I'm interested in this as well, as I've run into the same problem and didn't have time to solve it, before the game needed to ship (client work). How do you import your sounds? By using Flex annotations?
Far as I can tell, is that the best approach to compression, is to use programs that are designed to do it and import the compressed file into Flash. I use Photoshop to compress images for example. By using Audicity it seems like you got that covered.
At the moment I'm working with the sound buffer directly in a hobby project. I'm surprised to see that you need a multitude of 11.025 khz. The sound buffer needs to be filled with a bytestream and uses 44.1 * 2 * 4 per milisecond. That stands for:
- 44100 samples per second, (44.1 khz)
- 2 channels, left and right
- 4 bytes per sample, as every sample is an integer, which gives a resolution of 2^32.
So that would bring it at 44.1 * 2 * 4 * 1000 = 352800 bytes per second, or 344.53125 kilobyte per second. An odd bitrate to say the least. I suspect they don't use the whole range of values the integer provides and cap it at at some value to get a 320 kByte bitrate. This is just a suspicion though.
That bieng said, I think most problems come from using a samplerate that differs from 44.1 khz. I don't think resolution is much of a problem as Flash could easily normalize the value and find the appropiate value within it's own resolution. I suspect (and this is still an untested hunch), that most problems stem from using a lower samplerate, as you'd need to fill in blanks. When the mp3 runs at 1/2 or 1/4 (I noticed that 11.025 is 1/4th of 44.1kHz) you can easily duplicate it, but running at 3/4 would have you choosing between two samples almost every converted sample.
I hope this gives you some more insight, it did for me at least. I haven't checked if the soundbuffer used by the VM runs at this bit- and samplerate, or just the soundbuffer that is used by the API ,although the latter would make little sense to me.