Extract Vocals (demucs 2-stem)

Isolate vocals from a mixed audio file using demucs in 2-stem mode. Fast and focused — produces vocals.wav and accompaniment.wav. Optionally drop the accompaniment if the user only wants the vocal stem.

When to use

• The user wants vocals isolated from a song (e.g., for editing, re-recording, or remastering).
• Preparing a karaoke track — extract the accompaniment, discard vocals.
• Extracting a vocal guide from a podcast with music bed.
• Quick turnaround on a single stem (faster than 4-stem separate-stems).

Do not use this skill for:

• Full stem separation (drums, bass, other) → use separate-stems.
• Batch processing many files → use separate-stems with GPU for throughput.

Inputs