Tutorial · Tutorial · ~6 min

How to Extract Instrumentals from a Song

Karaoke backing tracks, remix stems, licensing reviews, practice tracks — an instrumental version of a song is one API call away. No DAW, no audio engineering, no manual EQ.

InstrumentalsKaraokeVocal separation

Get your instrumental

Step 1 — Understand what you'll get

'Instrumental' means the full mix minus the vocals. The model estimates the vocal stem and subtracts it, leaving everything else: drums, bass, guitars, keys. On clean sources this is indistinguishable from an official karaoke track to most listeners.

Step 2 — Submit the job

Python

import requests, time

API = "https://api.mlslabs.io/v1"
HEADERS = {"X-API-Key": "YOUR_API_KEY"}

# sample audio included on the dashboard — or point at your own track
job = requests.post(
    API + "/vocal-separator/jobs",
    headers=HEADERS,
    json={
        "input": "s3://bucket/my-song.mp3",
        "output": "s3://bucket/out/",
        "mode": "music",
        "tracks": ["vocals", "instrumental"],
    },
).json()
job_id = job["job_id"]
print(job_id)

Step 3 — Wait for the instrumental

Python

while True:
    st = requests.get(API + "/vocal-separator/jobs/" + job_id, headers=HEADERS).json()
    if st["status"] == "done":
        print(st["output"]["instrumental"])
        break
    if st["status"] == "failed":
        raise SystemExit("failed — credits auto-refunded")
    time.sleep(2)

Step 4 — QC the result

Note: Vocal bleed is the one artifact to listen for: some reverb on the vocal can survive separation. On dry, well-mixed sources it is rarely audible.

A/B against the original: vocals should be gone, groove intact
Check for 'underwater' artifacts in the low mids (chorus/echo bleed)
If bleed is audible, re-run with mode=music and a higher-quality source
For karaoke use, confirm the lead vocal is fully gone in the first verse

FAQ

Common questions

Is the instrumental identical to an official karaoke version?

Extremely close on clean sources, but not byte-identical: expect a few artifacts you can hear on reference monitors. For practice, streaming and remixing it is excellent.

Can I get just the vocals too?

Yes — request both stems in the same job (tracks: ["vocals", "instrumental"]).

Does this work for any genre?

Yes — the model handles pop, rock, hip-hop, EDM and classical. Dense, heavily-processed mixes are the hardest.

Keep exploring

How to Extract Instrumentals from a Song

Step 1 — Understand what you'll get

Step 2 — Submit the job

Step 3 — Wait for the instrumental

Step 4 — QC the result

Common questions

Related guides

Vocal Separator API

Make Karaoke Version In Python

Audio Source Separation Explained