# facebookresearch/demucs

Code for the paper Hybrid Spectrogram and Waveform Source Separation

Repository: https://github.com/facebookresearch/demucs
Canonical: https://ross.abutalabs.com/products/demucs
Language: Python
License: MIT
License Family: permissive
Archived: true
Last push: 2024-04-24T20:25:19+00:00

## Health v2 (maintenance only)
Score: 10/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2504, "days_push": 861, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, archived
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 10359, forks 1586 (observed 2026-08-28T04:10:41.785334+00:00)

## What it is
Demucs is a state-of-the-art music source separation model from Meta AI that splits songs into stems like drums, bass, and vocals using a hybrid spectrogram/waveform architecture with Transformers. This original repository is no longer maintained; development continues in a fork at github.com/adefossez/demucs.

## Use cases
- separate vocals from a song
- extract drums and bass stems from music
- remove vocals to make a karaoke track
- split a music track into individual instrument stems
- run music source separation on audio files
- research hybrid spectrogram and waveform separation models

## When to choose
- you need high-quality stem separation of vocals, drums, and bass from songs
- you want a proven, widely used open-source separation model with pretrained weights
- you are doing audio research and want a reference implementation of hybrid separation models

## When to avoid
- you need active maintenance or new features - use the adefossez/demucs fork instead
- you need the sparse attention model from the paper, which is not released
- you need real-time separation on low-resource hardware without a GPU

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, audio-processing, deep-learning
- domain: machine-learning, deep-learning, media
- platform: python, cross-platform
- tags: music-source-separation, stem-separation, audio, u-net, transformer, facebook-research, vocal-removal, gpu

## Member repositories
- facebookresearch/demucs (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:41.785334+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:19:07.125871+00:00, confidence not recorded.
  - readme: https://github.com/facebookresearch/demucs (fetched 2026-08-28T04:10:41.785334+00:00, sha 108a47272d14)
- Data as of 2026-08-30T08:39:29.467469+00:00.