What do the 32/64/128 bits in sound cards mean?

[invictius]

To me, midi-wise, it seems to be the quality of the instruments. A vibra 128 sounds more realistic than an awe32.

Also... is it practical to have several sound cards in the one windows 98 machine? I know it's probably impossible to have more than one isa.

[Remilia Scarlet]

Most of the nice have integrated DSPs, so my best guess is the size of an integer on those these days. Sorta like saying a Playstation is 32-bit or a Genesis is 16-bit.

[kb1]

It's the number of bits used to represent the amplitude of a single sample. If running at 44,100 hz using a 32-bit sample, that's 44,100 32-bit numbers per second, corresponding to 44,100 possible changes in voltage level (samples) going to your amplifier, and, therefore, 44,100 possible speaker positions per second (assuming that your speaker could move that fast, and that your amp can deliver that response).

The "32" describes the resolution of different possible voltage/amplitude levels. 32 means that you have 2 to the power of 32 possible levels, or about 4 billion (4,294,967,296). This number is signed, so you can values from -2147483648 to +2147483647. 0 = the speaker at rest. For a concert pitch of the note A (440 hz), being rendered full volume as a sine wave through a 44.1 Khz, 32-bit sound card, an entire cycle is about 100 samples. You'll have 25 numbers going in a sine wave pattern, from 0 to +2147483647, then 25 numbers decreasing to 0, then 25 more going negative towards -2147483648, then the final 25 finally rising back to 0. This repeats 440 times per second, giving you a 440hz note.

In other words, if we assume that your sound card, preamp, amp, and speaker could faithfully recreate this signal perfectly (which it can't), it could be roughly stated like this:

44,100 times per second, a 32-bit sound card reads one 32-bit number which creates one of 4-billion possible voltage levels that, when amplified, magnetically move the speaker to one in 4-billion possible positions along it's possible travel. A level of 0 is the speaker at rest. A maximum value positive number is either the speaker pushed out all the way, or if wired backwards, in all the way. A negative voltage moves the speaker in the opposite direction.

The real effect of having more "bits" is noticed the most in quiet passages, where more bits allows more tiny speaker movements. When everything is loud, it doesn't make a lot of difference. You may notice that, on some music sources, when the song fades out at the end, it gets trashy sounding. That can be caused by a lower bit rate, when recorded, or when played back. Also, with higher bit rates, multiple manupulations of the sound will incur less error, and therefore less noise.

[AndrewB]

Keep in mind that these gadgets are largely named for marketing reasons and have nothing to do with what kb1 mentioned.

[fraggle]

If you search hard enough you might find some explanation or excuse for them, but if we're honest they're marketing and mean nothing at all.

[Remilia Scarlet]

kb1 gave a good explanation of bit depth. But anything over 24-bit in audio gets to be pretty meaningless because of limits of circuit design in terms of signal-to-noise reponse (SNR), and because that's approximately the limit of the human hearing system. At that point, more bits just means accuracy during volume adjustments, or during certain algorithms, such as IIR filters, and not necessarily final perceived quality.

As far as dynamic range goes, at 32-bit, the extra 8 bits act as headroom for the dynamic range. 64-bit audio chains just give you more headroom. An anecdotal story I can offer in attest to this happened to me about 8 or 9 years ago. In Sonar, which has a 64-bit audio path, a bass line I was writing sounded fine. But, any time I rendered it to a 32-bit WAV file (my usual practice during export), it became distorted and clipped. The reason it sounded OK in Sonar was because of the extra headroom. The accuracy/dynamic range sounded identical during softer parts, otherwise.

Anything that advertises audio bit depth over 64-bits is just that: advertising. Messing with WAV files over 32-bit floating point probably won't buy you much quality, either.

[kb1]

Yes, marketing can call the product anything they like, and, yes, a 128 bit depth for sound samples is quite ridiculous overkill. I assumed the OP was describing a spec with "32/64/128". If we're talking about specs, vs. a product name, the bit-depth spec is the number where you'd expect to see a power of 2.

yukib1t: When you say "headroom", do you simply mean more math precision for less error over multiple calculations, or are you describing support for a range above 100% amplitude, to allow intermediate results to avoid clipping?

[Remilia Scarlet]

Both. It does provide additional accuracy for the calculations, but it can also lead to having a bit more amplitude headroom. Or rather, a bit more perceived headroom due to the accuracy. It's harder to clip audio in Sonar than, say, in Renoise, assuming you're doing the right things with levels to begin with and not being stupid with them. This was why when, in my little anecdote with Sonar, I dropped the 64-bit engine, it clipped like I was hearing in my resulting WAV.