Source Separation
Separate audio into individual stems (vocals, drums, bass, other) using Demucs on GPU.
/v1/audio/separate
curl "https://audio.toolkitapi.io/v1/audio/separate?audio_url=https://example.com/song.mp3&model=htdemucs&stems=4stem"
import httpx
resp = httpx.get(
"https://audio.toolkitapi.io/v1/audio/separate?audio_url=https://example.com/song.mp3&model=htdemucs&stems=4stem",
)
print(resp.json())const resp = await fetch("https://audio.toolkitapi.io/v1/audio/separate?audio_url=https://example.com/song.mp3&model=htdemucs&stems=4stem", {
});
const data = await resp.json();
console.log(data);# See curl example{
"status": "ok",
"vocals_url": "https://storage.example.com/audio/vocals.mp3",
"drums_url": "https://storage.example.com/audio/drums.mp3",
"bass_url": "https://storage.example.com/audio/bass.mp3",
"other_url": "https://storage.example.com/audio/other.mp3",
"task_id": "abc123"
}Try It Live
Description
How to Use
1. Provide a publicly accessible audio URL.
2. Choose a model: `htdemucs` (default, good quality) or `htdemucs_ft` (fine-tuned, better for music).
3. Choose stem separation: `2stem` for vocals + accompaniment, or `4stem` for vocals, drums, bass, and other.
4. Each stem is returned as a separate downloadable audio URL.
About This Tool
Separate mixed audio into individual stems using Meta's Demucs deep learning model on GPU. Choose between 2-stem (vocals + accompaniment) and 4-stem (vocals, drums, bass, other) separation.
Perfect for creating karaoke tracks, remixing songs, isolating instruments, and audio analysis.
Why Use This Tool
- Karaoke — Remove vocals to create instrumental backing tracks
- Remixing — Isolate drums, bass, or vocals for remix production
- Music analysis — Analyze individual instrument parts
- Sample extraction — Extract clean instrument samples from full tracks
Frequently Asked Questions
What's the difference between the two models?
Can I separate more than 4 stems?
How long does separation take?
Start using Source Separation now
Get your free API key and make your first request in under a minute.