GOOD STUFF! REALLY ENJOYED THIS ONE. SHOUTING MY COMMENT LOUD DUE TO THE NOISE FLOOR THAT MY NEIGHBOURS 210DB VACUUM IS CURRENTLY CREATING

I’m a trumpeter, and my dog is therefore a trumpeter’s dog. He comes and sleeps in my studio when i practice. Yet at the very end of this video, during the trumpet interlude, he woke up, alarmed, and howled. You’ve got some impressive skills, congrats! ;)

Only a few problems: There are no mic pre-amps that get even close to 144dB dynamic range. So the noise floor of the pre-amp (or the active electronics in an condensor mic), seldom better than -128dB, will be noticable before you're reaching the real limits of the 24 bits. And then there are the physics of audio in free air. The maximum dynamic of free air audio is (more or less) 196dB. That's the point when the drop in air pressure gets to the point of vacuum. Anything above that will be a shock wave, not a resonance. The real value of 32 bit floating point to me is when you're manipulating and mixing the sound. It means that it's virtually impossible to saturate the mixing busses ... never mind how many channels you sum or how far you boost a certain frequency.

This is the best explanation of 32bit I've seen. Thanks!

When I was still teaching, students would often ask why I didn't use 32bit float in studio sessions. I'd always tell them I only use 32bit float for crucial recordings of live performances or interviews where there's no possibility of a retake. As some other commenters have noted, there are electrical / physics factors which may impact how effective our equipment is at reproducing huge amounts of dynamic range. For me, 32 is just a slight safety net in "one off" recording moments.

There is not really hardware DA converters that can handle electrically wise more than 20 bits because the electrical voltage difference is to small for the last 4 bits of the 24 bit. The reason we went to 24 bits is only because computers work with x times 8. So inside the box you can play around with 24 bit values or 32 or 64. But at the moment the bits has to go back into real audio voltage everything above 20 bits is almost impossible to create very accurate.

3:19 Forgive me; if there's noise noise in the quiet part of a 24-bit-fixed recording and not in the 32-bit-float it has nothing to do with the recording format per se. It's most likely because of either: a) mismatched gain settings or b) using preamps with vastly different noise characteristics at equivalent gain settings. The limiting factor is the device in your chain with the LOWEST dynamic range, not the highest. 1500db DR of a 32-bit-float file is irrelevant if your mic and preamp can deliver 'only' 130dB. If your signal is in the noise floor of either your environment or your analog signal chain then no amount of extended dynamic range above or below zero will clean it up. Perhaps the Sound Devices recorder splits the incoming singal across two or more preamps, each operating at a different gain range. If this is the case then you might realize more usable dynamic range, but in what scenarios and how often are the benefits realized? Look, by all means go out and buy a 32-bit recorder for the 0.1% chance you'll make use of its capabilities. But don't fool yourself into thinking that this is what makes or breaks a recording. Know that 99.9% of your recording needs can be served by the meager 144dB available through 24-bit fixed and a judiciously-set preamp gain.

Reaper also supports 32-bit FP files. 32-bit is amazing for recording source audio for sound design, bumblebees or tanks my Zoom F6 has my back without me having to worry about gain. :)

Reaper can also import and record in 32-bit and 64-bit FP. I think the files has to be in .wav format, but it shouldn't matter as you should be able to use .wav files in Reaper even if you are using a Mac.

I’ll give you a like for the kitty, . . . . . . . . . Oh and the info was really good as well.

FL Studio shows 32bit float as its best setting so I used that for years. Always trying to get a loud output, while checking that against Loudness Penalty to make sure I fit here on YT. I finally added a SSL Comp to the mix Buss, for volume reduction, and started pushing up the mix channels into it. Was shocked to see my files recorded at or above clip would still pull down as you had shown, the real benefit to 32 bit float shows up in that respect. But of course, I’m going to extremes as I learn where I need to be. Too hot into the SSL, which will pull it down to fit YT. I used to think I needed limiters to keep the recorded files from clipping, actually that info is still there in the WAV file because of 32 bit float. This information is hard to find, but you’ve explained it nicely. Thank you.

Thank you sooo much for your clear explanations about post processing. Everybody says that "you can fix it in post" but nobody shows how !!! You didn't either BUT you mentioned the DAWs that can handle it 😍 I'm wondering if the list has expanded one year after your video ?!

Just subscribed! I'm a "Fake Filmmaker" and after watching several videos from your channel this week, and not subscribing, I watched this one; All I have to say is THANK YOU, THANK YOU for saying something that makes sense about 32-bit float...I got beat up by a whole bunch of audio professionals (in a Facebook group) after saying how excited I am about 32-bit float (although about half of them agreed with me about it being a "game changer" for "location" recording.)

What an amazing explanation and demonstration. I finally feel I understand the differences and their benefits, thanks to you guys. What a great contribution to the audio community. May Santa leave a garbage bag full of money under your tree next Christmas.

That's great! Second video I watched from this channel, and learned something new here as well. You, good sir, have my subscription.

Personally I do 44.1/24 for recording (with modern plugins and internal super sampling 44.1 is enough and it saves a resampling round), 32bit FP If I am exporting stems or a final mix to be sent to someone else for mastering. Final renders are always 44.1/16bit for me (I still do CDs, people seem to be more willing to pay up for something they can hold, going by the fact that most people listen to my stuff on KZfaq or pirating them as low quality mp3s via some sort of downloader). Beautiful explanation as always!!! P.S.: Still waiting for the Flopcat ASMR album. P.S. 2: Reaper does 32bit FP, too!

I played trumpet in high school and I’m pretty sure I played that song. Regardless of the sheer music, I’m fact. Great explanation of bit depth, thank you.

great vid! You missed Cubase Pro, it can record in 32-bit floating point, and actually supports 64-bit float as well of course not many plug-ins support that bit-depth yet

Awesome video and brilliantly explained. This is why new loop pedals are coming out with 32-bit floating (Boss RC-600 for example). Fantastic as it means less degradation for overdubs and when recording and using as an audio interface, your doing to get great quality!

Very realistic and intelligent presentation - quite well done.

Thanks for the great demo lecture on 32 bit float. Anxious to hear & see your review on the Mix Pre 6ii recorder in the near future.

thank you. great explaination. and a cute cat. : )

Great video James, certainly like the extra headroom my mixpre 3-2 allows!

It's also good for cranking your preamp to get some analog breakup, then just bring the level down after recorded.

Hey James, thanks for the vast amount of time, to pull down my headphones, before shouting into my ears :D. I couldn't hear the rest of the video, but I'm sure it was great. Kidding, of course. Cubase btw, asks the same question. While the default export setting is 32bit in Cubase 11 pro, the recording and import limit seems to be 24bit. Internally Cubase can even work at 64bit float. Maybe a perfomance issue, when dealing with countless tracks and plugins. Dunno.

Good video, thank you for the clear explanation about the 32b FloPping cat standard

Wow what a wellmade video. Easy to understand and with great examples!

Steinberg's latest offerings support 32 bit 'integer' hardware encoding (UR-C Series) and in software (Cubase - Nuendo) support up to 64 bit floating point precision.

Its really funny how you consistently get Flopcat fitting to the topics in your videos :D

32 flops per cat you say?

Instinctively I always did this but not I see why it's been such a lifesaver! 32 FTW

I’m a bit confused, don’t the preamps have way way more impact on the noise flore then the converters?

Nice Vid. I’m on 48/24 bit and am very happy there. Very informative and well explained. Thank you. Stay safe n well. 👍

you guys are both informative and funny, well done!

Auditon and Premiere both can work with files from disk that are 32bit float. Audition can work with 32 int and float. Both (as with any modern DAW/NLE) with use a 32bit float internal resolution for any mixing or audio manipulation. This allows you to, essentially, never clip the internal summing architecture of the software. You only need to truly worry about the the output gain when outputting to a file to disk or to the audio device (the meters in most DAWs show the integer value of the audio output in dBFS). Once you write a file to disk at 16 or 24 bit, or try to play back audio through a 16 or 24 bit DAC, you can easily clip. Essentially, exceed the available bits for 16 or 24 bit. One interesting thing that Audition, and to a lesser extent Premiere, does is to work with temporary files at 32 bit float resolution. For certain formats, on import, the files will be converted to 32 float in a temporary cache system. Any manipulation of the audio data will be stored in the 32 float temp file system, and only be rendered to disk on a save operation. This allows unlimited undo and unlimited gain manipulation until the audio is rendered to a file on disk as a last step. Additionally, Audition can record 32 float format to disk if the recording device supports it.

Gabba Industrial Deathmetal can only be measured on the Richter Scale.

32 bit sounds ideal for classical or very intimate settings - maximum dynamic range!

thanks for this

Thank you, I nearly understood some of that.

The shouting made me spit out me brew. Cheers lads, entertaining as always.

Thank you so much guys, this video exceeds 32-Bit of PRO level!!

But 99% audio interface's AD converter is 24-bit, And whatever the 24 or 32 bit been used, the clipping occurred before the digital part(when the input signal voltage is too high to the AD circuit), and also the signal noise ratio can't be improved by it(SNR is input signal level minus the AD part self-noise level) I haven't read much information about 32bit AD, if i wrong please correct me. Moreover Akash Murthy made wonderful videos about fix&floating bit explanation, I'm not sponsored but they're great.

What an interesting video and very informative. I must admit, the whispering bit gave me ASMR tingles. You should do an ASMR channel. Very good. 😂😊

There is 1 problem. There is no such thing as a 32 bit floating point ADC. One issue is the thermal noise floor, you could make a 32 bit codec but around 14 bits would just be noise. A very good 24 bit ADC has a 100dB signal to noise ratio. Cool it with liquid nitrogen for better performance, better yet put it on the cold side of James Webb. I guess you could put some sort of analog auto gain control on the front end, but that would need to run at above a 10mS cadence, not 48KHz. edit: You could try using companding with all its issues. Put an 8:1 dynamics compression on the input and 8:1 expansion on the output (or code equivalent), but I do not see a way around the second weakest link in an audio chain, the ADC.

"This is the way." -James, to his mirror while brushing his teeth

I checked that MixPre-6 II specs. It has unprecedented AD conversion at -142dB (almost the whole range of 24bit audio)! Impressive! Ableton Live records in 32bit, but I always thought (and still think) that it is impossible to go beyond the converters dynamics. RME Fireface 800 has only -112dB and I guess difference between 24bit and 32bit audio recording won’t that spectacular as in your video. But I will test it...

Thank you for the information. This is excellent news for recording live bands. Rarely do they play at the volumes set in sound check. Usually by the time the first song is over there are clip indicators on all channels.

You‘re great guys😅😂💪🏻💪🏻…love your cat content made with serious intresting topics

I was recording 32bit at 96k into Sonar (now Cakewalk) back in 2009, ummm, on a Windows system with a total of 4GB of ram and a HT chip. Of course, that meant, in my then unique situation, I could only record 8 tracks at a time before dropouts and other random glitches started. If I had a better system, today, I would never again bother with 24 anything ''in'', only 32. That would be a nice SSD and a 4Ghz+ 8 core with 32GB of ram and massive outboard A/D but I don't see that happening in the near future, so, whatever. I've dropped back to 24/48 for everything except for archival works that I can't foresee being worked with for several years to come (ie ideas that have hit brick walls, children's performances, family historical stuffs, etc). I will say, however, there is something really sweet about 32/96 - it just seems so smooth to listen back to, can't really quantify the quality but you KNOW when you switch from that down to 24/48 in a BIG way. All these folks who say ''There's no detectable audible difference above 24/48'' are liars, period. Everything about 32/96 sounds better than 24/48 but it's just too much of a resource hog to be worth it for most commercial and/or non-critical applications. If I were running a pro recording studio, nothing would be tracked at less than 32/96 because mixing with it is a dream. You can get away with a lot more bounces and sub mixes; things that at 24/48 start to feel rough around the 3rd time they've been sent through any kind of processing just keep on going and going and going while acquiring little to no artifacts that would otherwise be very obvious at 24/48 or below. I still double process things in a 24/48 mix but I really try my best to avoid EQing or including within any 3rd pass/layer anything that will get processed a 3rd time. I don't mind a bit of phase or even sample level ringing but bit level anomalies from multi-passed resampling can give you headaches that even a kernel engineer can't figure out how they got there, how to get rid of them or even how to mask them.

Reminds me of that time I was on the receiving end of a project that a composer who had zero clue of how to use 32bit float sent. The whole thing was a nightmare. Dude was mixing everything with KRK's backed to a wall with another concrete wall about 4.5 inches behind his back (all of that in a rectangular room, no idea why he used the width and not the length of the room), the room modes were so insane that he'd mix the sub basses at around - 36db so they'd sound good to him and had to put the vocals and snares over 0 so they'd cut through. That'd have been fine had he sent the 32 float session but he insisted on converting the project into 24bit stems before sending them because "He has this really amazing soundcard that made things sound much better when converted ( a Berhinger MDX), didn't want us to mess the mix and also...his wifi was too slow to send full-on 32 float sessions) he never checked his clip gains before converting so everything was either really quiet or clipping like a mf. Since the project was entirely made of 3 free vst synths, one kontakt drum, and one vocal take I suggested he had every track hitting around - 6 on average before convertion and we'd redo the balance over at my studio. Once I had tracks with the legal acceptable quality required to get started it took about 13 revisions for him to validate the mix because they didn't sound good on his "state-of-the-art system" while the director of the movie went with the very first mix because the revisions sounded more and more erratic. When all of it was done, he sent this huge email saying everything would have been fine had we used the 32 float mixdown in the movie. Had to explain none of the equipment in the whole production could or needed to work above 24 for our purpose. if this is a cautionary tale, I'm guessing the moral is : More bits isn't better if you have zero clue of what the extra bits are doing. Great explanation in your vid btw

Great video (as per your normal standard) Well explained technical information plus Flop Cat and Marks trombone farting!

I'ma need that 512-Bit float for my cosmic quantum shifting drummer, average snare hit is at 9z+(600x572y)e/14-Db so 32 just isn't gonna cut it fam. xD

Best audio-related channel!!! Always good quality information and in a super funny way.

Super interesting video and topic. I'm ready for 32bit! But my DAW won't support it, so I'll record my rescue cat with 2 channels and and put a very transparent compressor on the higher gain channel is case my cat does a dog barking impersonation... Great job...as usual (and great info). Phil NYC Area

U guys are hilarious.

Most of these talkers only talk about the circumstances and repeat the marketing slurry, without telling a bit more consciously how such a 32-bit float recorder works. The matter is quite vague described in the Sound Devices patent US9654134B2 (although it can be done more easily - with only two sets of preamp-ADCs, like in Zoom F6 / F3 recorders, etc.). In a nutshell: you don't need preamplifiers with dynamics greater than 128dB at all. The trick is to use several sets of preamp & ADC, where each preamplifier has a different gain, and then quickly switch the streams to the loudest undistorted one in the digital domain, so the "regulator of the gain" is indeed a microprocessor. And the 32-bit float format makes it even easier to store such informations (as real digital data stream + gain).

Fl studio and ableton run 32 bit files as well

Cubase Pro supports 32bit integer recording and 64bit float processing.

Bravo🙏👍👍

Teacher: So, what did you learn from this video? Me (an american who has been taught that 65db is the level of a vacuum cleaner): ...british vacuums must be loud and shitty

Great video 👍 Looking forward to your upcoming SoundDevices MixPre-6 review. Please let us know if Steinberg Nuendo can import 32 bit floating in MacOS.

Btw, it doesn't have to be floating point. As long as it is 32 bit, integer or floating point will work mostly the same for most uses (integer technically being superior). Cubase is fully 32 bit compatible, btw.

Think about it: The first element after the mic in the signalchain will be the preamp, it's dynamic range comes nowhere near the range of 32bit digital float audio. So the preamp will allways be the bottleneck in this theory. I understand that in 32bit float the gainstaging will be handled for you. But in extreme cases the analogue circuit will still overload long before the digital will.

Hey guys. Love the channel👍 I am a drummer of quite some years and have just in the last few years gotten the idea to start recording on my own. I've bought quite a few things to record in 32 bit. My thinking is that I can capture and use all the small details that have previously been left out with gain levels and compression etc. Is this thinking wrong? Thinking of course of ghost notes and the fiddly bits. Maybe the filddly bits are all meant for the jazzy parts?

Great Video! Unrelated question: Will there be a HEDD monitor review at any point? Cheers, Ben

Isn't the more interesting question: What dynamic range has the analog mic pre section of the 32BitFP unit? I remember that they have stacked ADCs which gives you more DNR, but has inferior specs regarding THD because of the two stages switching.

There’s one military airport in my, otherwise quiet, area. Still, I might look suspicious there with any recorder, not to,mention a 32 bit one.

Love the video ( as I always do) BUT @03:56 when your shouting, you had perfectly good levels going into your mix pre6 ( it didn't blink orange or red once) so were you clipping the signal in post? for a proper test why didn't you saturate the circuit in the mix pre and clip the "F" out of it so people could really hear what it sounds like. Just asking .....

Very useful info, once again! I have a couple of questions, though: 1- This might sound stupid but... can you record a 32-bit floating-point file with a 24bit ADC converter audio interface? When using an audio interface with built-in preamps, what are you really controlling with the "gain" knob? The analog signal gain (clipping chance 1), the gain feeding the converters (clipping chance 2) , or both at the same time in a 1 to 1 ratio (both clipping at 0dBs)? Anyway, I´m confused. Do you need a 32 bit ADC to achieve the "impossible to clip" dream when recording? Wouldn't the analog preamp clip anyway? How does it work in a field recorder that can record 32-bits?? infinite headroom analog preamps + 32-bit ADC? 2- I take the 32-bit floating option is also good for grouping/summing tracks into buses when mixing. Does that mean we can gain stage at the start of a mix session, and never worry about clipping in groups or master faders again? What about plugins? Do they need to also work at 32-bit floating to maintain the dynamic range of the signal that's passing through? 3- Are there any benefits in keeping everything at 32-bits until the last step of the mastering stage (dithering/noise shaping)? Just like editing a video with 4K video files: A 1080p export still looks better than when using 1080p files directly. I hope someone around here finds these questions useful. You're killing it in this channel!!! I would love to see a Dan Worrall - PDP crossover!

A bit surprising that Cubase or several other popular 32bit supporting DAWs weren’t mentioned in your list, especially when Cubase supports 64bit which wasn’t covered at all in this video. Maybe an updated video in the future could expand on this topic a little.

As I’ve got a Mix pre 10mk2, Im interested to hear your review/what you think

👍

pro tools also can use 32bit floating point

32 bit floating Cat 😖

That seems cool and all, but... Isn't there information missing in this video about the recording process before A/D conversion? Like, how about A/D converter clipping? Electronics noise and clipping? I mean... If it was that easy and miraculous.... Why is there still gain pots around at all? 🤔

So, my ex speaks at around 140dB, since it's painful to listen to her for any amount of time.

Isn't 16-bit PCM supposed to have a "sign" bit to dictate the polarity of the given pulse, and the rest of 15 digits to encode the magnitude or amplitude of that pulse?

Awesome as always! 😂😂😂 Can you adopt me please? I’m a simple guy, 8 bit are enough 😅

Interesting, I've been wondering about whether you actually ever could get to a state where you just forget the input gain with 32bit..... As well as there not being many DAWs that seem to support this for recording at the moment (I wasn't aware of that*), I wonder if there's really hardware that can safely/accurately cover this enormous range....though I see that little mic seemed to cope with Flopcat's 7,000,000dB purr so maybe that's been answered. Another excellent video, cheers. *Cubase seems to suggest it can import/record 32bit but suggests running in 64bit fp when doing so....so not sure.

So when starting a new session in Pro Tools would it just be worth doing EVERYTHING in 32Bit float? It sounds like a pure win - win in terms of quality and dynamic headroom. Or will the performance cost be enough to make going above 24bit a problem? Being able to only have to focus on not clipping my pre-amps on the way in would avoid a huge amount of stress, especially when dealing with drums.

Cubase does 64 bit float. Good luck using some plugins.

And the question is: Do you use a Zoom F8N Pro or some other recorder? Thank you What about microphone choices for interviews? Thank you And sound software? Thanks and Regards

Your example is A1 3:40

what do you even need preamps for in a decice like this?

Put a 16-bit mic too :)

Will you be auditioning a Zoom F8n to compare it against your future review of the Sound Devices unit ?

QUICK MATHHHH

I've always bounce my tracks out in 32 bit float knowing the first time I bounce it it's gonna be a brick wall.

Hmm, this needs another take. I get how 32 bit processing is beneficial, but there's no way the signal at the start has anywhere close to that resolution/headroom. There must be some AGC or other trickery before/in the ADC to avoid clipping and noise. 24 bit recording itself isn't effectively 24 bit, that's the theoretical digital limit without taking SNR etc into account.

Great and funny video but a bit lacking on the other relevant dynamic ranges, as others have pointed out. I'm intruiged by how the two ADC signals get precisely overlayed in such a 32-bit recorder

I feel very sorry for the person who measured the sound pressure level from an atom bomb. 1 meter A weighted right?

@1:40 you should have added the extra 8 bits to the left - not the right!

Actually 16 and 24 bit are integer files, not fixed point, since there is no decimal point at all. I mean I guess you could argue the point is fixed at the location after the lowest number, and we capture nothing after the point, but that’s not how the term fixed point is used in computing. (Which is a ridiculous nit-pick out of an entire video packed with great information.)

If I'm not wrong, you forgot to mention one tiny little detail: you'd need an interface that supports 32-bit recording. Not too many available.

Very interesting topic! Can't wait for the hi-fi fans to start swearing 32 bit sounds better and "more precise". 😂 I can totally see how floating point audio can be extremely useful in field recording. However, it might have prevented some confusion if you mentioned that the gain controls are deactivated in 32 bit mode. It needs to be underlined that (usually) 32 bit audio it's not just 24 bits + 8 additional bits, but as you mentioned in the video it's a very different way of storing numbers. It could have also been a 24 bit floating point format. It's mostly the format that matters, not the number of bits. But then again, 32 bit is already the standard for floating point numbers. What I can't really wrap my head around is how they managed to implement a converter that reaches this level of performance. RME's high-end converters are stated to have a SNR of around 120 dB and anything beyond that will run into problems regarding self noise of components of the input circuitry. So, there's no way to even make use of regular 24 bits using a traditional implementation. I guess they are using multiple converter chips with a different gain each and then assemble the floating point number digitally - but I could be totally wrong.

You seem to have missed Cubase off your list of 32bit float capable DAWs - actually it can capture and manipulate 64bit float too. Even Steinberg's humble UR22c interface captures 32 bit audio at 192kHz.

Here we go with more fuel on the 32-bit-float vs 24-bit-fixed fire.

My excessively loud and large production dumps have typically way more than a mere 32 floating bits. Data ‘is’ stored... or ‘are’ stored..? Hhmmmmmmm.

My question is: if 32 bit float has 1500 dB of dynamic range, why not offer a different version of 32 bit float where instead of each division of loudness being twice as loud as the last one, maybe chop it into smaller divisions like 10% louder? You’d still have 150 dB of range, but with finer, more accurate sampling.

Just to be clear... 32 bit floating point recording works ONLY with 32 bit AD converters right? Recording with a 24 bit AD converter in a DAW configured at 32 bit FP will not accomplish this... am I right?