24 or 32 bit float

[WernerF]

Not that it really matters but, the other thing that has not been mentioned here is that an audio file that has been recorded at 32 bit can not be clipped. You can always bring the waveform back down to a point below clipping with the clip gain trim. This is, of course, provided that you have not clipped the converter or mic pre's on the way in. Of course this is pretty irrelevant to anyone who follows the principles of proper gain staging but worthwhile mentioning nevertheless.

[Bob Olhsson]

Originally Posted by daeron80

...Its main advantage is that it doesn't need to be dithered after each digital process...

Actually it does need to be dithered but I'm told it's difficult and exactly how that math works is way beyond my pay grade.

The way I think of it is audio written full scale 23 bit with a gain control that can be adjusted after the fact. This allows you in some, but not all, cases to perform signal processing that results in numbers that exceed full scale while retaining the ability to reduce the level after the fact back to where it won't clip a converter. Digital audio emerges from an A to D converter chip dithered to a 24 bit stream. There's no reason to record more than that. Once you start crunching numbers, you then need to worry about both dither and clipping. Using 32 bit float can certainly speed up a work flow but it isn't the magical solution lots of people seem to think it is.

In the earliest days of the internet we had some of the most knowledgeable engineers in the world posting to usenet about the technology. I only learned enough to be dangerous but thankfully some of us who were curious did get to learn quite a bit during the late 1980s.

[drumzalicious]

Basically 32 bit float is good for if you're going to be doing a lot of editing where you will be changing the gain of certain files, using audiosuite plugins (think Vocalign, Pitch Correction, etc), and stuff like elastic audio.

[daeron80]

If you have a 32-float file, processes it at 32-float, and save it back to 32-float, no audio program I know of will offer to dither it. Some, like iZotope products, will offer dither when saving a 32-bit processed file to 24-bit fixed. So maybe it would be clearer to say that the advantage is, you don't have to worry about dither when you stay 32-float because it's not an option.

E.g., when I do a bunch of noise reduction work on a file with RX, I begin by saving a version of the file at 32-float. Then I do my overall EQ pass, broadband NR pass, and then fix all the little individual problems. When I save my work every few minutes, I don't have to think about whether or not to dither it. It just saves to 32-float and I go on, which is conceptually the same as what happens in Pro Tools when an audio stream gets passed along from plugin to plugin to a buss to more plugins at 32-float. When the NR work is done, I save it, then save-as to 24-bit with TPDF dither. That file goes back into the Pro Tools session the original came from. I find that that workflow ends up sounding better than if I save the 32-float processes of RX back to 24-bit a few dozen times over the course of working on it. Better to keep it at one rate than to convert it back and forth repeatedly. This is probably one reason HDX is perceived to sound superior to TDM in many cases, since the TDM signal path sometimes has to convert the signal to 24-fixed and back a few times.

If I were to decide to do all that NR work in PT instead of RX, I would want the session rate to be 32-float. My experience is, one or two layers of conversion are no big deal. After a few, quality may start to suffer noticeably. After a dozen or so, it starts getting ugly. That's why I say that a movie session with lots of sound design work going on may benefit from the 32-float session format. It all depends on how much of the processing is going to take place offline or is going to be reprinted after successive changes.

[Gonk]

I haven't read in-depth on the processing side of things, but though 24 bit is more than good enough in terms of the dynamic range of human hearing but I have found 32 bit extremely handy (alongside clip gain) for overages something 24 bit is not so kind with.

When setting up a session quickly and if those odd spikes come up when a singer gets too close to the mic, the track can be easily "fixed" with clip gain as long as the pre amp was not overloaded and distorted for too long.

So in that respect I have found 32 bit an advantage.

[janmuths]

Originally Posted by WernerF

Of course this is pretty irrelevant to anyone who follows the principles of proper gain staging

Couldn't agree more.


J.

[daeron80]

Originally Posted by WernerF

Of course this is pretty irrelevant to anyone who follows the principles of proper gain staging.

Actually, this is only true with respect to clipping. If all one is doing is tracking and mixing, simple and straightforward, then it's true. For anyone doing offline processing where using a fixed bit format results in multiple layers of either dither or quantization distortion, it becomes a factor. You can't avoid quantization distortion with proper gain staging. You can avoid it with dither, but at the cost of a raised noise floor. Raise that floor just so many times, and you begin to cloud your sound. For film people who are printing submixed elements, printing them again with effects, printing stems with those included, printing mixes of stems, and so on, the problems can creep up. That's where a floating point file format can make a difference.

[janmuths]

Originally Posted by Gonk

When setting up a session quickly and if those odd spikes come up when a singer gets too close to the mic, the track can be easily "fixed" with clip gain as long as the pre amp was not overloaded and distorted for too long.

So in that respect I have found 32 bit an advantage.

Doesn't clip-gain work equally well for 24bit files, assuming you use it to bring the level down a bit?
And if the pre-amp or converter (=24bit) clips any recording should be equally distorted, no matter if the software stores the converter's output as 32-float or 24bit. Or am I missing something here?

Cheers,

Jan

[WernerF]

You can only restore a 32 bit file to an undistorted waveform with clip gain. It is not possible to do this with a 24 bit file. Clip gain can make the 24 bit file lower in level but the waveform will still be clipped if it was clipped in the first place.

[daeron80]

Try it.