[adsr]

I doubt that would work well in practice. The reason is that timing is extremely important for this to work, if you're even one sample off you'll here a faint sound, and more than a few then it's quite obvious. So if the sound is generated by a separate OS process, you'll never know exactly when the sound is generated and you will not even have control over exactly when your own sound is played back in relation to the other process, think sub ms accuracy.