Everyones favorite 32 bit float

[The Documentary Sound Guy]

7 hours ago, Wandering Ear said:

I think there are a few concepts that are getting mixed up here. Recording a 32 bit float file does not mean you using dual A2D converters. Dual A2D converters are a technique used to increase the dynamic range a converter system can measure, regardless of the processing and storage pipeline that comes after conversion. This dynamic range in conversion is actually the most important part imo. 

Thank you for correcting this.  I think I was inadvertently the cause of this misconception.  This point bears highlighting.  I didn't explain it thoroughly enough when I brought it up.

[Wandering Ear]

1 hour ago, The Documentary Sound Guy said:

Thank you for correcting this.  I think I was inadvertently the cause of this misconception.  This point bears highlighting.  I didn't explain it thoroughly enough when I brought it up.

I don’t think you were intentionally mixing the two concepts. In many conversations I have around the subject I find there is often a lack of consideration of the entire signal chain, and just a focus on this one specific element. I think the marketing materials have a lot to do with it. 
 

If I had the time available I would love to run some actual experiments measuring the dynamic range of recorders on the market so I didn’t have to rely only on specs, design theory, and some personal experience. I’m a little short on resources at the moment, but if anyone was interested in partnering with me I would love to put real world measurements on paper.  I’m sure I would learn a lot about the implementation details that don’t really show up “on paper”.   Who knows, maybe I’d even find a benefit that I don’t see in the specs, and haven’t discovered in practice yet. Until then I’ll keep relying my intuition and experience to set good gain levels and keep enough headroom to capture a great recording without clipping. 

[The Documentary Sound Guy]

5 minutes ago, Wandering Ear said:

I’m a little short on resources at the moment, but if anyone was interested in partnering with me I would love to put real world measurements on paper.

I'm interested, but equally short on resources (time especially).

 

[Roland Harris]

11 hours ago, Wandering Ear said:

The MixPre 6 touting no clipping and 32 bit float has 142dB of dynamic range (based on their specs I think they use a single 32bit converter, but have not verified this). 

 

I think you are adding to the confusion! The MixPre ii series use multiple ADCs, as Sound Devices have long acknowledged. E.g. https://www.sounddevices.com/how-is-a-32-bit-float-file-recorded/ . As I said above, some of the very cheap so-called 32-bit recorders do have a single (24-bit) ADC recording into 32-bit float, essentially to no useful effect and somewhat disingenuously.

 

Cheers,

 

Roland

[The Documentary Sound Guy]

Does a 32-bit floating point ADC actually exist?  My assumption is that it probably does not due to the inherent nonlinearity of floating point math; I think it would be quite hard to design one, which is presumably why "32-bit floating point" recorders use 24-bit integer ADCs under the hood.  I would love to know if anyone knows of or can locate an ADC chip that is *natively* 32-bit float.

[Izen Ears]

On 8/20/2025 at 11:47 PM, inspire said:

Here’s a processed version from a different recording as an example. I separated the channels, set the levels properly, and applied a bit of additional post-processing.

 

That sounds pretty good, the fact that you can hear conversation right before the huge rev. I recently recorded racecars and obviously I had to ride the trim between voice level and motor. The motors sounded identical to your recording, but it was a Deva 5.8 in 24-bit.

[Derek H]

Does a floating point ADC exist? No, not currently. I believe how Sound Devices does it with the mix pre 2 is by using multiple 32bit fixed point ADCs in tandem. 
 

The biggest advantage I see to 32bit float is using it as an alternative to a limiter. 
 

Somebody check my thinking here but If I understand it correctly a 32bit float recorder is designed so that the entire dynamic range of the analog input is digitally recorded in an optimized way. Also if I understand correctly there are different approaches to how 32BF recorders handle analog preamp gain since even though there are multiple ADCs working to optimize different signal levels there’s still just one analog preamp per channel.  I believe Sound Devices gives you physical control over the preamp gain still so that the signal can be dialed in to an approximate setting at the risk of being able to overdrive the preamp, something 32BF cannot repair. The other approach that some manufacturers take is to permanently fix preamp gain setting to ensure that a typical source (mic) cannot overdrive the preamp. In the later scenario I suppose any gain applied to the “input” is actually done digitally, post-ADC, and akin to recording an automation envelope in a DAW. It’s non-destructive and only serves to position the signal within the 32BF scale to get the signal where it needs to be for the downstream device whether that’s headphones or some other device you’re feeding. 
 

These are two very different approaches for handling preamp gain. 
 

 

[Wandering Ear]

9 hours ago, Derek H said:

 

The biggest advantage I see to 32bit float is using it as an alternative to a limiter. 

 

I like this way of distilling down to a single sentence. 

[Wandering Ear]

17 hours ago, Roland Harris said:

 

I think you are adding to the confusion! The MixPre ii series use multiple ADCs, as Sound Devices have long acknowledged. E.g. https://www.sounddevices.com/how-is-a-32-bit-float-file-recorded/ . As I said above, some of the very cheap so-called 32-bit recorders do have a single (24-bit) ADC recording into 32-bit float, essentially to no useful effect and somewhat disingenuously.

 

Cheers,

 

Roland

You are right, I forgot they pointed out they use multiple 32 bit fixed point converters. Thank you.
Barring that, which I will update, the rest of the point I make I believe still holds. The MixPre has an input dynamic range of 142dB, and Neverclip has a dynamic range of 134dB. I believe this is still the fundamental limitation of the systems we are discussing, and 32bit float does not overcome this. 
Is it an improvement?  Yes.   Is it a fundamental paradigm shift in how we approach audio recording?  Imo no.  It’s just a new tool that we can use to eek out a little more than we could do before. My experience with it has left me feeling like that improvement is incremental (and relatively small), and the marketing materials are selling it in a way that does not ring true when compared to top end professional systems out there today.  I have yet to see a technical explanation to convince me otherwise.   I’m open to being wrong, and I’d love to hear how this is providing a greater than 8dB benefit over Neverclip?

[inspire]

15 hours ago, Derek H said:

The biggest advantage I see to 32bit float is using it as an alternative to a limiter. 

In case MixPre II Series Limiters are disabled in 32-bit float mode - it just don't need it and there is no possibility to use them.
32-bit float is essentially 'clip-proof' and in most cases gives better results than 24-bit with limiter. I'm personally not experimented much with MixPre II with 24-bit with limiters.

 

I once had a recording session with a choir. We did several rehearsals beforehand, and I set my levels carefully with just a little bit of headroom. As we all know, leaving the levels too low is not ideal either, so I didn’t want to keep too much headroom. Then came the important moment – the actual performance in front of the audience. The audience started applauding enthusiastically, and the choir, inspired by the energy, suddenly sang much louder than during rehearsals. If I hadn’t been recording in 32-bit float, that recording would have been ruined – the meters were deep in the red. But thanks to 32-bit float, the performance was saved, and when I listened back carefully, there were no audible problems at all.