Audio Stems
Source Separation API — split any mix into clean stems
A BS-RoFormer model trained to separate vocals, drums, bass, piano, guitar — and environmental sound like cars, rain, planes and thunder — from mono or stereo mixes.
Before & after
See the difference
One mixed file goes in — four synchronized stems come out, ready to rebalance or remix.
What it does
Most stem tools stop at music: vocals and instruments. Source Separation goes further. The BS-RoFormer architecture has strong representation of harmonic, transient and noise sources, so it separates musical stems cleanly and also isolates environmental sound for film and game SFX work. Every stem comes out phase-clean, artifact-free and ready to feed spatial workflows.
Full API referenceimport requests
url = "https://api.mlslabs.io/v1/source-separation/jobs"
resp = requests.post(
url,
headers={"X-API-Key": "YOUR_API_KEY"},
json={
"input": "s3://bucket/mix.mp3",
"output": "s3://bucket/stems/",
"stems": ["vocals", "drums", "bass", "guitar", "environment"],
},
)
job = resp.json()
print(job["job_id"])Key features
Why teams pick this API
BS-RoFormer architecture
Transformer-based separation with strong representation of harmonic, transient and noise sources. Clean output: no phase artifacts, no residue.
Multi-class stems
Vocals, drums, bass, piano, guitar, orchestral instruments and more — in a single pass.
Environmental sound
Isolate cars, aircraft, rain and thunder for film and game SFX work — a class most stem tools ignore.
Upmix-ready
Stems designed to feed spatial workflows: independent gain, delay and placement control for 2D to 3D upmixing.
Use cases
Built for real workflows
2D to 3D immersive upmix
Separate a stereo source, then place each stem independently in space — the input side of the stereo-to-Atmos pipeline.
Music production & remixing
Split songs into vocals, drums, bass, piano, guitar for remixes, covers, mastering prep and accompaniment.
Film & short-video SFX
Isolate cars, planes, rain and thunder as usable sound effects with full control.
Legacy content enhancement
Decompose and remix old stereo audio for smart speakers, car systems and cinema.
Live & interactive audio
Fast separation in real-time streams for live, meeting and in-car immersive experiences.
Pricing
Credit-based pricing
Every request draws from your credit pool. This API bills 18 credits per minute of audio.
| Billing unit | Cost in credits | Notes |
|---|---|---|
| Processing time | 18 credits | Per minute of audio |
| Failed jobs | 0 credits | Credits auto-refunded |
| Batch processing | Same rate | One queue, per-file callbacks |
Learn
Guides, tutorials & comparisons
Batch Separate an Audio Library
Batch separate an audio library into stems: folder-prefix jobs, per-track output, and catalog-wide s...
Read TutorialDemucs Alternative (Cloud, No GPU Needed)
A Demucs alternative for production: BS-RoFormer-class source separation in the cloud, no GPU or Pyt...
Read TutorialIsolate Environmental Sounds
Isolate environmental sounds from video: separate rain, crowd noise, and ambience into clean stems f...
Read TutorialMake a Karaoke Version of a Song
Make a karaoke version of a song in Python: separate the vocal stem, keep the instrumental, export a...
Read TutorialSeparate Audio Stems in Go
Separate a song into stems in Go with the mlslabs Source Separation API: vocals, drums, bass, other.
Read TutorialSeparate Audio Stems in Node.js
Separate a song into stems in Node.js with the mlslabs Source Separation API: vocals, drums, bass, o...
ReadAI Upmixing vs Matrix Upmixing: Why Stereo-to-Surround Sounds Different
Matrix upmixing spreads two channels around the room; AI upmixing separates the sources first, then ...
Read GuideBS-RoFormer vs Demucs: Which Source Separation Model Should You Use?
BS-RoFormer vs Demucs vs Spleeter: architecture, SDR scores, speed and VRAM. Community-aligned bench...
ReadFAQ
Common questions
Which stems can you output?
Vocals, drums, bass, piano, guitar (acoustic/electric), brass, strings, other instruments, and environmental categories (cars, planes, rain, thunder).
Does it work on mono input?
Yes — mono and stereo are both supported.
How does it compare to Demucs?
Demucs is a strong music-stem baseline; we add environmental classes and spatial workflow integration — see the comparison page.
Can I separate commercial tracks?
You must have the rights to process the content. The API does not distinguish, but your license to the source material is your responsibility.
Related APIs