inflates your ego. 32 Bit files also don’t destroy your CPU, require a server room of hard drives, and crazy hardware to process. There’s no better time to let go of one more thing in the solo capture process. 32 bit has proven we don’t need to go any high in bit depth now unless some weird 4D sound capture that would exponentially benefit us greatly were to present itself. Maybe In the future? Until then, there’s no need for 48 bit float after what is being discussed here.

[@FilmmakerIQ]

Ask any experienced engineer and I GUARANTEE they have NEVER clipped due to the technical limits of 24 bit. It's not that the human voice has a predictable range, it's that 24 bit has a range that a substantial portion of what's even physically possible. But the real proof is 32 bit float feature wasn't introduced on the top of the line audio devices first and trickled down to the cheaper devices. It was introduced on the lower end first. Because if you have experience, it solves no problem. And it doesn't reduce risk at all if you actually do some basic things.

[@elephantgrass631]

@@FilmmakerIQ And if you don’t have experience you should feel bad about using 32 bit? There are a lot of beginners out there who wished they had a Time Machine that could get that burnt take back while they were practicing. The fact that you could still burn a take sucks so eliminating that possibility helps. Sure it may be easy to most for setting levels, but for those that are starting out, 32bit is a great way to hone in while you get a safety net.

[@FilmmakerIQ]

Point me where in the video I said you should feel bad. Once you learn the simple basic technics, 32 bit float offers no benefit.

[@elephantgrass631]

@@FilmmakerIQ Let me guess, if you have to rely on 32 bit, you shouldn't be doing any audio recording right? "Ask any experienced engineer and I GUARANTEE they have NEVER clipped due to the technical limits of 24 bit." Good for them. I hope they feel better about themselves and can now slap the "pro" moniker on their participation trophy. 32 bit offers a safety net. A welcome one for more people who are starting out to get into this and/or just aren't as interested in audio but know that they have to capture it, knowing the chances of getting a burnt take are pretty much nil. We've finally hit the limit when it comes to audio capture. This is a good thing. Let it happen.

[@RJasonKlein]

This was very useful information for my newbie skill level, and presented in a detailed, yet easy to understand, way. Thank you!

[@KarlRock]

Incredibly well explained! Thank you!

[@meck10101]

The problem is when you boost those dynamic changes (the quiet ones) you also boost the noise floor and the unwanted sounds so you should also consider getting a mic/recorder that can handle such quiet sounds without noise!

[@FilmmakerIQ]

24 bit's quantization noise floor is extremely low - you won't notice it at all if you follow the approach I laid out. Ambient noise floor is a function mic placement - no recorder will fix that.

[@Humcrush]

No thrashing or trashing, but I still think you unduly discount the benefit of one less thing to worry about. Removing one factor, gain, doesn't mean you automatically become lazy about every other aspect of capturing good location sound. In truth, it frees one to better focus on placement, etc. if one so chooses. As a director, something I try to install in co-workers is that mediocrity isn't evil. Mediocre is a good baseline platform from which to launch towards greatness, because when you fail, you only fall back to "good enough", not the whole way down to smacking the floor of unusable.

[@Humcrush]

PS. In video, extra bit-depth comes at a pretty steep file size cost, so there needs to be a pretty damn good reason to burn through the extra storage 12-bit or 16-bit would require over 10-bit. For audio, though, the file size difference between a 16-bit and 32-bit float recording is not really meaningful. Which is to say, there's not really much reason to NOT capture 32-bit float if your recorder offers it.

[@FilmmakerIQ]

I clearly don't see it as "freeing" at all. I don't see how turning a knob slightly takes attention away from placing a mic -it's all part of the same package of doing the job. That you would suggest that it does feels like you're advocating laziness. It's not just mediocre, it's you don't care. But the merchants will gladly take your credit card number to tell you otherwise.

[@Humcrush]

@@FilmmakerIQ Sure I care about getting good audio, but I also have to get a good image and most importantly I have to get a good performance. I rely on a tripod to keep my camera steady...is that lazy? If a 32-bit recorder allows me to devote even 0.5% more of my brainpower on the day to an actor instead of a piece of gear, then that's a win.

[@FilmmakerIQ]

False assumptions and analogies. IF your recorder allows it - I'm not saying you shouldn't use it - said so in the video. But I don't see it deriving any benefit or freeing at all and I certainly don't see it as justification to spend the money if you don't have to. But really, 0.5% more brainpower to turn a knob? I'm sorry but that's not a win, that's just making up bullshit excuses.

[@Humcrush]

@@FilmmakerIQ You say "I'm not saying people shouldn't use it", but then you're implying they are lazy if they take advantage of it. You seem to be drawing a false equivalency between not caring about managing gain, and not caring about final audio quality. Those two things are only connected because of a technical limitation in a previous generation of equipment. I used to have to care about how film was processed in order to care about color. Now I don't. Does that mean I don't care a lot about the color of my films?

[@j7ndominica051]

So, how much dynamic range can these new recorders actually do? Seems insane. There is no microphone than can withstand loudness greater than 150 dB. I don't think you get 10^38 discrete levels, because the steps get bigger as the level increases. It's the same number of levels as in integer but they get redistributed unevenly. Think about it: you have 2^32 possibilities in the data word regardless what meaning you assign to it. The instantaneous precision is 25 bits, including the sign bit, 24 above and 24 below zero.

[@FilmmakerIQ]

After reading up more and more on it - it's 32bit float is just 24 bit float with an exponent to add a ton range. Because of the nature of audio, the larger skips in distance between steps at the extreme loud end don't really matter anyway. I would wager these recorders aren't any better than 120db dynamic range - IF that. They just use a recording format where the allocation of that dynamic range just doesn't matter.

[@ovi.1000]

Hi John, Great explanation! I'd expand a bit the kind of advantage for 32bit float. Your suggestion for 24bit recording, to lower the gain to -20dB to avoid clipping (then raise the gain in post) works perfectly with high fidelity equipment that has lower self noise. Using a not so good microphone or input device that it has higher self noise, setting from the beginning a higher input level to (-10dB to -6dB) it mitigates better the self-noise instead, it increases the risk of clipping. In my opinion, this kind of situation could take advantage of using 32bit float. Please, let me know if you agree with this.

[@FilmmakerIQ]

Nope. ;) Recording -20dB would also record the microphone's "self noise" down -20 dB. Anything coming from the mic will be completely unaffected in terms of noise. The recorder's own self noise would be the only thing at stake here. But even old beaten up recorders have EXTREMELY low noise especially if they are capable of recording 24 bit. If you purchase a recorder capable of 32 bit, it's going to be even lower than that. From a practical stand point, not worth thinking about. So you'd get practically zero benefit on the noise front recording 24 vs 32 bit front. All you're getting is clipping protection. 32bit just pushes the gain staging process to post instead of at the time of recording. You're going to have to turn down the signal when you see it clip with 32 instead of raising the signal with 24 bit and recording at -20dB. The ratio between recorded signal to noise stays the same in both scenarios.

[@ovi.1000]

Thanks for reply. You're right about self noise. I have Zoom H1N and I have to test it with less gain (as you said it) and no limiter instead. I found the limiter introducing noticeable noise. BTW, I discovered I have as well a device which records 32 bit FIXED Point - It is a GoPro9. I haven't yet tested how much self noise it has.

[@FilmmakerIQ]

I can't find any evidence that GoPro 9 shoots 32bit fixed. How did you find that out?

[@ovi.1000]

@@FilmmakerIQ Subject: "GoPro 9 shoots 32bit fixed" First, about setting from GoPro Hero-9 Black, user manual, p. 143: "Tech Specs: Protune / RAW AUDIO This setting creates a separate .wav file for your video, in addition to the standard .mp4 audio track. You can select the level of processing to apply to the RAW audio track. RAW Setting -> Off (default) No separate .wav file is created. RAW Setting -> Low: Applies minimal processing. Ideal if you want to apply audio processing in post-production." When I open the file in Audition, I can see: - Source Format: Waveform Audio 32-bit integer / That should be 32-bit fixed - Bit depth: 32 (float) / I suppose this is the processing format chosen by Audition to open and process the file, once it was accepted and recognized.

[@FilmmakerIQ]

Kind of annoying that the manual doesn't say... But I guess it makes sense to record 32bit on something that practially has no audio gain control - 32 bit float would be overkill (and require another ADC)

[@PB-kf4qt]

Hi! Thanks for the info. I'm a filmmaker working on a Direct Cinema documentary and not using a boom operator, as it's a very intimate situation with several characters, and they don't feel comfortable with one. Quick question: Can recording in 32-bit help me bring distant sounds forward in a room with multiple people talking, while also preventing clipping from sounds too close to the mic? I know it's not a professional setup, but since I'll have to handle the sound myself, do you think a 32-bit recorder can manage this?

[@FilmmakerIQ]

No. Get yourself multiple mics and a multi track recorder.

[@PB-kf4qt]

@@FilmmakerIQ thank you!

[@hbp_]

I have an additional point for you that I just realized recently. I was looking into how different manufacturers build the 32bit float signal. There isn't so much information available publicly but Sound Devices has a patent covering one method. What they do is that they flip from ADC to ADC using a "switch". They also apply a transition gain to smoothen out the signal level change. I have noticed that the background noise (cars etc) with my F2 is getting suddenly louder and even goes back and forth quickly. I have heard this doesn't happen with F6 though, at least not noticeably. I might be wrong here but it sounds like the 32bit float signal might even need more work in post, if whatever you are recording is hitting the "swap point" in a way that the output is jumping back and forth. I'm almost like wanting to buy one of those recorders and try to find patterns that cannot be recorded properly 😁 I suppose Sound Devices might be doing this better than some of the other manufacturers.

[@FilmmakerIQ]

Weird - are you doing some kind of auto leveling in post. The way I've heard it described - 32 bit doesn't negate setting levels - it just pushes it into post.

[@hbp_]

@@FilmmakerIQ nope. I believe it might be just specific to F2 really.

[@Robkellysound]

I didn't realise I had missed this trend. When did 32 bit float become a recording format? Mixing is a different though, last I heard it was just a editing and mixing consideration.

[@FilmmakerIQ]

In 2019 a few of these devices dropped and now people are telling me it's game changing.

[@A-Mana]

I think the most useful two features of 32bit are 1- Can save your clip "in case" some distortion happened 2- can handle over processing better (just like RAW photo when editing) Maybe not super useful for some, but definitely an advancing in audio technology that I welcome.

[@FilmmakerIQ]

Yes you can save clipped audio but 32 bit float doesn't handle over processing better than 24 bit ( it's used in the processing step and then bumped back down to 24/16). It's not like RAW photo editing, the analogy doesn't hold up because digital audio goes through digital to analog conversion before it hits your ears. But if you're going to torture the analogy, 24 bit is already shooting RAW, 32 bit is just 24 bit but with the ability to capture a crazy range. Other than the never clipping bit, you will never be able to tell the difference between 24 bit and 32 bit recordings

[@kierenmoore3236]

The example I’ve had thrown at me. By others advocating for 32-bit, is recording raindrops and there being an unexpected thunderclap … Do you think that could possibly clip a 24-bit recorder on which the raindrop peaks are set to -30db … ?

[@FilmmakerIQ]

I hate that example because it doesn't actually reflect any real world use case. It's literally a talking point for those that don't know better. If you were intentionally recording raindrops why would you want a sudden extremely loud thunderclap in your recording? It would be absolutely shocking to the listener to hear that extreme dynamic range change (don't know if you've been close to lighting... That stuff can get extremely loud to the point of pain which is 140dB SPL... Rain might be around 40 SPL). Even if you recorded at 32bit float you still have to contend with the extreme dynamic range change. If your mic survived the close up lightning strike (most are only rated for 120dB SPL), when you go to lower the volume in post, your raindrops will drop down to nothing... And if you're original intention was raindrops, well they're gone... Either way you pick one or the other. Now if you're asking about distant thunder... That's easy. Plenty of recordings exist of rain and distant thunder and been done on equipment that didn't even have 16bit of dynamic range. With 32bit float... Okay you can capture thunder up till your microphone maxes out... But you're still going to have to chose which one you finally set the volume for.. the rain or the thunder, not both.

[@JulienALVES-yp5wm]

Well, sorry, but as a sound post guy, I disagree. It’s entirely possible to take the clipped samples of said thunderclap and bring them back to a 24bits window that is compatible with the rest of the sound… so yeah, you can have both, we do it literally everyday in sound post… Try to record gun shots in 24bits (no you can’t just avoid clipping by properly setting preamp, because the sound is way more defined by what’s coming after the very first clipped samples… which also count btw!) Difficult case, which we used to solve by using at least 2 tracks with different gains (same mic, one track not clipping to properly capture the very beginning of the sound, and another one heavily clipping to get the ending of the sound right, the bigger part actually) and the reconstruct a proper signal… Or you can go and record 32f and then simply recover the first clipped samples, as well as the late way too low... (mic noise is to be taken into account, though for the very low sounds…) So please stop advocating 32f is useless en field recording, it is dead wrong when we speak about sound effects recording, it’s not a pov, that’s just what we do all the time ! I would add that for video shoot, I totally agree, nothing will ever replace proper gain staging and preamp management. (Even if we can have 32f safety nets in lavs nowadays) Also worth noting that a full 32f production would require additional workflow steps and costs for ponctual benefits. So… no, certainly don’t go full 32f all the time just because it’s safer on principle. It is safer, but it’s only really meaningful for specific sound effects recordings…

[@FilmmakerIQ]

​@@JulienALVES-yp5wm You didn't understand what I wrote or said.

[@zwheels654]

This basically confirms my suspicion. Essentially, I have the same cushion using a -6db safety track and my recorder is almost 10 years old. But, it does everything I need it to for my productions. Why change it?

[@FilmmakerIQ]

Well -6dB isn't the same cushion as -20dB but if it works for what you're doing, keep doing it.

[@zwheels654]

@@FilmmakerIQ Agreed. That -6db is for the safety track, while I target my peaks on my main track for -12db. So that gives me almost a -20db pad but makes for a little less editing in post as the main levels are closer to what I'll actually be using them at.

[@theobviouschicken]

thank you for demystifying certain parts of the 32 bit sounds v 24 bit sounds. easy to understand without going into heavy detail. I am going to use 32 bit as I am a stupid that likes to put mics on motorbikes and record themselves talking in 100 kmh winds and a 100 db exhaust. I suspect that even 32 bit cant help me there. in the past I have been recording 16 bit with -13 db on the recorder works ok but when I roll the throttle and the wind gets up it clips ( ALOT ) no matter what ( expected ) got a new recorder+mic that can do 24 and 32. Thanks again for making the content and the examples.

[@FilmmakerIQ]

Be careful as SPL can increase or decrease depending on distance to the source. So even if 32bit float can cover the range of sound, your physical microphone may not be able to capture the sound of exhaust if it's right up by the tailpipe.

[@theobviouschicken]

@@FilmmakerIQ Thanks I have the mic inside the helmet to capture what I am saying and luckily ( unlucky for the public ) the harley is noisy enough for it to add a subtle flavour of hydrocarbon burning neanderthal undertone to the idiocy I am spewing onto the internet. Testing it this weekend with different gain values. It's one of those instamic things which wedding people use. Hoping it solves a few other issues with lav mics in helmets. Im also going to test it behind me out of the wind for b-roll audio ... b -audio.

[@FilmmakerIQ]

Just FYI, if the mic is inside your helmet, 24bit is plenty enough to handle road noise - so long as you follow the -20dB rule for peaks ;) But 32 bit will be fine if your don't want to mess with that.

[@theobviouschicken]

@@FilmmakerIQ I'll try that as well. the instamic does 24 bit stereo. but 32 bit mono which is ok but to my earholes I like proper stereo. Biggest issue is wind noise got a deadcat coming to fit but want to play now ! 👍

[@FilmmakerIQ]

Are you running two mics? If just one mic then mono is all you're going to get.

[@sergiomartinez5946]

I'm noticing a lot if people having misconceptions of 32bit float as being a "set it and forget it" miracle working format. You still have to make sure you're capturing a good signal. If I'm filming an interview and the mic is 50 feet away from the subject 32 bit float will NOT fix that.

[@sergiomartinez5946]

I just bought the Rode Wireless Pro because I needed a ecorder. I think for the price it's a good deal with everything it comes with and had timecode! One of the reasons I bought though is because it has 32bit float. I plan to do a lot of "one man band" recording. Though my strength is film I still am familiar with audio settings. When I record I plan to set audio levels correctly as I always do. However I'll be glad to have 32bit float for those uncertain times. I see this as a tool not a crutch. I'm hoping people understand this. That's one thing I think is a bit of a downside to technology getting better. While it makes things "idiot proof" it turns people into idiots because they don't understand the basics of whatever it is they are trying to do. Bad capture is bad capture even if you have 32 bit float. It's the same as people thinking you never have to have good lighting because youe camera is shooting in a RAW format. Your image will still look bad. Speaking of idiots, you've opened this one's eyes to how much dynamic range a typical 24bit recording can capture. Time to brush up on my audio technique. I want to add that tbe Rode Pro records 32bit Float to the unit itself. I'm connecting it to my camera and recording audio straight to my camera (which won't be 32bit float) which is why I would want to set levels correctly. I would only use the 32bit float recording from the units themselves only if I really screwed the pooch on my audio levels.

[@FilmmakerIQ]

I was under the impression that RODE GO II was 32bit float.. it is not. That was slightly misleading in their ads as it can convert to a 32bit file but not actually record at 32 bit. The Pro version can record at 32bit. Recording at 24bit is still useful. Most mics including the one on the Rode Go are only capable up to 120dB SPL. 24bit can cover 144dB... So the entire range of what that mic can do is already covered by 24 bit!

[@RavikantRai21490]

@@FilmmakerIQ So then John, this confuses me. Because if the clipping can happen right at the mic level, what even is the point of 32 bit float recorder with mics like that? And, assuming you have a mic that can do even better than that, you still hit the 144 db limitation of current mics, right? So....I still fail to see how 32 bit can "save" you in any scenario. This is kinda confusing to me even more now.

[@FilmmakerIQ]

That is right - you clip at the mic level, 32 bit float won't save you.

[@AllenCavedo]

Nicely done video. Good info and good demos. There are many 32 bit discussions out there but very few that demonstrate the dynamic range of 24 bit recording.

[@TylerDarlington]

Whether or not anyone disagrees or agrees with you, you make awesome videos with a well educated points of view. I appreciate your thought out arguments and your willingness to engage with your viewers. I appreciate the slight debate and it makes us, and I, smarter.

[@mlem567]

huh this technology should be great for my asmr metal headbanger channel

[@bizylex]

Your technique and explanations are great and I agree that 32bit float is not necessary in most situations especially if you dial in your settings correctly. But with that in mind I asked the company I worked at to buy the zoom f2 LAVs because I am not always the one filming, sometimes it’s an intern or an employee that doesn’t have any idea about audio recording . We used to use a normal lapel recorder but they could never manage to get clean audio because they treated it as plug and play directly, with the 32 bit float, I use it as an idiot proof way of being sure that the low budget rushed corporate/marketing videos sound decent without requiring someone knowledgeable on « set ». This company was never going to put more effort in their content, and while it lacks any substance and quality at least they can manage decent audio thanks to 32 bit float (I don’t work there anymore but discovered the benefits for people who want a set it and forget it solution)

[@MrArtiisan]

finally someone says it

[@kollegeturnschuh5181]

I use the MixPre-3 II for podcasting with remote guests (I ship the equipment to them), most times they are not tech savvy at all. So having it set to 32bit float it kinda makes it idiot proof as long as they are able to find the Record button :)

[@FilmmakerIQ]

Okay, I'll buy that argument. 32 bit float is for idiots :P

[@elvisripley]

I deliver 32 bit a lot so it shows up in the editing software loud and all software has more range to lower audio level and not as much to raise it. A lot of faders go from +12 of gain and -60 to -♾ of cut. You can always boost with a compressor or gain plugin but it is nice to just turn it down a bit if needed.

[@djp_video]

I might have missed it, but I don't think you even mentioned that even 24-bit exceeds the dynamic range of any real-world microphones out there. Going to 32-bit for recording is just wasting storage and processing power.

[@FilmmakerIQ]

I agree with the sentiment but the issue with that argument is recording levels are not tied to real world sound pressure levels. Also 32 bit float is actually very computationally efficient (it's from 1985 after all) and the space needed isn't really that much in the grand scheme of things. But you are right that 24bit already exceeds pretty much every other part of the audio equation. So much so that in order to derive benefit from 32 bit they have to link TWO digital to analog converters!! :D

[@winterhell2002]

Are there even recording devices precise enough to have 24bit/16.7 million distinctive pressure levels, let alone more? I've heard the physical limit in professional equpment can go around 20-22 bits.

[@FilmmakerIQ]

I think your question is a little misleading. The microphone signal is analog, in essence the microphone has infinitely variable "pressure levels". The infinitely variable signal is quantized and sampled using a digital system. Now at the same time there's also noise in the system and noise in the form of quantization noise which is the error that's made when converting the infinitely variable analog signal into a discrete digital form. That noise is why you lose a few bits and they say a 24 bit file only has about 20 useful bits. Which is still a lot of room to work with.

[@TheRealHarrypm]

As a videographer, who actually knows how to use my limiters I use 32-bit float on the Zoom F2 which is camara mountable, the reason why we like the new 32-bit float units is they also offer set and forget pre-adjusted output level for monitoring or in-camara recording ready for live use or for playback, its made audio idiot proof for field work as you don't always have to post process audio, its all done for you on the little DSP built in with only a few grams more weight to most rigs, and is salvation for SDI digitisation chains. But I get your point its like saying 10-bit is better then 8-bit sampling for analog video In the context of analouge FM direct capture for tape media for example it does not matter past 8-bit if the msps is high enough as its not a live signal the dynamic range potential is capped at an analog level like you can do 50msps of sampling at 12-bit but there is only 20msps at 8-bit of valuable data in real world use, but people think oversampling is just better, the same idea of thinking can be applyed to audio but its on the scale of GB's per hour not TB's per hour, you capture only slightly more information then you need, and distro the mimimum practical.

[@FilmmakerIQ]

Does that necessitate 32bit float though?

[@TheRealHarrypm]

​@@FilmmakerIQ As a output format not really, but due to to size of the F2 there really is no space for a physical control surface so it makes sence that 48khz 32-bit float is the only option internally, as far as I know its the only bit of kit this small with this DSP feature, though Rodes new wireless systems also have some intresting features.

[@FilmmakerIQ]

I have two sets of Rode Wireless devices. I run them at 32 bit internal recording but I have never actually had to use the feature set of 32 bit in that I've never had to lower the gain to restore a clipped audio. That's not due to skill, that's probably due to the internal software and perhaps limiters on board. So... I'm not sure how 32 bit plays into any of it or even if those Rode mics are ACTUALLY 32 bit or just able to put out a 32 bit file.

[@TheRealHarrypm]

@@FilmmakerIQ They are genuine duel ADC devices, not as good as sound devices but still soild modern all in one ICs would be nice to see board level tare downs though. IIRC there capsules have not changed since the new smartlav+ version which are very good at the price point. Its more DSP then it is the format nowdays thats ware we see the real world improvements more and more hands off as you have seen first hand.

[@FilmmakerIQ]

So I checked again because I had the Rode wireless plugged into a wireless reciever and it got a little hairy on a singer... well the Rode Wireless Go II is certainly NOT 32. What's annoying is Rode Software gives you the OPTION of exporting a 32 bit float file, but the devices themselves record 24bit... So... yay... marketing... Oh well, it's just a little bit over the top when the singer hits the highnote - probably sounded like that in the room as well.

[@jasonsmall5602]

32 bit float does not have more values than non-float. It has a higher range. There are still the same number of discrete values. Not all values in the range of float can be represented. There are holes.

[@FilmmakerIQ]

Someone else pointed that out. Added the error in the description

[@endokrin7897]

Gah! Too many smart people!! 😉👍

[@jasonsmall5602]

@@endokrin7897 nah, just educated in the subject matter.

[@EvilErwin23]

I'm not an expert, but for me the 32 bit floating point recording sounds for me like they solved a problem no one had with a solution thats basicly "to take a sledgehammer to crack a nut". Next they invent a new "Ultra-RAW"-Image format for cameras and say: "Look, with this you don't need to set the aperture anymore. You shoot and set the brightness of the picture in post-production. And now you don't have to worry about any random flash of light anymore."

[@ginglyst]

a new "Ultra-RAW" image format isn't required. Even with today's RAW format you can ignore exposure settings and still get the same results in post. Two prerequisites though: as long as you expose within the dynamic range of the image sensor and if you have the imagesensor characteristics data (obtaining that information is not so straight forward) So in real life it's just easier to set exposure correctly and tweak the RAW-files in post.

[@EvilErwin23]

@@ginglyst I was trying to make this point: With 32 bit floating point recording they replaced a work step you do before you record with a work step you do in post production. You gain the security that your recording doesn't clip but it would not clip in 99.99% of all cases if you do your first work step corectly.

[@FilmmakerIQ]

"ignore exposure settings and still get the same results in post" That's just flat out wrong and you explained why in your next sentence :P

[@ginglyst]

@@FilmmakerIQ oooh flat out wrong, you calling me out on that one?

[@FilmmakerIQ]

You called yourself out on it :P

[@benjaminreed6816]

People don't seem to like common sense anymore. This is a great video. What are your recommendations for an affordable recorder with decent preamps? Started on a zoom h5 gets pretty noisy recording really quiet sounds.

[@FilmmakerIQ]

The question is what is your noise source? My experience is that the H4 is pretty decent from internal noise. Are you picking up environmental noise? Is the mic introducing noise itself? Is the cable solid?

[@cuentaparadeciridioteces3648]

The preams on the h5 are pretty decent, in my experience setting the gain to 6 is pretty much the limit before the noise of the recorder starts to become noticeable. The problem comes with mics that need a lot of gain, such as the xy mics included with the h5, in those cases you'll have to push the recorder above 6. Of course this isn't a problem if you're recording loud environments. If you want to buy something new, I'd say you get a MixPre 3/6 or a Zoom F4/F8, the step up in preams quality is notorious with those, unlike a H4 and an H6 for example.

[@ralfbaechle]

@Filmmaker IQ - some folks like me rely on reasonably high audio levels especially on crappy volume limited audio such as some business laptop which just are lacking the umpf to play audio at an acceptable volume if a room isn't perfectly quiet. With that experience in mind I'm just begging everybody to use the dynamic range available to deliver audio that doesn't require the listener to artificially boost the audio on his side or play with the volume for a good audio experience.

[@FilmmakerIQ]

One of the worst audio levels experience I've ever had recently was live streaming a school board meeting where each school submitted a video progress report. To say the levels were all over the place would be an insult to something that was all over the place. I mean half these people didn't even believe in second takes... Luckily I offloaded the audio job to my friend who was helping me haha It is what it was.

[@stevenclark2188]

So beyond 8 bits in an image color channel isn't really perceptible, but it becomes perceptible when manipulating the dynamic range of something like a negative film scan in a digital darkroom process. Meanwhile the difference between 8 and 16-bit audio samples is pretty audible. Beyond that probably isn't perceptible but you would need the extra precision when recording to capture a wide dynamic range of audio without risk of clipping, because you'll expand a chunk of that range working at the workstation later to get your final product.

[@FilmmakerIQ]

Be careful trying to draw comparisons between sound and vision. Sound is much older bio tech and works quite differently involving different parts of the brain.

[@2by3]

You cannot compare 2 sound signatures concurrently but you can visually compare multiple color and gradient samples. Our brain can compare very short audio passages, still our sound memory is very limited.

[@stirado]

Thank you for another exceptional video. I've learned so much from your detailed explanations. Keep 'em coming.

[@mr88cet]

Great topic! Minor nitpick: time 5:59 20dB is 10x power. The amplitude is 10x, but power is v^2*R, so 20dB.

[@FilmmakerIQ]

Ugh - I got some confusion there with volts.... 20db is 100x POWER - it is 10x the amplitude. en.wikipedia.org/wiki/Decibel

[@mr88cet]

@@FilmmakerIQ, well, that’s interesting! I got out of EE school back in 1983 and have been a Computer Nerd ever since. So, I looking at that Wiki, I’m either remembering wrong, or there are different definitions of dB going around (beyond just what the definition takes as the reference amplitude or power level). I think it’s more likely the latter. I found a Quora chat “Why is a decibel always in the form of ‘20log()’?” (KZbin doesn’t seem to let me post the link, sorry.) Anyway, what I remember from EE school, 38 years ago, was that it was defined in terms of *amplitude* ratios. The gain of an amplifier is the output *voltage* over the input *voltage* . So, if its gain is 10, then the base-10 log of 10 is 1, meaning one Bel of gain, or 10dB. However, what’s most interesting is the *power* increase, and since p = v^2 * R, the *power* gain is double that, 2B or 20dB. So, I guess it gets back to there being several definitions of decibels going around! Yeehah…

[@mr88cet]

Either way though, great topic and great video! Thanks!

[@FilmmakerIQ]

Yeah I remember this trap from way back when I was first researching this! Tricky because I went to marketing school :P

[@rickh6963]

What the industry has accomplished is to give the hardware weenies another excuse why they can't make a film until they acquire this tech! ........ OH NO! I can't shoot my first really bad film until I get a RED 8K camera with 32bit floating point audio!!!! ..... Which probably saves us all from some really bad videos.

[@soundsaroundme8766]

I think the 32 bit float will be beneficial for a drop rig in nature recording, when you can't predict how loud or how whispering the nature is

[@FilmmakerIQ]

Sure... but that represents a tiny tiny tiny fraction of audio recording.

[@andres160982]

@@FilmmakerIQ dude thank you for the video. Sooo helpful and well explained. I'm a newbie and I was thinking about that above mentioned nature recording. So there is an advantage right? If I want to capture ambient sounds...

[@CUTproductionsLtd]

Your scientific videos of filmmaking are great and entertaining and I've learnt a lot about exposure I never knew. I'd just like to point out that this particular video whilst obviously common sense and true, for recording of dialogue and music, perhaps misses one fundamental point of where and why 32bit float might be used in a professional context, no matter the skill, talent of the recordist and quality of the digital recorder. In recording SFX (or for scientific analytical recordings perhaps) you might want to record such a wide dynamic range, wider than our own even, as to make it almost impossible to cover that with compression, limiting or gain-staging even with 24bit: For example soft rain during a thunderstorm with terrific lightning. Any recording chain will of course still be limited ultimately by microphone and ADC DR but given sufficient SPL ability, padding and technology such as multiple analog-to-digital converters, you can record this huge DR, where 24bit might not without distortion and or noise. Obviously 32bit float in prosumer recorders, is intended to appeal to the non or semi professional recordist and professional recording will normally be with gain-staged 24bit, working to proper head room standards such as -18/20db, but the ultimate DR of the final medium (usually much lower than people assume, for our comfort and intelligibility, even with classical music) requires often neither and is not necessarily the only reason to use such techniques in the first place.

[@FilmmakerIQ]

That is such an edge use case that it's ludicrous to consider. You are talking about soft rain with an incredibly close lightning strike... Close enough to physically hurt you... For scientific analysis maybe (even then these recorders may not pass scientific muster), but not for anything with an aesthetic purpose. Those extremely rare edge cases are really not worth discussing. If you need it, then you know why you need it. But for the 99.9% of users, you accomplish the same thing with proper gain staging.

[@CUTproductionsLtd]

@@FilmmakerIQ Nonetheless this is one of the abilities of 32bit float recording as opposed to processing. I mentioned SFX recording and in brackets 'perhaps' scientific recording, which in truth is not my area. You don't need to be dangerously close to lightning for it to exceed the DR of 24bit - what is 'ludicrous' is to imply I was suggesting that. Your claim, in the title of your video, is 'You don't need 32 Float..." without qualification, such as, 'for indie fiction filmmaking', and having been a professional location sound recordist and dubbing mixer in film and TV, I'm saying you are not right, in all cases and I know of professional sound designers and SFX recordists, who are using 32bit float to record SFX for filmmaking, for the case reasons I exampled. Similarly, for example, there could be many unpredictable wide DR news, documentary and broadcast scenarios where 32bit float could be very useful and the ability to bring sound out afterwards that might be lost otherwise - not everybody is shooting controlled fiction, even indies. Already I would think this exceeds 0.1% use for 32bit float recording you suggest. You could say you don't need 24bit.... too, for your definition, which I could easily make a case for on your terms also, 16bit is fine - then we can dismiss any new innovations as unnecessary gimmicks because we don't own them. 32bit float is not just about not having to set levels because you are too lazy or don't know how. It's like saying you don't need to shoot in raw, you can shoot in Prores and expose properly, like professionals have done for years, which is equally true - new technology offers greater flexibility for both the amateur and professional - In the interests of full disclosure and due diligent research, I would of mentioned, if only in passing, that in your explanation, unless you have a clickbait reason for such a dogmatic position and title, then I understand.

[@FilmmakerIQ]

You still miss the point in that 24 is ALREADY the insurance against clipping in uncontroled situation. 24bit IS the RAW high dynamic range. As for the lightning strike, yes you do need to be dangerously close. Thunder in close proximity is 120dB, 24bit is capable of 144dB. But that also depends on how you gain the signal. My whole problem with the rain and thunder scenario (which others keep bringing up), is it's not realistic. You either want to record the rain or you record the thunder in close proximity. There's no real practical use where you want both in there full and original volumes. BTW you never shoot LOG in news or broadcast, because you don't have time for grading in turn around. So you'd stay in a broadcast color standard with limited DR. Anyhow, as for SFX guys, I'm sure that case use and situations where they ACTUALLY see benefit is probably less than 0.1% of the TOTAL use among all people of these recorders. As I said, if you really need it, you'll know why. Otherwise, you don't need it. That's not just click bait, that's from experience and understanding how this stuff works. 32 but float isn't innovation, it's marketing. I already own F8N and the 32bit float in the latest version would add nothing beyond what I'm doing with 24 bit. And I've only used it once in a controlled setting, every other use was a set and forget scenario capturing whatever people in front of the mics decide to do. Anyhow, I'm just repeating everything I said in the video, so if you want a better understanding of my position, watch the video with less bias, really listen to try to understand the argument.

[@AT-27182]

Great explanations and demonstrations. Thank you. Audio specifications are necessarily tied to what we are capable of sensing, what nature produces and what our technology can delivery. Our technology has now reached the point that it is no longer the weakest link in the process and any improvement is just for marketing bragging rights and not for really getting the job done in any objective way. But marketing bragging does tend to create cash flows, so I imagine we will keep seeing new products that cost more or try to make us dissatisfied with what we already have. Nothing wrong if someone wants to buy audio jewelry or audio make-up, but it is important to know that it is just jewelry and make-up and that it is not technically necessary.

[@marvinglenn]

@7:30 32FP is not going above zero dB because it has the sign bit; it's going above zero dB because the reference point of what is treated as full-scale/zero-dB is not its maximum numerical value, like is selected for the linear encodings.

[@endokrin7897]

If you ask me, these manufacturers are hyping this up a BIT too much. 😁🤣👍

[@QuantumFirefly]

Okay, I'll byte. How much more?

[@brockelman]

Thanks for this. It's an outstanding tutorial and I'm glad I found it and your KZbin channel! Excellent work.

[@yash1152]

hey, i am not into audio recording. but the levels and storage etc are a thing of my interest. so, are there other floating point values as well? like say 24 bit floating or something?? anyways, awesome video. it's good to have ppl breaking the marketting hypes.

[@FilmmakerIQ]

Yes there is but they're not as well implemented

[@yash1152]

@@FilmmakerIQ thanks for answering

[@helmanfrow]

This video sums up the remarks I leave in the comment section of every other video touting the marvels of 32-bit float. I suppose I could simply link to this video but I do so enjoy my pre-bedtime troll.

[@TheREAPERBlog]

oh hey! 👋

[@helmanfrow]

@@TheREAPERBlog Well, fancy meeting you here!

[@lohphat]

There are some great explanations of IEEE 754 online but instead of the mechanics of it, there’s an easier way to explain the concept without having to go down the implementation rabbit hole. The exponent defines the power of 2 exponent “block” of where the desired number is located, then within the two values of the exponent, the mantissa encodes the closest location in that bloc of the desired number as a percentage between the exponents. The more bits a bailable in the mantissa, the more precise it is BUT by definition floating point a only an approximation. It is in a way a form of data compression where the entire spectrum f real numbers accessible between the scale of Fe exponent rage is encoded in n-bits. Thus, floating points not intended encode discrete values, but only approximations allowed by the available bits to describe the percentage of the numbers location.

[@FilmmakerIQ]

Thanks for that, but I don't think that was easier...

[@MichaelWeizenfeld]

To my point of view, the magic not in 32 bits itself, but in two (dual) ADC fitting more analog signal range and increasing SNR accordingly. This also should lower quantization noise. Btw weak side is quality of mic and preamp which usually hisses a lot.

[@FilmmakerIQ]

True but 24-bit already provides as much SNR as you can perceptual need, outstripping the capabilities of most mic and preamps to begin with.

[@MichaelWeizenfeld]

@@FilmmakerIQ I found in-depth video about 32 bit float kzbin.info/www/bejne/rqqwhnSNjc-lnqM It answers many of the questions that plagued us.

[@FilmmakerIQ]

There's no mystery that plagues me.... why don't you watch my video and directly address what I'm saying? I browsed through that video and didn't see anything contradictory...

[@patrickang318]

Having clipped an audio in an interview, I admit it was my lack of practice, and I still support everything you’re advocating here.

[@FilmmakerIQ]

You live and learn. I've done it too.

[@CED3]

Thank you for all of your amazing videos!

[@georgeogrady449]

What about 40bit sound

[@Terry-nx8kg]

There is 64 bit floating point audio.

[@alexandregneouchev2415]

Hey, sorry to ne that guy but +6 dB is double the amplitude, the pressure or the tension but not the power that would then je multiplied by 4. There is a lot of confusion about it and that would be nice id you corrected that.

[@FilmmakerIQ]

Already added in the description.

[@alexandregneouchev2415]

@@FilmmakerIQ my bad. Thanks.

[@forbiddenera]

Two things I never see discussed with 32fp either is floating point precision errors, which I'm keenly aware of as a developer. Sure, you can represent a huge range of numbers but you can't represent every number in that range and the farther you get away from 0, the larger the gap between numbers. Try doing 0.1 + 0.2 in any fp32 representation, it doesn't equal 0.3 as you'd expect because fp32 can't represent 0.3, only 0.299999 or something ❤ as you say in your description, there's still only 4 billion values it can represent between that bajillion range actually a bit less because of the sign bit. The second is the precision of steps between different bit depths, eg how dBFS is mapped to dynamic range. There's no reason 16-bit audio couldn't represent 65536dB of dynamic range, if the step size between volume differences was 1dB; 16 bit with 96dB range has a 0.00024dB step difference, 24bit at 144dB range has a 0.0000014dB step difference, although that's partly because dB is log.

[@TiagrajI]

I can hear the neighbours doing aeronautic stuff at midnight and they mix spaghetti as well at that time 😂

[@ootrip40]

Great video. There is one question that I haven't been able to get an answer to, which in my eyes is so obvious. Following the signalpath the first thing you encounter is the analogue preamp stage. If you don't set that inputlevel correctly (say, too high), what is the use of having that massive 32bit float range if the signal is already is distorted?

[@FilmmakerIQ]

When switched to 32bit float, these devices essentially lock you out from controlling the gain. It's preset. You can still overload it though.

[@ootrip40]

@@FilmmakerIQ I see, makes a lot of sense. So the dynamic range of the preamp is way smaller then the digital 32bit float range and will always be the bottleneck as I assumed. Thanks for enlightening me!

[@FilmmakerIQ]

I believe they actually link up two preamps to expand the dynamics range but even then it still doesn't come close to filling out the total available dynamic range of 32 bit float.

[@mateuszwojt]

I can think of one instance where the recording would benefit from 32-bit floating point - recording guns. Gunshots are notoriously difficult to capture, even with expensive gear because of the huge spike in dynamic range.

[@FilmmakerIQ]

I think that's a technique issue. Even with 32 bit float you also run into the possibility of blowing out the microphone capability: the gun is after all a creating a shockwave. You can also overload the internal electronics of these recorders even if 32 bit float is indistructible The NT2 I demoed the saw on is only rated to 132dB SPL... The best bet is to probably go multiple passes, get some distance for the blast and another up close for the chambering of the round and all those noises... Marry them in the mix...

[@DGaryGrady]

@@FilmmakerIQ That's an important point: The digital file format has nothing to do with the what the initial analog sound capture is capable of.

[@av6162]

Freakin' awesome. Thanks for restoring my understanding of audio levels. I'm trying to not be sold on getting NEW tech if established tried and true practices are still valid. Now back to finding a nicely used Sound Devices field mixer. Cheers!

[@ATTACKofthe6STRINGS]

Awesome video. I have a sound devices mixpre 3 I because 32 let’s me be absolutely lazy with my levels. As a solo guy who wanted a recorder for the choir concerts I participate in, often not having time to set levels as I’m doing all of this while preparing to perform as well, the piece of mind of knowing I “can’t” vet the levels wrong is what I was paying for. But, it is also nice to know that I don’t need 32 bit audio outside of that situation, since I’ll be able to set levels correctly and I can always learn better techniques to improve. I love your content!

[@FilmmakerIQ]

You won't have any issues with 24 bit and a choir if you set your levels the way I described. I've done it many times

[@ATTACKofthe6STRINGS]

@@FilmmakerIQ Oh I don’t doubt you one bit! The examples you gave and points you brought leave me with nothing to say besides “you’re right”. I’m honestly just personally acknowledging both the ignorance tax and ADHD tax I pay to avoid stressing myself into the aether, lol To the first point: when I first decided to record my choir concerts, I just didn’t know the things you taught me in this video. I was learning about 32-bit recording as I was making the decision about which recorder I wanted to buy. If I’d have known what showed here there’s s chance I would have made a better decision, or at least made the same one in a more informed way. That said, the reason I ended up going with 32-bit recorders was actually just to avoid stress. I knew I’d have to set up my equipment on my one time, on the day of my performance, in between whatever dress rehearsals or call times I had. I might not even get time to properly set anything up. With my ADHD diagnoses last year, and my previous experiences with high stress, fast-paced environments, I’m not sure I would have been able to commit to learning how to record location sound with a field recorder without the massively overengineered safety buffer of 32-bit recording. I was essentially about to learn to do it on my own without any mentors or guidance. The decision of which recorder to buy was, itself, a stressful experience because my brain was already running ahead to how I would feel if I failed to capture the moments I wanted to capture because I was just going through the learning process. Now that I know what you’ve taught me, I’ll be better informed if I need to make another purchase in the future, but I also know I have to pay the ADHD tax somewhere. Either I medicate (been working with a psychiatrist, and it’s been going well!) Or, I force my ADHD brain to slow down, which leads to me taking a much longer time to do something normal people do to avoid missing things normal people wouldn’t miss. Or, I just avoid the problem altogether with 32-bit recording. It doesn’t matter what I do, how much I learn, or how well I prepare, I almost always mess up something moderately important up, no natter how many times I’ve prepared and practiced. While I know now I never needed 32-bit, I paid the price for the piece of mind that there is nothing I can do to mess up the most important part of what I want to record. Thank you so much for putting out such amazing content! EDIT: fixed a typo

[@DroseMr]

Can you get 32bit foat by connecting a 14bit mic like hollyland lark m1 to zoom v3 xlr via vxlr pro to 3.5 line to gh6 mic jack recording in camera? Im new. Thank you

[@FilmmakerIQ]

No. You should review my video on audio recording. 32 bit float is a recording format, that's all. I don't think anything in your audio chain does 32 bit float

[@Sayphonik]

General rule of thumb… you can’t make a bad signal “better” but you can make a great signal worse

[@marin_real_estate_photography]

Just wanted to say thanks for this (older) video as I am contemplating buying a Tascam DR-10L and i did have FOMO when it comes to the lack of 32-bit float on those old recorders.

[@DQSpider]

With you 100% on this. If the gain isn't set right, no amount of bits will save you from the source signal clipping in the preamp or being dumped into the noise floor. I am legitimately impressed by how clean that Zoom sounds with 30 dB added too. All internal preamps?

[@FilmmakerIQ]

I didn't do anything special... It's the same series from Zoom as the F6 with 32 bit float so I imagine they share some similar electronics.

[@Humcrush]

This isn't true. It doesn't matter where you set your gain on a 32-bit recorder with multiple analog gain stages. It's not unlike white balance in a RAW image...it's just metadata.

[@DQSpider]

@@Humcrush remind me to never hire you to do audio

[@FilmmakerIQ]

You can still screw up white balance in a RAW image so bad that RAW won't save you.

[@Humcrush]

@@DQSpider 1. You never should! 2. Have you worked with a something like a Mix Pre II?

[@tommihommi1]

I mean there's a reason many professional audio recorders can record safety tracks at lower gain. 32 bit float gives you that all the time every time (if you want), at not much if any extra cost, even in cheaper, more amateur-friendly equipment, and you might gain a few dB lower noise floor in some circumstances, as well as simplify the workflow a little. And unlike with raw video, the files aren't that much bigger, aren't much harder to edit etc. Massive game changer that marketing makes it out to be? Maybe only for extremely niche stuff like recording racecars, fireworks, thunderstorms, if even that. Nice to have? Sure.

[@FilmmakerIQ]

So... just record lower gain... What I find annoying is the constant barrage of "it's a game changer" when it's not. Nice to have? Fine. But does it justify replacing gear you already have? absolutely not.

[@jambononi]

The biggest thing for me is when I get a line in after testing the audio and then, say a dj or sound tech, changes the output once the event starts. I've lost gigs because they changed it (and genuinely they were also recording and didn't press record etc).

[@FilmmakerIQ]

Okay, I legitimately get that. Still think a -20dB headroom can still save the day in those situations.

[@jeremiahkuehne2400]

@@FilmmakerIQ Sure, and then there are the situations where there's a quiet signal and giving an additional -20dB headroom is just adding additional unnecessary noise when you can just record 32-bit...In a perfect world, someone who is able to monitor the audio would have no problem with 24-bit but a 32-bit recorder is a small price to pay for basically having an assistant.

[@gtrdrumsplayerduarte]

The most important stuff is noise floor from every component.

[@18thAvenuePodcast]

Thank you so much for this video! I’ve sub to your channel. I need to learn whatever you are willing to share about audio recordings! I am new to filmmaking, but been playing with cameras for a decade now, just never took audio serious until I decided to turn professional. So I am hear to learn. Love the video. I can tell what I have for gears if you had like!

[@Gregg0Palmer]

Excellent discussion! The issues I have seen with clipping usually involve transient peaks (laughing, coughing, door slams) , these can reach significantly greater gain values than your dialog. When not gained with significant headroom you can find yourself with clipping in the recording.

[@gillesvanleeuwen]

You didn’t mention the safety track.

[@FilmmakerIQ]

You don't necessarily need a safety track. I've literally never used one.

[@pauli9649]

When I record an event, I set one room mic to record the entire evening to catch any impromptu happening because I can always add video from photos or some B-roll, but not the audio. It's usually somewhere where I have no easy access to it ( high above ppl, back of the room, ... you get the idea ). And I'm going about my business recording with another setup, something else most of the time. This technology would solve my issues :) and you can't do it any other way. Imagine a wedding where people may speak with (PA 90db), without PA (40db), and sometimes music blasting as well (120db), which gives you about 80db spread. The dynamic is too great for any hard-set levels. Yes, the limiter helps but not 100%. So my point is, if you can monitor and adjust, use whatever you want, when you can't or don't know how to, use 32bit float. That is why we have full auto on cameras as well. If you can't adjust manually, or you don't know how to, you can still record video in focus and now audio "in focus" as well. :)

[@FilmmakerIQ]

Sigh... I've literally done all that with 24bit audio, I don't need to imagine it, I've lived it... you've got 144dB range to work with - maybe more realistically 120dB - 24 bit actually can cover your 80 dB range just fine. And like you noted limiter helps and extends that range even further. I literally recorded a metal chop saw and a whisper in the video... You got to get used to the idea of recording lower than you think. My F8n helps by showing me 0dB with 20dB headroom - but a lot of cameras and recorders have bars that go into yellow at -18dB - that's where you should be landing with your peaks. People have a bad habit of going way higher into upper end of the yellow. The problem really is, you're not going to get good audio recording that way (people on mic, off mic...) regardless of how many bits you record - it's going to be a jumbled mess - the lows are going be buried in ambient background noise anyways - 32 bit doesn't magically make the rustle of plates suddenly softer. If you record that way - you get what you get. 32 bits doesn't save you. (Also I would be puling the feed off the PA - so it's pretty predictable and for people that go off mic, switch over to room sound from a camera... because no matter what you do, the sound from the PA patched into the recorder will sound infinitely better than any recording of the sound in the room from the PA) But you bring up a good point - there's already an "Auto levels" mode on some of these recorders - just use that.

[@pauli9649]

@@FilmmakerIQ have you EVER in your entire life not clipped any audio because of wrong gain? Even before all your years of experience?

[@FilmmakerIQ]

I have not clipped any audio since I learned the technique I talked about in the video... and I've recorded probably hundreds of hours of event audio where I left the recorder unattented. It's not hard to do... the issue is you underestimate the capability of 24 bit recording.

[@pauli9649]

@@FilmmakerIQ ..."since I learned the technique"... :) exactly. This is for someone who didn't learn the technique or doesn't want to. You are not driving a stick shift either, right? You would get better mileage and more control over your car, but the convenience is not worth it. ;)

[@FilmmakerIQ]

Your analogy to stick shift is pretty weak. Setting the levels with 24bit is already driving automatic I don't understand why you would push back on someone showing you how to literally solve the problem for free just by learning basics without resorting to buying new equipment. You really bought into the marketing.

[@shutterrecoil]

So Mad Men is short at times when studio dialogs were not recoreded?

[@FilmmakerIQ]

The Mad Men clip is a reference to how companies market a product without distinction. Don't take anything about the dialog was recorded for the show itself.

[@TheUglyGnome]

Answer: Because limitations of your analog circuit (mic preamp) becomes the problem even before the limitations of 24-bit digital audio.

[@FilmmakerIQ]

That's why they put more than one in there! Because you need two.

[@adagiotexas]

The whole point of these new 32 bit float recorders is that you DON'T have to do gain staging at all. If you are a run and gun, solo filmmaker, this takes a lot of burden off your shoulders.

[@FilmmakerIQ]

It's only useful in situations where you CAN'T - being a veteran run and gun solo filmmaker for years - gain staging isn't something that's hard to do.

[@John-e4p1x]

Adagio you dont get it. Take 1 minute on the set to set the levels, and its done. OR... do what? let the 32 bit uh... be whatever level it is... then spend hours later in poast listening to everything to normalize it??? What? Its like saying "the camera record raw, so I dont need to do anything on set! Just point and do it later!!!" you are srupid. John is right. By the way i own a few 32 bit recorders, made by sound devices. I never ise the 32 bit it would be a waste.

[@elektrozil9728]

paused at Tzar Bomba, and swiveled my swivel chair.

[@sokolum]

Always thought every 3db doubles the output.

[@IDKOKIDK]

as far as signal, not as far as hearing. try it now, change the volume by 3db, it doesn't sound twice as loud or soft

[@FilmmakerIQ]

My mistake, 3dB is double the power - 6dB is double the amplitude (which means 4x the power)

[@mr88cet]

@@FilmmakerIQ, yes, so doubling the amplitude gives a 6dB increase in power. There appears to be (surprise surprise) several definitions of dB going around, whether they are defined as log10 of power ratio or log10 of amplitude ratio.

[@chuxmix65]

Dynamic range. Max Plank would give a thumbs up! Same basic principles apply with light, film, and/or sensors. Bit depth. Great video! I understand audio much better now!

[@princebanini]

Great advice. I will definitely consider your teachings in this video. You earned a subscriber

[@EquatorialVillager]

I think 32bit float is a tool. Just like autofocus is for cameras. Having it doesn't mean you get to ignore fundamentals and not having it doesn't mean you're severely limited. Eventually, most recorders will have it (32 bit float) and then we can go bicker about the next new feature someone dreams up. The most important takeaway from this video IMHO is for a new buyer trying to quantify how much benefit that single feature is worth today. If they are choosing from many of the latest options from Zoom, Sound devices, they may already have 32-bit float recording. If the previous/older/used model is available for significant $ less then that may be something worth looking at (depending on the other feature differences).

[@FilmmakerIQ]

I don't know if it will become universal because it seems to require additional hardware (a second ADC).

[@jorge.rubiales]

I work recording interviews that aren't staged nor scripted, and under no circumstance I can make a guest repeat an aswer. So I have to be on top of my game every second of the gig, and if something happens I would rather have those extra 10db for safety (I'm talking +10 because you're really working the limits of the microphone, not the converters anymore). Some people give you a false sense of security during soundcheck, and when they get emotional they start overmodulating their voices. Also, I wouldn't use a limiter on a recording of this kind, never. That just imposes your criteria over the post guys, and it will not be appreciated (you're basically choosing the recording dynamic range for them). Nice video though, and really for a lot of dialogue uses (specially if scripted or non-critical applications) 24-bit with decent enough preamps-cable-mics is more than enough to get clean recordings.

[@FilmmakerIQ]

My bread and butter are interviews and I guarantee you that 24bit is plenty if you practice the headroom I talked about. Nothing I shared is only for controlled environment. Essentially if you can't make 24 bit work it's because you're setting your levels way too hot. The limiter is only a last resort, if you're crossing 18db of headroom to actually engage the limiter it's probably not going to be a good take but at least you have something useable.

[@katbryce]

32 bits is still 32 bits, whether is is floating or fixed, with fixed they are evenly distributed throughout the range. With floating, you get more small numbers and fewer larger numbers, but overall the same quantity of different numbers.

[@FilmmakerIQ]

Yeah I biffed that. 32 Float is a about range, it still only has the accuracy of 24 bit.

[@Finn_Talks]

The takeaway I got from this is that turns out I don't know jack any shit about audio. Thanks, John. I learned somethin today.

[@pdrg]

It's like shooting an Instagram selfie with a 40Mpx camera

[@petermsull]

You're looking only at headroom but there's an even more important benefit. In linear bit distribution the (bit) points are fixed. So unless you're using all the possible gain, up to the LSB, your *resolution* suffers: a low level sound might be only 8 bit or possibly worse. The benefit, and why I played a part in pushing the production of 32 bit FP, I'll explain by way of an example. Say you're recording a scene where one person is shouting and the other is crying/whimpering. The level would, in linear land, have to be set (obviously) for the loudest and the quieter would be at a much lower *resolution*. Now in floating point the bits, well, float, meaning our bit load is being redistributed constantly (save a couple 'safety bits') and thus what may barely register on the meters is at *FULL* (or nearly full; for all practical purposes full) *resolution* and not at a significantly lower *resolution* like 10,8 or 4 bit. To sum: I pushed like hell for floating point (as a sound effects guy not a production mixer) because to me it was all about *resolution* and the gain benefits I look at as merely a mathematical byproduct; I have recorded sound effects for film all over the world for many decades and what frustrated the heck out of me (as I began with Nagras) was low level = low *resolution* which is most frustrating when recording things (largely) in nature that didn't provide enough level to gain the benefits of whatever bit depth. Thus, after a decade of bitching and moaning, I began my crusade in earnest. I have a parable I've used with Jim Cameron and teamsters if you like.

[@FilmmakerIQ]

I do not buy the resolution angle at all. That's not how audio works. And I demonstrated that you can capture a shout and a whisper in the exact same file using the same gain using 24 bit because the range is not as extreme as you make it out to be.

[@Md2802]

If you're recording at 24bit, you have 144dB of digital level. Your analogue front end will have a noise floor well above that. Even the best commercial microphones have a maximum dynamic range of around 130dB; and you're not even getting that in the real world, unless you just happen to be bumping right up against it's maximum sensitivity. Not to mention preamps, inserts, and converters (which usually perform even worse). In a 24bt recording, you'd need to be peaking at -96dBFS to have 8 bits of resolution - which would leave you with 48dB of signal. Assuming the best case scenario that your hyper-quiet (130dB) mic is gain staged so that it distorts right at the same point your converter clips: at -96dBFS you're only getting 34dB of dynamic range (below which is noise). 8 bits is more than enough to capture that.

[@endokrin7897]

Peter Michael Sullivan is a fun guy at parties.

[@Md2802]

@@endokrin7897 As an audio engineer myself, I love getting into the weeds with guys like Mr. Sullivan - especially as we're getting more and more drunk. Nothing better than a good argument where there's a mutual sense of respect. And as someone who's evidently been in the industry for decades, and is clearly still passionate about audio, Mr. Sullivan has my respect.

[@forbiddenera]

22:12 1000x more power, not 1000x louder. Gotta be careful with that, but otherwise this and your sample rate videos are some of the best explanations I've seen on the subject.