CUHK · I connect natural language to audio operations — say what you want, and the model separates or edits the sound.
Two things you can ask for:
"the dog barking, without the music" -> promptsep (isolate a sound by description)
"remove the siren; add rain after 3s" -> sonoscript (edit the clip from text)
-
promptsep — language-queried audio source separation Isolate a target sound from a mixture using a natural-language query; a text encoder conditions the separation model.
-
sonoscript — text-guided audio editing Insert, delete, and replace sound events from text instructions, with time-aligned edits and evaluation.
NumPy core, optional PyTorch backend — offline-first, CI green.