Tutorial · Python · ~6 min

How to Separate a Song into Stems in Python

No CUDA. No model weights to download. No OOM errors on long tracks. About five minutes, using our sample audio.

PythonStemsSeparation

Separate stems

Step 1 — Sign up and get your API key

Sign up (free tier included)
Copy the key from the dashboard
pip install mlslabs

Step 2 — Pick sample audio

Sample	Use for
concert-stereo.wav	Live performance, 4 stems
pop-song.mp3	Commercial-style mix
city-rain-night.wav	Environmental stems
movie-scene.wav	Dialogue + SFX separation

Step 3 — Submit a separation job

Python

import mlslabs

client = mlslabs.Client("YOUR_API_KEY")
job = client.source_separation.submit(
    input_url="s3://bucket/pop-song.mp3",
    output_url="s3://bucket/stems/",
    presets="music_4stems",   # music_4stems | music_6stems | environmental | all
    stems=["vocals", "drums", "bass", "guitar", "piano"],
)
print(job.job_id)

Step 4 — Poll for the result

Python

job.wait()
print(job.status)  # done | failed

Step 5 — Download and verify

Bash

# each stem gets its own download URL
curl -L -o vocals.wav <vocals_url>
curl -L -o drums.wav <drums_url>

ffprobe -show_streams vocals.wav | grep -E "channels|duration"
# => channels=2, duration matches source