When would I use 24 bit instead of 32 bit float

[Goat Rodeo Cowboy]

I see it as a bit of over-kill, but when recording live events... meetings, seminars, speeches, sermons, etc, I have long used 32-bit float to capture maximum detail and to minimize residual noise. (If music is involved I may record at a higher sample rate.) After noise reduction, compression, equalization, etc... I convert down to 44.1/16 bit or whatever is needed.

My theory is that that dithering and other distortions will be minimized during the editing process.

The latest versions of Adobe Audition now permit recording at 24-bit. (I have always assumed that 32-bit float results in something kin to 24 bit data.)

Is there any reason to record at 24-bit rather than 32-float? My original USB device... an M-Audio Transit... has a switch setting for 24 bit. Using it with the older Audition resulted in a file that opened up as 32-float.

I know if I go to the Adobe help files looking for an answer... I may be a day and half before I come up for air... I will chase too many stray rabbits while I am peeking and poking my way around in HELP. ;D

 

[Emmett1057]

You can't record at 32 bit FP resolution. It's 24 bit resultion within a 32 bit FP file. The only reason for 32 is if you're going to do destructive processing to the file. Even with 24 bit files, the mix engine is still 32 FP.

Emmett

 

[Goat Rodeo Cowboy]

You can't record at 32 bit FP resolution.

If I were smart enough to engage you in a debate over that statement, I wouldn't have asked my question. ;D So, I shall not argue with you. But I will share with you WHY I even asked the question. I am trying to convince myself I will transition from Adobe Audition 2.0 all the way up to Audition CS6. And when I click on that little red record button, the window pops up and offers me some options. These are FILE options, and the drop-down menu for bit-depth is: 8, 16, 24, 32-bit Float.

When told that "Curiosity Killed the Cat", I always ask: "What was the cat curious about?"

So what different happens when I click the button for 24 vs. clicking the button for 32-bit Float? Until I do a File Save, does audition store and deal with the audio as 32-bit Float in memory and then do a conversion to 8, 16 or 24 when I click on save, or does the virtual file (in memory) and .tmp file (on disk) have "dumbed down" audio during all this process if I have selected 16 bit in advance?

Since my original post I did lose myself for a few minutes in Adobe Forums and Help Files. I gather someone doing spots and promos voice-over work where the "level" of all recorded material is up in a rather consistent and healthy gain setting, 32-bit Float may not offer much if any improvement. If on the other hand you have crowded a family into your studio and they are having a bit of conversation to be recorded and sent to Grandmother for her birthday, and some members of the family speak softly and have to have their contribution expanded significantly, then the 32 bit Float format would not lose the original audio down in the "too few significant bits" swamp.

In my case where I often record lectures, seminars, sermons and events that also include off-mic contributions or questions from the audience, making the original recording with software that processes and saves the file at 32 bit Float might be very desirable?

 

[Emmett1057]

Under no circumstance does 32 bit FP offer any advantage during the recording process. It does, however, require a bit more juice from the machine, so it can actually be a bad thing.

I'll try to simplify it down as much as possible. Think in terms of money. Cash. For this example, change only. Let's say 24 bit audio follows our money system, in that you have quarters, dimes, nickels and pennies. Now lets say 16 bit is the same, but with no pennies and 8 bit has no pennies nickels or dimes, just quarters.

Here's how this works...You can never get more accurate than a penny, no matter what. At least not in the physical world. In the virtual world, however, it's easy to deal with a fraction of a cent. That's like floating point. You cannot have something physically smaller (more accurate) than a penny. It does not exist. Just like, in this case, you cannot have something physically more accurate than 24 bit. No matter how you break the dollar, you will never be able to divide it by more than 100. But let's say you wanted to make three purcahses. Including tax, one was 17.734 cents, one 24.211 cents and one 11.899 cents. You can quickly round in your head that you would have to spend 55 cents. But that wouldn't be the exact price, it's just what you would spend. Now after that's done, let's say you decided to three more. 12.856, 14.723 and 5.099. Again, quick math gives us 34 cents, for a total of 89 cents. That's about how 24 bit math works if you're doing processing. Now, when processing in 32 bit FP, it can store all of those numbers and use exactly the right amount, in this case, a total of 86.522 cents. Now, you still can't pay anything smaller than a penny, so you would pay 87 cents. Two cents closer to being accurate. 32 bit FP is the same way. It's much more accurate processing, but you can still begin and end with something no more accurate than 24 bit. Floating point is virtual. It's the way math is done, not something you can hear.

So, if you're only recording and saving, there is no advantage to 32 bit FP. It's the equivelant of one transaction. Something that, with tax, is 49.775 cents. No matter what you do, you will only have accuracy to the nearest cent. It doesn't matter if you have the most advanced calculator in the world, our money doesn't go any smaller. 32 bit FP is a good thing when you're making multiple transactions (processes).

Now, to complicate matters, the mix engine in Audition is 32 bit FP. So if you work in multitrack to perform your processing and gain changes, there is no advantage at all to recording in 32 bit FP because the mix engine is taking care of that and will yield identical results to processing a 32 bit FP file, except that with 24 bits, you'll need slightly less HD speed to cover the streaming.

The bottom line is that, with Audition, it is physically impossible to record any better than 24 bit (which is 144dB of dynamic range). Even the best converters in the world will not come anywhere near that much dynamic range (my Apogees do about -114dB) and analog gear won't come near the converters (usually 90-100dB for a REALLY good analog unit), so 24 bit it already overkill, at least for recording.

Sample rates are another discussion of wasted space and there's no reason to record above your final delivery format, unless you're using ancient or junky converters.

 

[Goat Rodeo Cowboy]

I've been searching The Web for some info as to why anyone would find 32 bit float useful. It is obvious that the few venturesome souls who have deposited some flotsam out there are "writing somewhere ABOVE their paygrade". Some of them may know what is happening, but don't have the vocabulary to describe it. Most of them don't have the "know" or the vocabulary.

There must be some valid purpose or Adobe and Pro-Tools would not be spending as much time and energy adding it to their tool chest as they are doing.

Here is my current guess as to why 32 bit float is even on the table. Emmett... we have been focusing on the wrong end of the gain problem. 24 bit gives us an absence of noise at the bottom of the gain scale that is fabulous. We who come to the recording studio and process with some time in the broadcast world have always worried about keeping the gain at every stage as high as possible because noise could slip in anywhere. I don't know about you but it is a struggle for me to record anything down 25 to 40 dB below the Zero Reference level. 24 bit audio gives us the freedom to record way, way, way below the "digital clipping level" at the top end.

In a production world, a voice-over studio, we tend to record voices and they tend to be well regulated voices that never venture way, way off the top end of gain.


Let's assume you are recording a jazz group of 15 to 20 instruments. It is getting toward the end of the session and the group has decided to break the monotony by recording something that may not find a place on the final output, but it is a piece they love. Everything in going fine until and they take turns with their favorite riffs, the Tenor Sax gets a bit carried away and lets it "all hang out". And on playback you realize he may have had a slightly cracked reed that is putting some overtones into the mic that go un-noticed at the time. On top of that, the mics you assigned to that area of the studio tend to have a dip in frequency response so during the recording process so upper frequency EQ is being applied in the rack while recording.

Only later do you realize that particular track for that instrument is un-useable because the riff has gone into digital clipping.

Here is my theory on 32 bit float: It is MOST USEFUL during the mastering/mixdown process because effects that push a track louder than zero are not destroyed, requiring a return to the last available back-up recording. The 32 bit float is something like scientific notation and mantissas from some long forgotten math classes of years ago. So the overly loud track can have its gain reduced back down into the loudness range the standard digital system can deal with. So why ever use it in recording? In my example case of the wild tenor sax player, THE ORIGINAL TRACK is destroyed by digital clipping under the digital systems we are used to. There would be no available back-up. By recording in 32 bit Float the original recording may be recoverable.

We are so early into this that even Adobe and Pro-Tools haven't figured out how to turn the justification into verbiage. Maybe in a few weeks or a few months we will be saying: "Thanks for something we didn't even know we needed." or we will be saying: "You guys blew it on that one, didn't you."

 

[Emmett1057]

Hey GRC,

The idea is nice, it's just not correct. You are 100% right with your understanding of the mix bus and summing that exceeds 0dBFS. That is absolutely sound.

But during recording, it's still a 24-bit recording. The extra decimal places are all zeros. Here is a benefit for recording in 32 bit FP...I record a VO. I use Audition's Edit View to process or Pro Tools' Audio Suite (not RTAS). That process clips the file. If the file is 32 bit FP, I can simply normalize it down to 0dB without losing any info. Alternately, a mixdown that's done to a 32 bit FP file does not have to be re-bounced if there's one little clip spot, which would be especially handy in working with very long files.

That being said, the exact same results can be achieved by recording a 24 bit integer file and converting to 32 bit FP before processing. There is no difference between that and a file recorded directly to 32 bit FP.

Audition has been using 32 bit FP for YEARS, as have most others. Pro Tools has been using integer files, so they're really pushing the rebuild of their mix engine that has now "caught up" with everyone else.

32 bit float files clip during recording, just like everything else.

And a lot of it is marketing hype, in the end. Just like recording at 192kHz. Or even 96kHz. Just because you can do it, doesn't make it better, but companies really want you to believe that it is.

Emmett

 

[Goat Rodeo Cowboy]

Thank you for the information, Emmett.

Now let me go a step further, and let ME shoot down my own THEORY. Emmett has been insistent that there is no valid reason for recording at 32 bit float. And in reviewing what I wrote earlier here is what finally sifted out: The editing program (Audition, Pro-Tools, etc) is NOT converting the audio to digital. Up ahead of the editing program there is a sound card or other ADC hardware taking the analog signal, including one that is too high in gain, and converting it to digital and then handing it off to the editing program. It doesn't matter how smart, how skilled, how talented the Audition, the Pro-Tools or other software of your choice might be, if the sound card as executed digital clipping to the sound that is too loud, then there is NO extra floating digits for the software to apply and remedy the damaged sound.

So, unless you find some magic somewhere in your sandbox, recording above 24 bit gains nothing.

That leaves me with two questions to ponder:

If I am in multi-track mode and I have converted the other tracks to 32/float, and I bring someone in to record one additional track, would I want to record it so that it would save as 32/float, or do I convert it later.

If I am working with a DAW (I don't have one, so that is why I have to ask the question for someone else to answer) or other mixing device that has internal mixing at 32/float and it offers a digital out via USB or FireWire so I can capture the sound using Adobe Audition or Pro-tools, would there be any reason to make that hand-off as a 32/float data stream? Can that even be done?

Oh.... it's a curse to have a curious mind like mine. It keeps me up much too late... way too many nights. ;D

 

[Emmett1057]

Hey GRC,

You did a fine job of shooting down your own theory. One thing to note. Recording at 32 bit float gains nothing. BUT, to say recording at anything higher than 24 bit gains nothing would be slightly inaccurate. Working in 32 bit FP, we've learned that you still have 24 bits of dynamic range on a floating scale (144dB). If you actually recorded at true 32 bit integer, you would have 192dB of dynamic range. The former Pro Tools HD engine worked at 48 bit integer, providing for 288dB of true resolution. However, with all integer-based systems, 0dB is the absolute max. The debate could go on forever about which is better and the truth is, if I intentionally try, I can break either system. Both have weaknesses. So while recording at a higher integer bit depth would give you more dynamic range, as discussed earlier, we're not even getting close to 24 bit dynamic range in the real world, so anything more, while technically better, would be an absolute waste.

Now, your first question...It really doesn't matter. In fact, you don't gain any benefit, even in processing, by using a 32 bit FP file over a 24 bit file. Because the mix engine is written in 32 bit and the processing is occuring there, it really makes no difference.

For your second question...A DAW is simply a digital audio workstation. Pro Tools and Audition are both DAWs. I can only assume from context that you're referring to a standalone recorder/mixer. Those are nearly extinct and none use FP math. However, if we do a hypothetical scenario and say that there was a standalone mixing system that worked in 32 bit FP, it would still not be able to stream anything in FP because FP is a calculation, not a data stream. We're going very deep into the rabbit hole here...

This is an imperfect analogy, but I think it will make the point. Think of 32 bit FP like an elevator in a building with 100 floors. All of the data is stored in the elevator iself, but the elevator can slide (float) up and down that scale as needed. But again, the amount of actual audio resultion never exceeds 24 bit. It just slides up and down this scale, giving the appearance of more dynamic range. In reality, there's still the same amount of dynamic range as 24 bit, but if you clip, you can just slide down the scale a little bit. When it comes to audio, that scale is nothing more than a calculation. The calculation decides where it needs to go on the scale, but something must always calculate it. To stream, you need fixed integer numbers. You need something "real". To get the same scale to stream, you would have to have data for the whole scale at once, which would equal about 250 bits and would need a very large, very fast connection, well beyond anything in existance. In our analogy, instead of an elevator that slides up and down the scale, you would need 100 floors worth of elevators stacked on top of each other, but only one would have data.

Nothing wrong with having a curious mind. I have no formal training in any of this, I just like to work things like this out.

Emmett

 

[jefflaurence]

Guys this is without a doubt the best explanation of 24 bit and 32 FP relationship I have ever seen.

Now I have a blazing headache...I am going to take a couple Advil and go rummaging around in"those boxes" for my Otari 5050 and a nice soothing reel of 3M 206...

Aaaaahhhh relief!

 

[Goat Rodeo Cowboy]

Learning a few thing about digital sound has not been easy. Since I don't do this professionally everyday, I don't have vendors all anxious to pamper me with knowledge. I did miss out on one opportunity to work that formula backwards. In the mid to late 90s I was working for a large, national computer distributor (Back before Dell proved to the industry that distributors not necessary!!!) and one of my customers was The Watchtower Society world wide headquarters. You may recognize them by he name Jehova's Witnesses. They were buying a LOT of PCs... heavy duty servers along with some solid workstations. They would bug me to procure accessories including sound cards for their MACs. I made an effort to get up to speed, but advised them that if they had another source for MAC accessories that knew the ins and outs of the product they should stick with them until I felt more confident.

I was hammering away at home trying to learn PC audio. I had the benefit of regular visits from the factory rep from Sound Blaster. Nice guy, but for what I wanted to know... he was pretty close to useless. I didn't know how to ask the right question, and he didn't know what question I should have been asking.

A few years later I took over the sound functions of a church. They were recording sermons direct to CD using a stand-alone Marantz recorder. Well, the THOUGHT they were recording. I immediately recognized their CD masters were the pits! I went on a crusade to learn everything a mere mortal could know about making original one-up CDs, ripping CDs, editing the ripped audio files and burning copies. I was thrown out of some of Atlanta's best computer stores for asking questions that they were sure should never be asked and should certainly never even be asked!

By the way, that same humble little Marantz recorder is still grinding away with success, It was just picky about the blanks it was willing to make audio deposits on.

In that era the Internet and discussion groups were full of folk lore, half truths and full lies. By about 1999 the people who could have carried on intelligent discussion apparently moved on to other topics.

I don't have a quarrel with Emmett that going beyond 24 bit and 44.1 sampling offers no improvement to the normal workflow in a studio set-up. But here is the thing I do with some regularity these days that sometimes is well served by treading where angels fear to tread. Salvage and restoration. Audio and pictures. And being a masochist, I am now moving a bit into video.

I take audio cassettes of family events from 40 years ago and turn them into CDs. And I don't do rip-and-dub. I am experimenting with going to extra high bit rates and bit depth to help identify noise from program and make the cuts at the right places.

I am currently working with someone's picture of their grandmother. It was on a Genealogy website and was the size of a business card. Any normal effort to blow it up to 3-1/2 by 5 or 4 by 6 results in jaggies, artifacts, and general chaos. I am down to one final layer that needs the detail brought out of the mud and muck. Grandma was quite a beauty. (Is the picture old? Grandson is in his mid 70's.)

I'm trying to learn if I can do to sound what I am learning to do to pictures. That is why I keep zooming in on sound files in a way that would not be productive or cost effective in the typical studio situations you guys work in.

GRC

 

[Emmett1057]

Interesting thing about higher sample rates...There are two trains of thoughts amongst people who have a very good understanding of digital audio. Neither one is right, neither one is wrong. The differences are too small to measure and well below the threshold of human hearing, but still worth thinking about.

1. To go from 96k to 44.1 for a final delivery, enough math and filtering will be done that the audio will acutally be degraded further than if it had stayed at 44.1 the entire time.

2. Recording at 96k does capture additional sound in the form of frequencies outside the range of human hearing. If your computer can keep up, why not capture everything you can, even if you can't hear it?

My personal feeling is that it's not worth using the extra space and power for no advantage, in the end. It doesn't really matter. The differences, either way, are far too small to have any real impact on the audio.

FYI, for restoration, iZotope RX2 Advanced is exceptional. And speaking of iZotope, you can't get into bit-depth without going into dither, as well. Really good dither article here: http://izotope.fileburst.com/guides/Dithering_With_Ozone.pdf

Emmett