Fons Adriaensen wrote:
16 bit means that there are 2^16 possible values for a sample. So the signal is quantised to the nearest level. Except in some special cases, the error (a rounding error) is random and appears as noise. For a 16-bit card, that noise will have a level that is 98 dB lower than the maximum amplitude sine wave it can produce. Let's assume the card is not really 'perfect' and you actually have 95 dB of dynamic range.
Where does that 98dB come from? A factor of 2 is roughly 3dB, so 16 bits should mean 3x16=48dB, no? Taking this figure, your example where the maximum level is set to 110dB will leave 62dB for pure noise, i.e between the level of a TV set and a handheld electric mixer (1) so perfectly audible. BTW, I generated a minimum amplitude signal in audacity and played it with my speaker volume set to max and it was perfectly audible (even with my window open and the birds singing outside). Since quantification noise is half that, it should be audible too.
If you don't believe this then ask yourself why speakers having an integrated amplifier and a digital input are so popular in professional circles. There is no 'volume' control on those (at least not one you'd normally use) the only way to play at low levels is by not using all the bits.
I doubt those use 16 bits input. Even low-end hi-fi digital recorders support 24 bits, which gives -72dB for the noise and starts indeed to be acceptable. But most end-user will simply set their system to "CD quality" (or leave it at the default which is usually that same 16bits 44kHz, whatever name the app chose to gave it).
But there's not reason why a software mixer shouldn't use floating point, or a fixed point format (e.g. 32 bit integers) that provides enough room above and below.
Well, 16 bits integers should be faster to process. The difference is not important when your computer is doing nothing but audio processing anyway, but we're talking end-user system here and sound mixing is simply a background process that should not take too much time from the real work of the computer, even if this results in less than optimal sound quality. You might argue that the difference in computing requirements is very little, but keep in mind that the "real work" could be a computer game where every cycle counts. This is especially true in the most common case where there is a single sound source: PA can send the sound directly to the sound card without wasting any time adjusting the gain in software first. Jerome (1) https://en.wikipedia.org/wiki/Sound_pressure#Examples_of_sound_pressure_and_... -- mailto:jeberger@free.fr http://jeberger.free.fr Jabber: jeberger@jabber.fr