A bit of this; a bit of that…

[Jason Nicholas]

I was at a café earlier today sat next to two young guys talking about camera gear. One of them had his shiny new Sony video setup. He kept going on about how 32bit Float would basically allow him to never have to really consider what he’s recording anymore. He told the other fellow that it used to be 24bit but that was an ancient standard…like from back in the 90’s. I never thought I’d feel dated by my bitrate.

[JustusWirth]

Until he delivers his 32 bit files to post, they begin to wonder why everything is distorted in their export and then come whining to some sound engineer who would have done a better job recording in the first place. 

[The Documentary Sound Guy]

"Never have to really consider what he's recording" is definitely a sign of a skilled cameraperson.

[borjam]

Douglas Adams described it as “covering it with a SEP field”. A brilliant way to make problems disappear.

[The Documentary Sound Guy]

...and that is why we all still get hired.

[balanceistisches]

On 5/15/2026 at 6:58 AM, Jason Nicholas said:

I was at a café earlier today sat next to two young guys talking about camera gear. One of them had his shiny new Sony video setup. He kept going on about how 32bit Float would basically allow him to never have to really consider what he’s recording anymore. He told the other fellow that it used to be 24bit but that was an ancient standard…like from back in the 90’s. I never thought I’d feel dated by my bitrate.

This is truly an interesting conversation. The modern technology of 32-bit float audio offers many advantages, particularly when it comes to professional audio recording and editing.

Let's take a brief overview.

24-bit audio is of very high quality and has been used in professional equipment for many years. It offers a wide dynamic range. This means that very quiet and very loud sounds can be recorded with high fidelity. 24-bit was the standard for a long time and, technically speaking, is not obsolete. Many professional recordings, films, and music productions still utilize 24-bit audio today.

[CS audio]

been wondering on this topic for a while

[inspire]

I think the truth is somewhere in the middle.

Personally, 32-bit float has become my preferred option in many situations — but definitely not because I suddenly stopped caring about proper recording practices. I still pay attention to exactly the same things I would with 24-bit recordings: mic placement, RF, preamps, gain structure, environment, all of it.

What 32-bit float gives me is mainly peace of mind. I simply don’t have to worry about headroom quite as much.

And to be fair, it has saved my recordings multiple times already — especially in loud and unpredictable situations where things can change very quickly.

But I also agree that some people talk about it almost like the laws of physics no longer apply.

[Bouke]

Meh.
I would think that if you properly gain stage 24 bits is even overkill.
Only reason that I can see for 32 bit is that you can record as hot as broadcast / finishing levels. That might be wanted if your do fast turnaround work.
(the 8 bits on top of 24 to get to 32 bits only kick in where 24 would have hit zero / ceiling, so there is 'hidden' headroom.)
I've been an editor all my life, most NLE's don't give you much gain without jumping hoops, and waveforms don't draw nicely if you record conservative. So a 'hot' recording is nice to have, saves tons of time.

But from a pure technical / quality perspective, I don't see the need for 32.

 

[Derek H]

From a post production perspective I’m sure 32 bit float seems unnecessary but it’s very different when you’re recording in the field. Bouke, have you ever had to record someone who whispers one line and then screams the next? Without warning? Or mic up 4 people plus plant mics for a free-driving scene where you have to leave the recorder in the trunk, press record and walk away without a rehearsal? There are many excellent use-cases for 32bit float and there will be more and more going forward. 
 

I have a feeling 32 bit float will become standard on the acquisition side of production and move beyond its current position as a gimmicky marketing feature for prosumer gear. Is just a matter of how soon. Post production software will hopefully catch up. If we’re sticking with 24bit in cases where 32 would be advantageous just because that’s what the Avid wants then I think we need to start pushing back.

 

 What already seems to be happening is experienced sound recordists are choosing when to use 32-bit float when it’s helpful in a particular situation and then doing additional conforming work after filming in order to hand over 24bit files to post that are immediately usable. Hopefully that will be a temporary workflow as post gets used to dealing with 32-bit float. 
 

Also, I’d say there is a low-level advantage with 32-bit float recorders as well and it’s not all about overload. Most recorders that have 32bf also have dual ADC inputs and used in conjunction with 32bf results in eliminating the digitization noise floor so that the only bottleneck is your microphone and preamp quality. Feel free to fact check me on that one and it’s maybe more a function of the dual ADC design but 32bf allows you to capture the entire dynamic range of a dual-ADC input. 

[The Documentary Sound Guy]

Most of our pro recorders have had dual ADC inputs for years ... technically speaking, we could have had the 'advantage' of noisefloor ages ago if we'd been willing to adapt our recording and gain staging to allow more headroom when we switched to 24-bit.  Quantization noise has never been the limiting factor; it's always been the electronic noise floor that we've been gain staging around (assuming properly engineered equipment at any rate).  Or, put another way, the analogue dynamic range of our equipment has never surpassed the dynamic range of 24-bit, even on previous dual ADC designs.

The only 'advantage' of 32-bit float is that it's *forcing* us to abandon the convention that we always record at -20dBFS because dBFS has no meaning for 32-bit float.  We still have to record within the proper limitations of the analogue dynamic range available to us, and that amount of dynamic range has been adequate to accommodate 'gainless' designs for quite some time.  We just never updated the gain staging techniques we learned to accommodate gear from the analogue era when we were trying to squeeze our recordings into 60-80dB of dynamic range.

Do you doubt me?  Consider that a whisper to a scream probably ranges at most about 80dB (a VERY quiet whisper is 20dBA, a VERY loud scream is ~100dBA).  Consider that, for most of my professional career, recorders with 120+ dB of dynamic range has been more or less standard.  So, if the scream is intended to be recorded at -20dBFS, the whisper lies at -100dBFS, and the noise floor of the recorder / preamp is 20dB below that.  And, the quantization floor for 24-bit is a further 24dB below that.  You *can* hear the pre-amp noise floor if it is only 20dB below the whisper, but you definitely won't be hearing the quantization noise below the preamp unless you are gaining the whisper WAY up in the mix.

BTW, I suppose I'm leaving out the dynamic range of the pre-amp, which is often the true limiting factor.  My point is that a design without variable pre-amps à la 32-bit float was certainly possible in 24-bit, and with good enough pre-amps on a 24-bit recorder, it's very possible to gain-stage much, much lower than convention dictates and achieve similar results to 32-bit.

[inspire]

The fact is that I recorded a choir concert (Live), where rehearsals were sung in one way, and the actual concert was much louder (in the open air). Yes, it can always be corrected, but 32-bit often saves the day when everything is in place but turns out differently.

At 32 bits, you don't need a limiter - it's simply not necessary.

[Bouke]

19 hours ago, Derek H said:

From a post production perspective I’m sure 32 bit float seems unnecessary

It depends. Of course I know what 'can' go wrong in the field. I've seen and heard close to all possible issues that 'can' occur, in my 30 years of working in the dirty trenches of post, with close to any kind of jobs...
But my remarks were about keeping headroom vs taking risk. Or, as I did on my occasional shooting (I did a lot of multicam work, just me directing / switching / operating PTZ and sound recording), I doubled the signal and set one channel some 15 dB lower than 'normal' for the occasional peaks.

For sound, I'm not into it as deep as you guys, but I know for sure that you can 'never' expect the one after you being 'logic' about your workflow. 
You all know that on some jobs post never goes back to the originals. I once have sent a picture lock to grading, where there were some dissolves. Grading demanded VERY LOUD that ALL CORRECTIONS NEEDED TO BE REMOVED.
Now these dissolves were between shots where some had a truly horrible white balance, so I added the shots with handles after the show to get the same grading.
That was told to the director who attended the grading, was put in the slate image in front of the show, was printed on the cover of the tape and mailed to the facility.
And you can guess what I got back... (Luckily, otherwise it was aired.)
This was a high end production, the most expensive grading facility in the country.

For the rest, I think TDSG is correct.

 

[Derek H]

TDSG, (I like the abbreviation btw). I love your post and agree that if it were acceptable to adapt our gain staging approach to take full advantage of 24 bit’s dynamic range then we’d be fine. However, and sticking with the whisper/scream example, if I turn in a scene that is gained correctly for the scream and let the whisper live down at -100dBFS post will email me and ask where is the sound. Or they won’t ask and they’ll just ADR it.  I know this to be true because I’ve experienced it. I recorded a scene where an actress portraying a mentally ill patient starts out in a barely audible breathy whisper and by the end of the scene is having a full-blown maximum volume meltdown, kicking and screaming. I rode the levels best I could but I knew what was coming so played it pretty conservative. Got an email back asking why levels were so low… I swear post would rather fix distorted peaks than deal with low levels. 

[The Documentary Sound Guy]

Yup.  The advantage of 32-bit is social, not technical.  It's making it acceptable to deliver ungained audio, and post is learning to normalize what they receive.

To be fair, a scene like that definitely *does* need mixing.

[Elgotron]

I'm directing a documentary for 2 and half years. I have to manage image AND sound. And I work in a environment of art brut workhop with persons with disabilities which means that anyone can scream or cry hyper loud at any moment even when I'm shooting a very quiet conversation... Well, I have to say that , in my case, very low budget movie, sound and image management in the same time, totally impredictible environment.  32bits saved my life : )

[Derek H]

Here's some very recent field experience to add to this discussion. I just finished recording a few documentary scenes with people on canoes singing very loudly. Used 32bit float pocket recorders (Tascam DR10) on multiple people and it worked mostly great EXCEPT one guy who was incredibly loud managed to distort his recording while singing. This happened despite the record level set to the lowest setting. So it proves that 32 bit will not always be the savior it's made out to be. You have to abide by the physical limits of your microphone capsule and preamp stage.  I'm still not sure if in my case I hit the SPL limit of the stock Tascam lav or if I hit the input limit of the DR10. TBD.

[The Documentary Sound Guy]

First nations singing & drumming is the loudest thing I've ever recorded.  I did manage a clean recording on a COS-11D with my Lectro gain at 1 ... I don't *think* it hit the limiter at that level, but it was close.

[inspire]

I had a similar real-world experience while recording drag racing. Some of the cars were much louder than expected and later in post I had to pull the gain down by around 40 dB on certain recordings. Honestly, at first I thought those takes would be unusable.

But surprisingly, the recordings still held up very well after reducing the level that much. In my case it really showed me that 32-bit float can provide an enormous safety margin — as long as the microphone capsule and the analogue stages themselves are not being overloaded.

To me, that seems to be the key difference: if the problem is only digital level, 32-bit float can be amazing. But if the microphone or preamp hits its physical limits, no bit depth can really save that anymore.

[Roland Harris]

On 5/22/2026 at 11:53 AM, Bouke said:

Meh.
I would think that if you properly gain stage 24 bits is even overkill.
Only reason that I can see for 32 bit is that you can record as hot as broadcast / finishing levels. That might be wanted if your do fast turnaround work.
(the 8 bits on top of 24 to get to 32 bits only kick in where 24 would have hit zero / ceiling, so there is 'hidden' headroom.)
I've been an editor all my life, most NLE's don't give you much gain without jumping hoops, and waveforms don't draw nicely if you record conservative. So a 'hot' recording is nice to have, saves tons of time.

But from a pure technical / quality perspective, I don't see the need for 32.

 

 

Exactly: spot on.

 

Having got a bit frustrated with the lack of practical tests comparing so-called 'dual-ADC 32-bit float' recorders to 24-bit recorders, I bought one last year to test (the Tascam FR-AV2: modest, but far enough up the food chain to have decent preamps and, unlike some of the cheaper '32-bit' recorders, to actually have dual ADCs) to test along one of my 788Ts and a MixPre-3 (1st gen, 24-bit). The latter, of course, has a very different topology to the 788T and, although often overlooked or misunderstood, this means that - even though 24-bit - gain setting is irrelevant in terms of recording: recording the quietest sound with minimum gain (6dB) and adding 70dB in post sounds identical to recording with max gain (76dB) in terms of self-noise and dynamic range. Of course, doing this with the 788T sees self-noise increase as you drop below 20dB gain. So whispers to jet planes (assuming mics and inputs are not overloaded) can be recorded equally well with the 24-bit MixPre recorders as with the later 32-bit models (and I assume, given the similar dual ADCs per channel, in the 8-series in their (original) 24-bit mode). Where the dual-ADC 32-bit float recorders do have the advantage for extreme and unpredictable dynamic range over 24-bit recorders of the MixPre etc. topology set at minimum or very low gain for extreme headroom, isn't so much in the monitoring (since the headphone output of the dual-ADC 32-bit float recorder will clip if you monitor for a whisper and the sound transitions to a scream) but, rather ironically, in the delivery since, of course, a recording of a predominantly quiet sound can be delivered at a decent level and the level dropped in post for the loud sounds (that seemingly look clipped). This doesn't really add to what Bouke said, other than pointing out that not all 24-bit recorders are the same, and, of course, my blog-post on the tests includes downloadable samples if anyone wants to listen to the comparisons: https://drbadphil.com/?s=fr-av2

 

Cheers,

 

Roland

[inspire]

Reading these comments made me reconsider one of my own experiences recording drag racing. I previously mentioned that I had to reduce some recordings by around 40 dB in post and was surprised how well they still held up afterwards.

Originally I assumed this was purely a “32-bit float saved the day” situation. But after following this discussion, I’m now wondering if a well-designed 24-bit recorder might actually have survived that situation too — assuming the microphone capsule, analogue stages and ADC were never truly overloaded in the first place.

I have to admit I haven’t really tested this properly side-by-side, because for extremely loud and unpredictable situations I usually choose 32-bit float recorders for the added safety margin and peace of mind.

Still, I think this discussion has been very educational in showing that the real limits are often more about the analogue side and recorder design than simply “24-bit vs 32-bit”.

[Jason Nightall]

I

On 5/15/2026 at 5:58 AM, Jason Nicholas said:

He told the other fellow that it used to be 24bit but that was an ancient standard…like from back in the 90’s. I never thought I’d feel dated by my bitrate.

 

Netflix delivery specifications require audio masters to be 24bit. If 32 bit files were submitted they would be rejected.

 

 

[Bouke]

6 hours ago, Jason Nightall said:

Netflix delivery specifications

Now, does Netflix already support an open source tool for their 'strange' loudness specifications?

 

[Johnny Karlsson]

"48k/24bit sample rate and bit depth for Original Language Mix or M&E Mix - applies to Stems and Mix Masters."

 

I could be wrong, but to me, this looks like it's specs for the final master mixes delivered from post(?)