This demo seems to do just that:

http://acko.net/files/audiosynth/index.html

Edit: It's not exactly as "dynamic" as I thought. The different samples are generated once and then played in a loop. So the audio data is not generated in real time, but it is generated.