Introduction
What this endpoint does
It splits one mixed song into five stems and returns each as its own WAV file.
Inputs that matter
audio_url is the only input: an MP3 or M4A song up to 10 minutes long.
What you get back
Five audio assets, each labelled in its stem field: vocals, instrumental, drums, bass, other.
| Field | Value |
|---|---|
| Input | Audio |
| Output | Audio |
Request parameters
| Parameter | Required | Description |
|---|---|---|
audio_url | Yes | The song to split: an MP3 upload or a direct link to an MP3 or M4A file, up to 10 minutes. WAV, FLAC and other formats are refused at submission and not charged. at least 1 characters |
Pricing
$0.108 / song
One flat charge per song, whatever its length, up to 10 minutes. All five stems are included. Failed requests are not charged.
Use cases
Prompt guide & pro tips
Introduction
Mureka Stem Separation is the stem splitter from Mureka, the AI music company. Give it a finished, mixed song and it hands back the parts: the lead and backing vocals on one track, the full accompaniment on another, and that accompaniment broken down again into drums, bass and everything else.
Request
{
"audio_url": "https://example.com/song.mp3"
}
The song must be an MP3 or M4A file. WAV, FLAC and other formats are refused at submission.
What comes back
The task returns five audio assets, one per stem, each with a stem field:
| stem | contents |
|---|---|
vocals | lead and backing voices |
instrumental | everything except the voices |
drums | percussion from the instrumental |
bass | bass line from the instrumental |
other | the rest of the instrumental: guitars, keys, pads, effects |
Every stem is a 16-bit, 44.1 kHz stereo WAV at exactly the length of the song, so all five line up from the first sample in any editor.
What we saw in testing
We split a 60-second bossa nova duet and a 30-second excerpt of the same song. Both finished in about ten seconds or less. Drums, bass and other summed back to the instrumental sample for sample. In the instrumental-only passages the vocal stem stayed close to silent, about 55 dB below the singing parts, so little of the band leaked into it. Each 60-second WAV was about 10.6 MB.
Task flow
The call is asynchronous: create the task, poll until it finishes, then download each stem from SpicyAPI storage. Failed requests are not charged.
Tips
- Start from the best copy you have; a heavily compressed MP3 carries its artefacts into every stem.
- For a karaoke bed use
instrumental; to rebuild the mix with a different drum sound usevocals,bassandother. - Stems are uncompressed, so a three-minute song comes back as roughly 160 MB of audio in total.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "mureka/stem-separation/audio-separation",
"input": {
"audio_url": "https://example.com/reference.mp3"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.

