Tomasz,
First of all, I have a few disclaimers that I should state. I can only say for certain the results of the tests we have conducted with one specific computer and a few specific test setups. As computer technology changes continuously, I cannot claim anything as certain. Also, any statements made about how Windows works is only based on deductive reverse-engineering. I cannot and have not seen the code, so I can't claim anything for certain.
Quote:
Originally Posted by tszyn /img/forum/go_quote.gif
Elias,
Many thanks for your prompt and detailed reply. It is really hard to find such well-written explanations on this topic anywhere on the Web.
1.
Your findings would seem to indicate that we don't need to worry about the mixing/volume-related distortion introduced by KMixer as long as we use a sound card that's capable of 24-bit output (e.g. the Realtek ALC882 on-board audio solution). As I understand what you wrote, some distortion will still be there, but will be very low -- much lower than with 16-bit cards. Am I correct?
|
Actually, even the 16-bit USB audio devices we used experienced very little distortion with Windows Volume Control (which I'll refer to as WinVC from now on). I believe what happens is, the USB device declares itself to the computer as being a device capable of x-bits, and then WinVC dithers its results to x-bits.
We have not tested sound cards. Although I would suspect that their performance is similar to USB audio devices, I would not say so with any authority. In fact, the one sound card which we did analyze performed sample-rate conversion, even on its digital output.
Quote:
Originally Posted by tszyn /img/forum/go_quote.gif
2.
Is there any way to verify whether my on-board Realtek ALC882 is in fact playing 24-bit audio after I decrease the wave volume setting?
|
Yes, you would need a bit-analyzer. Otherwise, I could give you a test file which could tell you (with high probability, but not certainty) whether the output was being truncated to 16-bits. I would be difficult to determine if it is dithered to 16-bits, however.
Quote:
Originally Posted by tszyn /img/forum/go_quote.gif
3.
You explained that when two 16-bit samples are mixed in Windows, the result is a 24-bit sample. I don't quite understand what's the advantage of 24-bit calculations in this case. If mixing two samples is done by adding two integers, where's the need for extra precision?
I mean, with 16-bit audio, each sample is an integer from the range 0..65535. Suppose the first sample is 12345 and the second is 20000. Can't you just add the two together, getting 32345? (Of course if the result were greater than 65535, you would get clipping, but it does not seem that you can avoid that by using 24-bit samples.)
|
Your assumptions are correct. If two 16-bit files are added together, they will not need 24-bits to represent the sum. What PROBABLY happens in WinVC is, the summing takes place and a 32-bit word results (with no audio information in the last 16-bits). WinVC probably truncates to 16-bits in this case, which will not affect the audio. Again, I don't know this for sure, but based on our experiments, this seems to be what happens.
Thanks,
Elias
