ENDEMenu
All tools

BLVCK.FRST / Stem laboratory

Stem Extractor

Four stems from one file – separated in your browser, without upload or account.

008Stem Extractor
How it works
  1. 01

    Choose or drop an audio file and press “Separate stems”. Supported formats depend on your browser; WAV, MP3 and FLAC are a good start. Maximum 300 MB per file, at least one second.

  2. 02

    The separation uses the Open-Unmix UMX-HQ model (four stems: vocals, drums, bass, other). The first time, your browser downloads about 71 MB of model files plus a 15 MB runtime (ONNX Runtime Web) from blvckfrst.com; after that they come from the browser cache. “Delete model files” removes them.

  3. 03

    The file is processed in segments of about 14 seconds in a background worker; the progress shows the segment and, after the first one, the estimated time left. “Cancel” (or Escape) stops at once.

  4. 04

    Duration: on a current desktop roughly a few seconds per minute of audio and more on phones. For tracks of about ten minutes or longer the tool shows a warning: the separation can take long and needs a lot of memory.

  5. 05

    Each stem has mute (M), solo (S) and volume; Play plays all four in sync, Space plays/pauses, the bar or a click into the display jumps. Mute, solo and volume only affect listening, not the export.

  6. 06

    Export: each stem as WAV (44.1 kHz, 16-bit, stereo) or all four as one ZIP. Files are created locally and saved as a download.

  7. 07

    Separation is an estimate: results can contain artefacts and bleed between stems. Mono files are treated as stereo with two identical channels; other sample rates are converted to 44.1 kHz.

  8. 08

    Needs a current browser with WebAssembly, module workers and Web Audio (e.g. current Chrome, Edge, Firefox or Safari 16.4+). Long files need a lot of memory, especially on phones.

Runs in your browser. This tool runs locally in your browser. Nothing is uploaded; there is no account.

Your audio file is decoded and separated locally in your browser (Web Audio, a local worker). It is not uploaded. Only the model files are downloaded from blvckfrst.com – no audio is sent. After you start a separation, your browser keeps the model files in its cache so they are not downloaded again; “Delete model files” removes them. No settings are stored.

Privacy: browser tools

Terms of use

The Stem Extractor uses the Open-Unmix UMX-HQ model (four stems: vocals, drums, bass, other). Separation runs only in your browser; your audio file is not uploaded.

Separation quality comes with no warranty. Results may contain artefacts, be incomplete or unsuitable for a given track. The tool is provided as it is.

You may only process audio files for which you hold the necessary rights (e.g. your own recordings or cleared tracks). The tool does not grant you rights in third-party recordings and does not replace a licence or clearance. You are responsible for lawful use of the input and of the exported stems.

Exports (e.g. WAV stems) are created locally on your device. Whether and how you reuse them is up to you — including any third-party rights in the original.

Credits

Open-Unmix (UMX-HQ)
This tool uses the Open-Unmix UMX-HQ music source separation model.
Authors: Fabian-Robert Stöter, Antoine Liutkus et al.
Weights: Zenodo, DOI 10.5281/zenodo.3370489, MIT licence.
Code: sigsep/open-unmix-pytorch, MIT licence.
Citation: Stöter et al., “Open-Unmix – A Reference Implementation for Music Source Separation”, Journal of Open Source Software (2019), DOI 10.21105/joss.01667.
Licence texts: Open-Unmix (MIT) · ONNX Runtime Web (MIT)