Musicgen Duration, Stereo, Melody

[chigkim]

It's awesome to see MusicGen on MMLX.
Can we have the duration parameter to set how long the generated audio should be.
The original musicgen has:
model.set_generation_params(duration=8) # generate 8 seconds.
Also it has model for stereo as well as feeding your own audio melody. It would be great for MLX Musicgen to support those models.
Thanks!

[barronalex]

As a work around you can use the max_steps argument to the generate script to set the duration.
There are ~50 steps per second e.g max_steps=200 generates about 4 seconds of audio.

Agreed, it would be great to have stereo support and audio conditioning! I might take a look later this week.