Topic: What is 32bit audio?

[mozmoz8]

Does anyone know what it means if something is 32bit 96kHz? Does it sound better than 24bit 96kHz?

[sunjan]

Does anyone know what it means if something is 32bit 96kHz? Does it sound better than 24bit 96kHz?

To my understanding, pretty much all digital sound devices have a maximum of 24bit input.

But as an example, some soundcards support processing at 32bit:
http://www.emu.com/products/product.asp?product=10447&nav=technicalSpecifications
"Bit Depths: 24-bit I/O, 32-bit processing"

I don't know much what this means in real terms. Maybe someone else can chime in?!

/Jan

[Petrus]

Many audio editors internally convert all audio to 32 bits before manipulating the signal, then round the end result to what it originally was when saving the file. Upconverting is easy, just add 16 or 8 zeroes to the end of each sample. The idea is to minimize rounding errors, if calculations were done in 16 or 24 bits, after heavy effects and editing the last one or even two bits would be quite random and you would loose up to 12 dB S/N ratio. With processing done in 32 bits and rounding result the end result is 16 or 24 "true" bits. With modern processor power usin 32 bits for calculations is not an issue anymore.

[digifish_music]

Does anyone know what it means if something is 32bit 96kHz? Does it sound better than 24bit 96kHz?

The technial answer above is the reason.

If you are seeing discussions of 32 bit 96 kHz as a listening format, that means that marketing has written the information you are reading.

No 32 bit does not sound better than 24 bit...and (this is where I will incur the wrath of the faithful), neither does 24 bit sound better than 16 bit(all else being equal).

For those outraged, please post a link to a single scientific paper that shows people can reliably distinguish 16 from 24 bit in double-blind listening experiments.

digifish

[DSatz]

> Does anyone know what it means if something is 32bit 96kHz?

Yes. It means that the sampling rate of the a/d converter is 96 kHz (which can be nice), and that the results are being stored in a stream of four-byte samples (which can be nice if your intention is to process the result in software that expects this format--no conversion is required)(but otherwise it's a big waste of space).

It's like the difference between 0.4 and 00000.4--same data, different representation. The second number is no more precise or accurate than the first. The second representation would be more convenient if you were going to enter this number into a database all of whose entries were in this same format--but what it expresses or contains is completely identical to 0.4.

No a/d converter in the world has anything even approaching 32-bit resolution, so any four-byte bitstream that's coming from a converter (as opposed to data generated by algorithm) will contain a lot of information which is just acting as a placeholder. In most cases for live recording it simply means that at least 1/4 of the storage space is being wasted. There certainly is no direct benefit in terms of sound quality.

Meaningful 32-bit resolution for any acoustically-generated audio signal is impossible. 32 bits represents a dynamic range of nearly 194 dB, which greatly exceeds the dynamic range of air itself--the ratio between the energy of molecular noise and the point at which the "rarefaction" part of a sound wave reaches zero pressure. (Air actually "clips" or "brickwalls" at very high sound pressure levels, on the negative-going side of the cycle.)

24-bit resolution is theoretically just barely possible, but no present-day converters achieve it--the best converters of today have a resolution of "only" 21 to 22 bits. So even in the best case "24-bit" merely describes the data stream in which those 21 or 22 meaningful bits are sent down the wire; it's not the actual resolution of the audio data in that stream, if it's coming from an a/d converter.

Again it's like the difference between 0.6 and 00.6--the content is one thing while the form of presentation is another (though obviously you can't fit a quart into a pint bottle), and mainly you need to know whether these "bitness" numbers are talking about the content or the presentation.

--best regards

[datbrad]

Great explanations !!

[Will_S]

For those outraged, please post a link to a single scientific paper that shows people can reliably distinguish 16 from 24 bit in double-blind listening experiments.

I'm not outraged, but there are plenty of studies where folks could hear a difference in the noise floor of a very quiet passage, turned up to the point that a full-scale signal would be quite painful.  I know that's not what you meant, but it is an audible (under the right conditions) difference.

As to more realistic listing scenarios, I don't know of any studies off hand, nor have I looked very hard, nor do I personally hear a difference on my modest playback equipment.  It's important to understand though that double-blind tests and the accompanying statistical analyses may be great for assuring that you don't falsely make a definitive statement that a difference does exist that is in fact inaudible, but are not as sensitive or appropriate for making definitive statements that a difference does not exist or is not audible.  Google type 1 vs. type 2 error if you care.  And statistics aside, there are also those who argue that the "stress" of testing interferes with the perception of subtle differences.  Sounds kind of like a cop-out, but perception is a tricky thing, so...who knows?

[waltmon]

I encountered the 32 bit scenario in Cool Edit Pro when I was editing 24 bit files...a bit off topic, but is there a way to disable the 32bit and keep it 24bit....?

  The volume seemed to be jacked some when I did my 4 track 744T mix-down...

Thanks,

Walt

[Brian Skalinder]

I encountered the 32 bit scenario in Cool Edit Pro when I was editing 24 bit files...a bit off topic, but is there a way to disable the 32bit and keep it 24bit....?

The volume seemed to be jacked some when I did my 4 track 744T mix-down...

CEP/Audition uses 32bfp internal processing, and it's not configurable - internal processing is always 32bfp.  (WaveLab, on the other hand, I think allows the user to select the internal resolution.)  There are only two CEP/Audition configuration options relevant to the internal processing resolution:

• Perform all editing at 32bfp, and dither the edited portion to 16-bit after every edit
• Perform all editing at 32bfp, and after processing, leave the results in 32bfp

If you're only applying a single edit to any portion of the WAV, the 1st option will work.  But I recommend the 2nd option.  It should be used if you're making more than one edit to any given portion of the WAV (to avoid dithering multiple times), and it also provides more options for dithering (e.g. selecting type of dither and noise shaping).

Not sure what you mean by "volume jacked".  Louder?  Probably, since you're summing multiple channels.  I don't think the volume issue you mention has anything to do with CEP/Audition's internal processing resolution. 

[digifish_music]

For those outraged, please post a link to a single scientific paper that shows people can reliably distinguish 16 from 24 bit in double-blind listening experiments.

As to more realistic listing scenarios, I don't know of any studies off hand, nor have I looked very hard, nor do I personally hear a difference on my modest playback equipment.  It's important to understand though that double-blind tests and the accompanying statistical analyses may be great for assuring that you don't falsely make a definitive statement that a difference does exist that is in fact inaudible, but are not as sensitive or appropriate for making definitive statements that a difference does not exist or is not audible.  Google type 1 vs. type 2 error if you care.  And statistics aside, there are also those who argue that the "stress" of testing interferes with the perception of subtle differences.  Sounds kind of like a cop-out, but perception is a tricky thing, so...who knows?

Inferential statistics are merely tools to judge probability, I know about type 1 and 2 errors (hypothesis testing). And as I just hinted, I am more interested in probabilities the p values from these tests give than a binary decision. I am a psychophysicist by training and among other things, I work in the area of product R&D in visual perception and have in the past done some psychoacoustic consulting for Rotel (in the days when 1-bit Sigma-Delta D/A converters hit the market). I have followed the whole 16 vs 24 bit and 44.1 48 kHz and 44.1 vs 96 kHz debates in the scientific literature for the best part of 20 years. 

People who cast doubt on the validity of double blind listening tests with repeated measures and objective response metrics are those outraged faithful. Perception is driven in a large part by expectation and it is very hard to convince people what they think they feel or hear is not real. If a sufficiently large sample of subjects perform at chance levels in distinguishing 16 from 24 bit recordings then I'd say that's a good case that the differences are inaudible. And there are many such studies (The Journal of the Audio Engineering Society in particular is a great place to start http://www.aes.org/journal/).

I have not seen any well designed research that shows people can determine the difference between 16 or 24 bit nor 44.1 vs 96 kHz recordings. To the contrary the opposite.

Now there are certainly good technical reasons for recording at 24 bit (or whatever bit-rate it actually achieves) and working in 32 bit float, but that is another issue.

digifish

[DSatz]

digifish, this thread is getting sidetracked by needlessly provocative postings. Let's try to stay on the subject, or else start a new thread called "How can I argue about 24-bit recording?", OK?

As to the point of all this: Very few people listen to 24-bit recordings as such; much more often, 24-bit recording is used for "sound acquisition," and the result is then normalized in level and dithered down to 16-bit for listening and delivery.

If you record in a 16-bit medium for 16-bit delivery, you must set your recording levels as high as possible to get the best possible signal-to-noise ratio. But in live recording, this exposes you to the risk that unexpectedly loud sounds will ruin your recording. The risk can be reduced through experience but never eliminated. It makes live recording a tense affair at times, especially when--as with most recordings that I make--one is the only person recording the event. (I record classical concerts and recitals professionally.)

24-bit recording is extremely useful for live performance recording because it allows you to set recording levels conservatively without giving up the full signal-to-noise ratio of the eventual 16-bit delivery medium (CD or whatever). It takes away the fear, increases the reliability of the process and restores the fun. To me that's been worth a lot.

--best regards

[boojum]

I will not say anything on this except that for the wide dynamic range I, too, record in 24 bit.  I set my levels quite low to insure lots of headroom.  I can always make it louder. but never can I remove the clipping.  With some of the better equipment, hardware noise is a minimal problem.  YMMV   

[Petrus]

If I understand corectly Neuman makes a "digital microphone" which internally digitizes the signal at 24 bits. Basically there is no level setting at all, it is fixed, as the S/N ratio of the electronics is very high. Full level is set at mic's clipping point and all lower levels just take care of themselves. It works... 24 bit dynamic range is anyway larger than any microphones DR.

[digifish_music]

digifish, this thread is getting sidetracked by needlessly provocative postings. Let's try to stay on the subject, or else start a new thread called "How can I argue about 24-bit recording?", OK?

As to the point of all this: Very few people listen to 24-bit recordings as such; much more often, 24-bit recording is used for "sound acquisition," and the result is then normalized in level and dithered down to 16-bit for listening and delivery.

If you record in a 16-bit medium for 16-bit delivery, you must set your recording levels as high as possible to get the best possible signal-to-noise ratio. But in live recording, this exposes you to the risk that unexpectedly loud sounds will ruin your recording. The risk can be reduced through experience but never eliminated. It makes live recording a tense affair at times, especially when--as with most recordings that I make--one is the only person recording the event. (I record classical concerts and recitals professionally.)

24-bit recording is extremely useful for live performance recording because it allows you to set recording levels conservatively without giving up the full signal-to-noise ratio of the eventual 16-bit delivery medium (CD or whatever). It takes away the fear, increases the reliability of the process and restores the fun. To me that's been worth a lot.

--best regards

This is on-topic as the original poster asked if 32 bit sounded better....and I quote "Does it sound better than 24bit 96kHz?". I was reducing the line of discussion to the end-game, and that is whether you can distinguish any format above 16 bit 44.1 kHz from that base-line.

It seems to me you are not reading threads thorougly as what you said I agree with in my last sentence...

"Now there are certainly good technical reasons for recording at 24 bit (or whatever bit-rate it actually achieves) and working in 32 bit float, but that is another issue."

digifish

[Petrus]

Oh yes, but what is the noise floor of the said mic?? I doubt if the S/N ratio is over 144 dB.  =:-)