32 bit float

[cichy]

why it is recommended not to deliver a 32 bit float file for mastering ?
what is the best results way of converting 32bit to 24bit file ?

[Lagerfeldt]

Might not be (easiily) readable and the actual sound data is more or less identical to a 24 bit file.

The 8 extra bits are basically reserved for extra headroom during digital processing.

[Cellotron]

32bit float MUST be converted to 24bit fixed point in order to output it to a DAC. As Lagerfeldt stated 32bit float makes for a nice storage format for allowing for extra headroom during a DAW app's internal processing - but imo it is essentially just bloated and unnecessary added data size with basically no audible benefits over just sending mixes at 24bit for mastering.

There are 2 ways to change 32bit float to 24bit fixed - either truncate it - or add dither at the 24th least significant bit when requantizing. My own tests have shown that truncation of 32bit to 24bit is generally completely inaudible in its effect - as long as it only happens as a single final step - as the area that quantization distortions are occuring from this conversion are well below the noise floor. Repeatedly doing these types of truncations - which in fact happens in some DAW apps during extensive signal processing - can potentially lead to some denigrations of the sound though - so in these cases you might want to make sure that the processors you choose are properly dithering back to the returned figures' bit depth.

Best regards,
Steve Berson

[24-96 Mastering]

There's only one potential problem that I can think of (and that I've come across before)... which is that with 32 bit FP files delivered to mastering, that the ME potentially doesn't know what the exporter listened to.

If there are peaks above 0dBfs (which in FP is possible), I won't know if they were attenuated or limited on some bus on the way out (before DA) or if they just ran into clipping.

I once got a song that had very significant peaks above 0dBfs and the accompinying notes said "can you make it a bit louder while keeping the sound?". That was a problem because 1. I didn't know for sure how loud the file played for the client and 2. since I don't know if and how the peaks / the level was reduced or clipped during playback, I don't really know what the mix sounded like at the client's place.

That's the trouble with 32bit FP. If there are no overs, it doesn't hold much advantage to 24bit. And if there are overs, I would rather have you drop your master fader and export 24bit... then we can be reasonably sure we're listening to the same mix.

[Cellotron]

When I receive 32bit float files I don't really worry about the above issues - I just load them into the DAW - truncate to 24bit on output - and just attenuate them as necessary if there are any overs on output - as the vast majority of tracks I master get sent to the analog process chain anyway. So far - no complaints from the clients that have sent 32bit IEEE files in regarding this - the reference approvals were just as easy to get as from clients that sent in other formats.

Best regards,
Steve Berson

[finetuner]

WaveEditor (Mac OSX) can read and convert 32bit float if your DAW doesn't.

[MASSIVE Master]

I notice that some apps open some 32BFP files from some apps and some don't.

[Lagerfeldt]

WaveBurner can too, it just converts it to 24 bits though. I also use Barbabatch.

Regarding clipping/not clipping don't we all just default to monstrous amount of clipping by default anyway? :-)