Thanks Estel.

About the N900 as a recorder : yes, that would be another great use case for the USB-audio hack. This is exactly what motivated Oblomow's experiments which got the ball rolling...

About not using hardware acceleration (I guess you mean the DSP ?) for decoding : yes, Oblomow also reported high CPU usage in his mplayer test, and I think it was the same in my MOC test. I'm not clear about the reason for that though : is it because we're bypassing pulseaudio ? Or put otherwise, if we could use the standard Maemo media player with the USB output, would it be using the DSP ? Or is some totally different software needed ?

About Easy Debian : I haven't really used this on the N900, so the question may be stupid... but will h-e-n and USB host still work for apps running under Debian ?