Mad + mpg321 just became my mp3 playback tool of choice based on output quality alone.
FYI, my setup is a dual-boot RH7.1/Win98, with an SB Live and the optical i/o daughtercard, feeding into a JVC RX-1028V A/V deck. Under Linux that optical S/PDIF output seems to be a nearly-transparent passthrough, meaning it isn't mixed. I'm not sure if it's the card or ALSA doing the sample rate corrections. Probably the card.
This has yielded what I believe is the cleanest possible pathway for digital audio, because it's using the much higher quality DA's in the JVC stereo unit. I believe that for a digital recording, the only point of significant signal loss is the MP3 encoding and decoding. Thanks to MAD, the vast majority of that is now the encoder.
Which leads me to the reason for posting: this implies that much of the perceived fidelity loss of MP3's involves flaws in soundcard DA's getting tickled by something in the MP3 algorithms.
I'm available for whatever testing anyone can come up with. Keep up the good work!
- Kevin Cody Jr