It's not about your speaker output, it's about your wav file outputs.


32 bit is significantly better than 16 or 24 bit, not merely because it has a better noise floor (which is why 24 bit is better than 16 bit), but because it is floating point instead of integer.


What this means is that it uses those bits more flexibly to differentiate your points, but really, even 24 bit is more than the average human can hear.  However, when you are still working on your tracks, 32 bit also can safely go over 0db without clipping.  What I mean is, the numbers will go over 0.  This will still sound like crap on a sound system, this is not what you want as your final output, but while you're still working, it's beneficial to have it so bouncing your tracks to wav doesn't kill your track if you accidentally clip.


When you're ready to save down your master, you're probably going to want to export it as 16 bit, but until then, yeah, keep your wavs as 32 bit and you absolutely want your internal mixing to be 32 bit.  It doesn't matter AS much for your recorded tracks, but unless you have a serious disk constraint, it's still worth just keeping it the same as the rest of your chain.

Actually, I disagree... until you output your master, I think your renders and channel exports should be 32-bit.  If you are sending your work off for contract mix-down or mastering, then it is really going to be far better to hand off a 32-bit file to the engineer - even if you don't clip ... and you still shouldn't - try to stay below -6dB even in 32 bit before you've got your mixed down master.  FP just does a much better job of expressing the curves of your wave (16-bit FP would actually be better than 16-bit integers for audio as well).

Your engineer may want to perform most of his work in 32 bit, but if you don't provide him with 32-bit files, that's going to mean a conversion... so, you get minor artifacts from this downsampling.  Not a massive amount of artifacts, but it really pays to try not to introduce more artifacts to your process than necessary.  That's not to say the engineer will necessarily do this, but forcing them to work in 24 bit may not be the best option either.


Once it's mastered, then 16 bit integer is probably sufficient, 24 bit is fine but probably overkill... but until you have a master, my opinion is that you should stick with 32 bit FP.

You are absolutely right, it won't improve the noise floor - it will improve the ceiling, which is not nearly as big a deal... especially if you're leaving mastering headroom.

However, what I say about not downsampling still has some relevance... 32 bit float is not commonly dithered by your DAW when you export (Ableton, f.ex.), but an integer format usually is by default in many DAWs.  If one is going to make a habit of exporting WIPs in non-FP bit-rates, then it would be a good idea to disable the dithering defaults so that you don't introduce more dithering noise than you need to.

Seems like we could make more efficient use of FP for storing audio than our current method... I don't think IEEE is the optimum solution here.