AudioSep model

This model was proposed in "Separate Anything You Describe".

image/png