Skip to main content

mureka/stem-separation/audio-separation

Mureka Stem Separation takes one MP3 or M4A song and returns five separate WAV stems: vocals, instrumental, drums, bass and other.

From $0.108/request

Workbench

Input

Audio separation
0/1

Drop an audio file here, or browse · 1 file · Max 90 MB · MP3

0 / 1 files
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. Song length does not change the price. at least 1 characters
Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

1 required field missing: Input audio

Output

Ready for a new run

Your result will appear here.

Configure the inputs, review the estimate, then run the model.

Related models and next tasks

Missing the model you need?

Take a finished mix apart into five tracks you can edit

Mureka Stem Separation takes one mixed song and returns it as five WAV files: the vocals, the full instrumental, and that instrumental broken down again into drums, bass and everything else. Each file runs the exact length of the song, so the five line up from the first sample when you drop them into an editor. The request is a single field pointing at an MP3 or M4A, and one song is one flat charge whether it lasts ninety seconds or ten minutes.

Split a song in the playground
Input
One MP3 or M4A song, up to 10 minutes
Output
Five WAV stems: vocals, instrumental, drums, bass, other
Audio format
16-bit, 44.1 kHz stereo WAV at the song's exact length
Billing
One flat charge per song; failed runs are not charged
Task ceiling
1,800 seconds from submission to a finished task
Labelling
Every stem carries a stem field in the task output
Structure
Two layers of separation from one callThe first split is voice against band: vocals on one track, the whole accompaniment on another. The second split works inside that accompaniment, pulling out drums and bass and leaving guitars, keys, pads and effects in other. In testing, drums, bass and other added back to the instrumental sample for sample, so you can work at whichever level the job needs without the layers disagreeing.
Alignment
Stems you can stack without nudgingAll five files start at the song's first sample and end at its last, at 44.1 kHz and 16-bit stereo. Drop them on a timeline at zero and they play back in sync; mute one and the rest still line up. That removes the fiddly re-alignment step that separated audio often needs before anyone can mix with it.
Cost
Priced by the song, not the minuteA two-minute single and a nine-minute album track cost the same, and all five stems are part of that one charge. Planning a catalogue job is therefore a count of songs, not a sum of runtimes. Songs over ten minutes are the one exception: an uploaded file that long is refused at submission, and a long track sent as a link fails during processing and is refunded.

What the Mureka Stem Separation API does well

  • A request with nothing to tune

    audio_url is the whole body, and the schema rejects any other key. There are no model variants, quality modes or stem counts to choose between, so two runs on the same file are asked to do exactly the same job. A misspelt field comes back as an invalid request straight away instead of quietly producing a different result.

  • Each file names its own stem

    The output is a list of five audio assets, and each one carries a stem value of vocals, instrumental, drums, bass or other. Map files by that value rather than by their position in the list, and the code that routes stems into your editor, store or mixer stays correct even if the ordering ever changes.

  • Clean vocal track in the gaps

    In instrumental-only passages of the test song, the vocal stem sat about 55 dB below the sung sections, so very little of the band bled into it. That makes the vocal track usable on its own for remixes, acapella edits and lyric work, where leftover drums or guitar would otherwise be the first thing a listener notices.

  • Fast enough to run inline

    A 60-second song finished in about ten seconds in testing, well inside the 1,800-second ceiling the task allows. That is quick enough to split a track while a user waits in an editor, and it means a batch of a few hundred songs is limited by how fast you queue them rather than by how long each one takes.

  • Uncompressed output, sized accordingly

    Stems come back as WAV rather than MP3, so nothing is lost a second time on the way out. The trade-off is size: each minute of a stem is roughly 10 MB, and a three-minute song returns about 160 MB across the five files. Download them promptly and store them where large audio is cheap to keep.

  • Refusals before the charge

    Formats the endpoint cannot read, such as WAV or FLAC, are turned away when the task is submitted, before anything is billed. A run that fails later is not charged either. The cost of a mistake in a batch is therefore a rejected row to fix, not a pile of paid runs to write off.

Who Mureka Stem Separation is for

  • Karaoke and sing-along products

    A backing track from any licensed song

    The instrumental stem is the song without its voices, at full length and full stereo, which is exactly what a sing-along player needs. Because the vocal stem comes back too, the same run can drive a guide-vocal toggle without a second request.

  • Producers and remixers

    Rebuilding a mix around new parts

    Keep vocals, bass and other, replace the drums with your own pattern, and the rest of the arrangement stays exactly where it was. The stems line up from the first sample, so the new part is the only thing that needs placing.

  • Musicians practising along

    Muting the part you want to play

    A drummer can mute the drums stem and play over the band; a bassist can do the same with bass. Separated stems make a practice track out of a finished recording without hunting for a karaoke version that may not exist.

  • Music apps and audio pipelines

    Stems as a step in a larger flow

    Because the request is one field and every output is labelled, separation slots into a pipeline as a single step: split, route each stem by its label, then hand the parts to transcription, analysis or editing tools further down the line.

How to call the Mureka Stem Separation API

  1. Prepare an MP3 or M4A, ten minutes or shorter

    Start from the best copy you have, because a heavily compressed file carries its artefacts into every stem. If your master is WAV or FLAC, convert it to a high-bitrate MP3 first; those formats are refused at submission.

  2. Send audio_url and keep the task id

    Upload the file or pass a direct link to it as audio_url, with an Idempotency-Key on the request so a retry after a network drop returns the original task instead of starting a second run. The response is a task record, not the stems.

  3. Wait for the task to finish

    Poll the task or register a callback URL. Short songs usually finish within seconds, but leave your client timeout well above that, since the task may take up to 1,800 seconds before it reports a final state.

  4. Route each file by its stem label

    When the task succeeds, read the five assets and use each one's stem field to decide where it goes. Download them promptly: the links are short-lived, and generated files are kept for a limited period, so your own storage is the lasting copy.

Mureka Stem Separation worked examples

  • Split a song from a direct link

    The whole request: one key, pointing at a file that answers with the audio itself. The task comes back first; the five WAV stems arrive on it once the run finishes.

    {"audio_url": "https://files.example.com/masters/demo-mix.mp3"}
  • A karaoke bed for a licensed track

    Send the song, then keep only the stem labelled instrumental for the player and the one labelled vocals for an optional guide track. Drums, bass and other can be discarded if the product never needs them.

    {"audio_url": "https://cdn.example.com/licensed/track-12.m4a"}
  • A catalogue run over a worklist

    Loop over a list of song URLs with one Idempotency-Key per row. Each song settles on its own, so a failed row is fixed and resent without rerunning the rest, and the budget for the batch is the number of songs times the per-song price.

Mureka Stem Separation API questions

Think of this endpoint as a fixed-shape splitter: one song in, the same five labelled stems out, one charge. Everything you control sits upstream — the quality of the file you send and whether it is under ten minutes — and downstream, where you decide which of the five stems to keep and where large WAV files will live. Get those two ends right and the call in the middle needs no tuning at all.

All Mureka models3