Should I be recording at 32 bit float?

[zeronyne]

So there used to be this odd camera…I think it was called a Lyko. It’s value proposition was that it captured so much information that you could decide the focal point and depth of field AFTER shooting.

 

So, for the sake of preserving dynamic range for future gain boost with minimal noise, should we be recording at 32 bit float?

 

This YouTuber thinks so.

 

And it’s not an academic Massenburg-style application of the fidelity (and much respect to George…I’m using his name to refer to his polarizing posts). The practical application is that you can boost very quiet signals without the typical issues.

 

Believe me, the issues with any music I’ve recorded at 16 or 24 are not related to bit rate.  Just curious as to your opinions. Again, I’m not debating perception or “better”ness. I’m just asking about the signal range advantage…is it worth the file size increase and the upgrade in interface?

[Philbo King]

It might be useful for specific applications like recording very wide dynamic range events (such as whispered dialog followed by artillery in a war movie, or seismic earthquake monitoring).  For music or audiobook recording it is probably overkill IMHO.

[Bill Heins]

I work at 96khz 64bit float in Nuendo....using float will help preserve more of the individual digits that make up an audio file during certain processing, especially important in mastering.

[Anderton]

8 hours ago, zeronyne said:

So, for the sake of preserving dynamic range for future gain boost with minimal noise, should we be recording at 32 bit float?

 

The video is not about recording at 32-bit float, it's about manipulating 32-bit float with two different gain channels, then figuring out which one to use. It's the digital equivalent of the analog hand-held recorders from Sony and Tascam that recorded two audio streams, with a secondary stream 20 dB below the main stream. Then if there was distortion on the main stream, you could take the undistorted sound from the -20 dB stream and edit it into the main stream. In that sense, you didn't have to worry that much about gain because if you weren't conservative enough, you could rely on the secondary stream. Of course, you had to match levels. I presume the main advantage of using digital is it can stitch the pieces together automatically.

 

So it's not about improving fidelity (unless it prevents distortion or excessive noise you would have gotten from sloppy level-setting). It's about making it easier to record, without having to be concerned about level-setting.

 

The bit depth used in recording is only as good as your A/D converter. There is no A/D converter that gives 24 bits of useable dynamic range, it's more like 20 bits (120 dB dynamic range) on a good day. Recording with 24-bit resolution is all you need until A/D converters figure out a way to keep electrons from crashing into each other and creating a noise floor (and engineers create perfect power supplies and circuit board layouts).

 

The bit depth used in the DAW's audio engine is what prevents rounding-off errors during processing. This has nothing to do with recording. It's about not modifying what was recorded. There's little practical difference between 32-bit floating point, 64-bit, and 64-bit floating point, other than the storage requirements. Consider this:

 

The dynamic range of 16-bit audio is 96 dB.

The dynamic range of 24-bit audio is 144 dB.

The dynamic range of 32-bit floating point audio is 1,680 dB.

 

Let's suppose some software engineers are so sloppy and incompetent that the 32-bit audio engine screws up constantly and introduces roundoff error-based distortion in the bottom 200 dB of dynamic range. Of course, they'll be fired, but your audio engine still has an effective dynamic range of 1,480 dB. That's more than enough to accommodate a near-infinite amount of  processors and calculations.

 

So, we can draw three conclusions:

 

• Given current technology, you will not obtain better fidelity by recording at more than 24 bits.
• It's possible to cobble together a system with dual conversion sections that greatly simpilifies setting lelvels and helps avoid distortion. There's nothing special about doing this with 32-bit converters, it's the process that matters. You could use 24-bit converters and still get pretty much the same result, in practical terms.
• You want your DAW's audio engine to have the highest possible bit depth. There's no measurable advantage of using more than 32-bit floating point in an audio engine. I believe in a lot of cases, audio engines use 64-bit resolution because hey, why not? Computers are all 64 bits these days.

 

And of course, there's always the caveat of believe YouTube at your own peril. Remember, there are a bunch of people who insist the earth is flat, and people haven't landed on the moon

 

 

[ElmerJFudd]

It provides a lot of dynamic range, which is great for acoustic music, classical music, jazz and other intimately performed styles.  I don’t think you need it for most popular styles. But if you have it why not.  If you take the time and use nice mics and pres 16bit or 24bit/48k does the job.  You’re going to dither down for video anyway.   And that’s going to get compressed for YouTube anyway.  

[Anderton]

Recording at 32-bit float doesn't give any more practical dynamic range than recording at 24 bits, because either one hits the noise floor before reaching the theoretically available dynamic range. Where it matters is the audio engine, and it will automatically treat whatever file you give it with whatever resolution you've specified in the audio preferences. 

[zeronyne]

3 hours ago, Anderton said:

Recording at 32-bit float doesn't give any more practical dynamic range than recording at 24 bits, because either one hits the noise floor before reaching the theoretically available dynamic range. Where it matters is the audio engine, and it will automatically treat whatever file you give it with whatever resolution you've specified in the audio preferences. 

Thank you for this and your more detailed explanation. It’s a huge blind spot in my understanding because I haven’t ever really considered it before. I stand informed.

[Anderton]

On 6/17/2024 at 5:54 PM, zeronyne said:

It’s a huge blind spot in my understanding because I haven’t ever really considered it before.

 

You and a LOT of other people! Glad you found it helpful.

 

FWIW you might be interested in checking out my eBook, The Musician's Audio Handbook (the link takes you to details, including the chapters and contents).

 

 

[Stokely]

I hear absolutely zero noise in any recordings I do.  I listen on decent headphones and my current interface/headphone amp is a Motu M4.     I have a very modest home studio, nothing special at all.   I've done 16 and 24 bit, 44.1 and 48 (haven't tried 96).

My thought would be--there's about 100 things I could improve before I worried about noise...my songwriting, my singing, my room, maybe my mic (which is just the same dynamic one I use live) and so on.

It has actually taken me time to undo my ancient mindset--I learned recording and mixing on analog tape, with analog consoles and everything else.  Noise was very definitely the enemy and it was there.   Setting recording levels for a vocal at say -18db causes my eye to twitch because way back when you were pushing it up toward (or even into) the red... I still do automate mutes on channels when they aren't playing, old habit and there could actually be some noise when you have a bunch of plugins sitting there...but it hasn't been very noticeable if there is.

 

[analogika]

The other question is: do you have an interface that can record in 32-bit float?

 

AFAIK, the very first 32-bit floating ADCs hit the market in early 2023, and unless yours is one of them, the whole question is purely academic, anyway. 

[harmonizer]

Is it possible that we will see more continued emphasis on how to deliver accompanying video with all audio, and less emphasis on using higher bit and sampling rates for the audio itself? If yes, the various capacities to always include that video data might overwhelm the questions about data consumption to support higher audio bit and sampling rates.