24bit format is signed integer, 32bit is floating point, so they're two entirely different number systems (I'm assuming most common formats here, some weird cases exist where you might have 32b integers, but rarely in audio), so it makes little sense to try to compare them as such. Going from floating point to integer is not truncation or bit-depth reduction like going from 24bit to 16bit.

If you're really interested, look up how floating point numbers are handled https://en.wikipedia.org/wiki/IEEE_754

 

The format of the sample does not determine the bitdepth and sample rate the sample is processed in. Your DAW doesn't care what your sample is rendered as - it'll convert it to 32b/64b float as soon as you drop it on a track anyway. Your rompler too will convert it to 32b/64b float as soon as you hit "load library".

Every effect applied to the sample inside the rompler/DAW will always, always, always do its processing in 32b or 64b float.

If your endgame here is to render your samples to save CPU, remove the rompler, load the sample, apply fx, render again, fuck, forgot something, load the processed sample, make changes, render again, load the sample ... then you'll be converting between 24 int / 32/64 float every time + several times.

Where you draw the line and whether any of this even matters to you (or anyone else) is your decision.

You know where 24/16 int distortion starts (-144/-96 dB), and you know you'll be adding it every time you render to 24/16 int, and you know how much post processing you'll end up applying. If you run a saturation effect at 24 dB gain on your 16 int sample, followed by an instance of OTT that brings up another 24 dB, then your 16 int distortion from your previous render is now -48 dB loud.

Does it matter? Not if your sample never gets quieter than -48 dB. Or if you post-gate at -48 dB.

Up
to
you.

Context is everything. I've put 50 dBs worth of compression and saturation on snares (and pretty much nothing else, except for sound design).

 

Last edited: Feb 26, 2025

There is no truncation, since 32 bit float uses a 23 bit mantissa. The main practical difference/advantage of 32 bit floating point is that the gain can be scaled.
https://www.sounddevices.com/32-bit-float-files-explained/

 

But for exchanging multitrack or stems with mixing/mastering engineer the track will end up in max 24-bit depth. Your master will finish as wav, anyway.

 

32 bit? lol. you guys are brainwashed.

16 bit is more than enough for everything

 

first you should choose 32 bit float. besides quality there is the big pro that you can't destroy your audio by clipping the recording. (this is in the box.... if it's external synths your converters can't record to floating point anyways)
to the other options, don't use them but there is a general rule that is "only dither once"! you never dither unless the final bounce of the track after mastering. every dither adds noise.

 

the headroom of 24 bit is identical with the headroom of 16 bit or 8 bit audio, the headroom is the difference between the level you averagely work in and the maximum, e.g. -15db to -0 db.

you know how noobs work: they push all faders to the max. it can easily happen to anyone that you clips your audio here and there, and working in float from the start to the end would have prevented you from that.

we use 32 bit since 1998 now, where size actually still mattered, and unless you use more than 2500 tracks in your projects, nowadays the 25% greater size should be really irrelevant.