🎵 AudioTextHTDemucs - Text-Conditioned Stem Separation

Upload an audio file and enter a text prompt to separate specific stems from the mixture.

Example prompts:

• drums - Extract drum sounds
• bass - Extract bass guitar
• vocals - Extract singing voice
• other - Extract other instruments
• Or any natural language description like "extract the guitar" or "piano sound"

Notes

• The model works best with music audio sampled at 44.1kHz
• Processing time depends on audio length (segments processed in 6-second chunks)
• The model was trained on stems: drums, bass, vocals, and other instruments
• You can use natural language descriptions thanks to CLAP text embeddings