Tools › Audio
Separate vocals from music
Split any song into vocals and instrumental, or into vocals, drums, bass and other, with the Demucs AI model running on your own graphics card. No upload, no account, no daily limit.
How it works
- We use Hybrid Transformer Demucs (htdemucs) from Meta AI research, one of the best open models for music separation, released under the MIT licence.
- The model runs on your graphics card through WebGPU. The first time, it downloads once (172 MB) and is kept in your browser for next time.
- The song is processed in overlapping 7.8-second pieces that are blended back together, exactly like the original Demucs program does.
Questions
Is my song uploaded?
No. The AI model comes to your device, not the other way round. Your file never leaves this tab.
How long does it take?
It depends on your graphics card; a time estimate appears after the first few seconds. Without WebGPU it runs on the processor and takes several times longer.
Can I use the result commercially?
The tool and the model allow it, but the song itself belongs to its rights holders. Use separated stems of songs you own or have a licence for.