Problem: deciding whether to depend on Demucs v4 for stem separation. Conditions: repository facebookresearch/demucs. Method: the README states the repo is no longer maintained by its author (a fork is named for bug fixes), v4 is Hybrid Transformer Demucs (default htdemucs, trained on MUSDB HQ plus 800 extra songs), 9.00 dB SDR on the MUSDB HQ test set, 9.20 dB for the sparse-attention fine-tuned model not released; htdemucs_ft is about four times slower; htdemucs_6s adds guitar and piano. Code licence MIT (Meta). Failure modes: unmaintained code may break with new torch/torchaudio versions (inference); the 6-source model has weak piano. Limitations: the licence covers code; the README does not describe training-data terms for the 800 songs or output rights. Version/date: fetched 2026-10-23.
Problem: deciding whether to depend on Demucs v4 for stem separation. Conditions: repository facebookresearch/demucs. Method: the README states the repo is no longer maintained by its author (a fork is named for bug fixes), v4 is Hybrid Transformer Demucs (default htdemucs, trained on MUSDB HQ plus 800 extra songs), 9.00 dB SDR on the MUSDB HQ test set, 9.20 dB for the sparse-attention fine-tuned model not released; htdemucs_ft is about four times slower; htdemucs_6s adds guitar and piano. Code licence MIT (Meta). Failure modes: unmaintained code may break with new torch/torchaudio versions (inference); the 6-source model has weak piano. Limitations: the licence covers code; the README does not describe training-data terms for the 800 songs or output rights. Version/date: fetched 2026-10-23.
Curator qualifications: All cited documentation excerpts located on the live pages by curator GET; no video sources. Correction folded 2026-10-24 from C3-B724-06 (curator-reviewed discovery audit): Problem. C3-B708-06 attributes 9.00 dB SDR to the default htdemucs model and says the 9.20 dB sparse fine-tuned model is not released. Correction. The README sentence about 9.00 dB does not name which model it describes, and the README lists htdemucs as the default and htdemucs_ft as a fine-tuned version. The Demucs paper's table gives HT Demucs trained with 800 extra songs 8.80 dB all-source SDR and the fine-tuned model 9.00 dB, so 9.00 belongs to the fine-tuned model (the sibling finding C3-B708-07 uses 8.80 for the default). No cited source states the sparse model is unreleased; the README's model list simply does not include it, which is an inference. Limitations. SDR is an average on MUSDB18-HQ, not a statement about any given mix. Version/date. README on main; arXiv 2211.08553.
Limitations: Maintenance status may change; fork was not read. Training-data licence of the extra 800 songs is undisclosed in the README. Corrected by C3-B724-06; where the correction disagrees, the correction governs.
Evidence status: statements located in the cited sources by a correlated desk review (Claude); not executed; not independently reviewed.
[sourced_fact] Problem: deciding whether to depend on Demucs v4 for stem separation. Conditions: repository facebookresearch/demucs. Method: the README states the repo is no longer maintained by its author (a fork is named for bug fixes), v4 is Hybrid Transformer Demucs (default htdemucs, trained on MUSDB HQ plus 800 extra songs), 9.00 dB SDR on the MUSDB HQ test set, 9.20 dB for the sparse-attention fine-tuned model not released; htdemucs_ft is about four times slower; htdemucs_6s adds guitar and piano. Code licence MIT (Meta). Failure modes: unmaintained code may break with new torch/torchaudio versions (inference); the 6-source model has weak piano. Limitations: the licence covers code; the README does not describe training-data terms for the 800 songs or output rights. Version/date: fetched 2026-10-23. · locator: header; v4 description; License; LICENSE headsource
Rights
component
license
hosting
redistribution
modification
training
Evidence and notes
documentation
unknown
permitted
permitted
unknown
unknown
Applies to this original summary only, not to the cited sources.