{"url":"/sota/music-source-separation-on-musdb18","task":{"name":"Music Source Separation","url":"/task/music-source-separation","note":null},"dataset":{"name":"MUSDB18","url":"/dataset/musdb18"},"category":"Music","categories":["Music"],"category_note":null,"description":"Music source separation is the task of decomposing music into its constitutive components, e. g., yielding separated stems for the vocals, bass, and drums.\r\n\r\n<span style=\"color:grey; opacity: 0.6\">( Image credit: [SigSep](https://github.com/sigsep ) )</span>","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["SDR (avg)","SDR (vocals)","SDR (drums)","SDR (bass)","SDR (other)"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"SDR (avg)":null,"SDR (vocals)":null,"SDR (drums)":null,"SDR (bass)":null,"SDR (other)":null}},"counts":{"rows":27,"rows_with_code":25,"rows_with_paper_page":27,"rows_dated":27,"rows_using_additional_data":9},"rows":[{"rank_in_archive_order":1,"model":"Sparse HT Demucs (fine tuned)","metrics":{"SDR (avg)":"9.20","SDR (bass)":"10.47","SDR (drums)":"10.83","SDR (other)":"6.41","SDR (vocals)":"9.37"},"uses_additional_data":true,"paper_date":"2022-11-15","paper":"/paper/hybrid-transformers-for-music-source","paper_url":"https://arxiv.org/abs/2211.08553v1","paper_title":"Hybrid Transformers for Music Source Separation","code":"https://github.com/facebookresearch/demucs","n_code_links":2,"syntology":{"n_ran":0,"n_unverified":10,"n_samples":10,"n_pointer_only_licence":10}},{"rank_in_archive_order":2,"model":"Hybrid Transformer Demucs (f.t.)","metrics":{"SDR (avg)":"9.00","SDR (bass)":"9.78","SDR (drums)":"10.08","SDR (other)":"6.42","SDR (vocals)":"9.20"},"uses_additional_data":true,"paper_date":"2022-11-15","paper":"/paper/hybrid-transformers-for-music-source","paper_url":"https://arxiv.org/abs/2211.08553v1","paper_title":"Hybrid Transformers for Music Source Separation","code":"https://github.com/facebookresearch/demucs","n_code_links":2,"syntology":{"n_ran":0,"n_unverified":10,"n_samples":10,"n_pointer_only_licence":10}},{"rank_in_archive_order":3,"model":"Band-Split RNN (semi-sup.)","metrics":{"SDR (avg)":"8.97","SDR (bass)":"8.16","SDR (drums)":"10.15","SDR (other)":"7.08","SDR (vocals)":"10.47"},"uses_additional_data":true,"paper_date":"2022-09-30","paper":"/paper/music-source-separation-with-band-split-rnn","paper_url":"https://arxiv.org/abs/2209.15174v1","paper_title":"Music Source Separation with Band-split RNN","code":"https://github.com/crlandsc/music-demixing-with-band-split-rnn","n_code_links":3,"syntology":{"n_ran":0,"n_unverified":10,"n_samples":10,"n_pointer_only_licence":0}},{"rank_in_archive_order":4,"model":"TFC-TDF-UNet (v3)","metrics":{"SDR (avg)":"8.34","SDR (bass)":"8.45","SDR (drums)":"8.44","SDR (other)":"6.86","SDR (vocals)":"9.59"},"uses_additional_data":false,"paper_date":"2023-06-15","paper":"/paper/sound-demixing-challenge-2023-music-demixing","paper_url":"https://arxiv.org/abs/2306.09382v3","paper_title":"Sound Demixing Challenge 2023 Music Demixing Track Technical Report: TFC-TDF-UNet v3","code":"https://github.com/kuielab/sdx23","n_code_links":1,"syntology":null},{"rank_in_archive_order":5,"model":"Band-Split RNN","metrics":{"SDR (avg)":"8.23","SDR (bass)":"7.51","SDR (drums)":"8.58","SDR (other)":"6.62","SDR (vocals)":"10.21"},"uses_additional_data":false,"paper_date":"2022-09-30","paper":"/paper/music-source-separation-with-band-split-rnn","paper_url":"https://arxiv.org/abs/2209.15174v1","paper_title":"Music Source Separation with Band-split RNN","code":"https://github.com/crlandsc/music-demixing-with-band-split-rnn","n_code_links":3,"syntology":{"n_ran":0,"n_unverified":10,"n_samples":10,"n_pointer_only_licence":0}},{"rank_in_archive_order":6,"model":"Hybrid Demucs","metrics":{"SDR (avg)":"7.72","SDR (bass)":"8.67","SDR (drums)":"8.58","SDR (other)":"5.59","SDR (vocals)":"8.04"},"uses_additional_data":false,"paper_date":"2021-11-05","paper":"/paper/hybrid-spectrogram-and-waveform-source","paper_url":"https://arxiv.org/abs/2111.03600v3","paper_title":"Hybrid Spectrogram and Waveform Source Separation","code":"https://github.com/facebookresearch/demucs","n_code_links":1,"syntology":null},{"rank_in_archive_order":7,"model":"KUIELab-MDX-Net","metrics":{"SDR (avg)":"7.54","SDR (bass)":"7.86","SDR (drums)":"7.33","SDR (other)":"5.95","SDR (vocals)":"9.00"},"uses_additional_data":false,"paper_date":"2021-11-24","paper":"/paper/kuielab-mdx-net-a-two-stream-neural-network","paper_url":"https://arxiv.org/abs/2111.12203v1","paper_title":"KUIELab-MDX-Net: A Two-Stream Neural Network for Music Demixing","code":"https://github.com/kuielab/mdx-net","n_code_links":1,"syntology":null},{"rank_in_archive_order":8,"model":"CDE-HTCN","metrics":{"SDR (avg)":"6.89","SDR (bass)":"7.92","SDR (drums)":"7.33","SDR (other)":"4.92","SDR (vocals)":"7.37"},"uses_additional_data":false,"paper_date":"2022-06-30","paper":"/paper/hierarchic-temporal-convolutional-network","paper_url":"https://ieeexplore.ieee.org/document/9812509","paper_title":"Hierarchic Temporal Convolutional Network With Cross-Domain Encoder for Music Source Separation","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":9,"model":"Attentive-MultiResUNet","metrics":{"SDR (avg)":"6.81","SDR (bass)":"5.88","SDR (drums)":"7.63","SDR (other)":"5.14","SDR (vocals)":"8.57"},"uses_additional_data":false,"paper_date":"2022-11-14","paper":"/paper/an-efficient-short-time-discrete-cosine","paper_url":"https://ieeexplore.ieee.org/document/9947051?source=authoralert","paper_title":"An Efficient Short-Time Discrete Cosine Transform and Attentive MultiResUNet Framework for Music Source Separation","code":"https://github.com/tsgouros09/Attentive-MultiResUNet","n_code_links":1,"syntology":null},{"rank_in_archive_order":10,"model":"DEMUCS (extra)","metrics":{"SDR (avg)":"6.79","SDR (bass)":"7.60","SDR (drums)":"7.58","SDR (other)":"4.69","SDR (vocals)":"7.29"},"uses_additional_data":true,"paper_date":"2019-11-27","paper":"/paper/music-source-separation-in-the-waveform-1","paper_url":"https://arxiv.org/abs/1911.13254v2","paper_title":"Music Source Separation in the Waveform Domain","code":"https://github.com/facebookresearch/demucs","n_code_links":1,"syntology":null},{"rank_in_archive_order":11,"model":"CWS-PResUNet","metrics":{"SDR (avg)":"6.77","SDR (bass)":"5.93","SDR (drums)":"6.38","SDR (other)":"5.84","SDR (vocals)":"8.92"},"uses_additional_data":false,"paper_date":"2021-12-09","paper":"/paper/cws-presunet-music-source-separation-with","paper_url":"https://arxiv.org/abs/2112.04685v1","paper_title":"CWS-PResUNet: Music Source Separation with Channel-wise Subband Phase-aware ResUNet","code":"https://github.com/haoheliu/2021-ismir-mss-challenge-cws-presunet","n_code_links":1,"syntology":null},{"rank_in_archive_order":12,"model":"D3Net","metrics":{"SDR (avg)":"6.68","SDR (bass)":"6.20","SDR (drums)":"7.36","SDR (other)":"5.37","SDR (vocals)":"7.80"},"uses_additional_data":true,"paper_date":"2020-10-05","paper":"/paper/d3net-densely-connected-multidilated-densenet","paper_url":"https://arxiv.org/abs/2010.01733v4","paper_title":"D3Net: Densely connected multidilated DenseNet for music source separation","code":"https://github.com/sony/ai-research-code/tree/master/d3net/music-source-separation","n_code_links":1,"syntology":null},{"rank_in_archive_order":13,"model":"DLMNet","metrics":{"SDR (avg)":"6.47","SDR (bass)":"7.29","SDR (drums)":"7.05","SDR (other)":"4.62","SDR (vocals)":"6.91"},"uses_additional_data":false,"paper_date":"2023-08-20","paper":"/paper/contrastive-learning-based-deep-latent","paper_url":"https://www.isca-archive.org/interspeech_2023/kim23i_interspeech.pdf","paper_title":"Contrastive Learning based Deep Latent Masking for Music Source Separation","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":14,"model":"Conv-TasNet (extra)","metrics":{"SDR (avg)":"6.32","SDR (bass)":"7.00","SDR (drums)":"7.11","SDR (vocals)":"6.74"},"uses_additional_data":true,"paper_date":"2018-09-20","paper":"/paper/tasnet-surpassing-ideal-time-frequency","paper_url":"https://arxiv.org/abs/1809.07454v3","paper_title":"Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Speech Separation","code":"https://github.com/facebookresearch/demucs","n_code_links":17,"syntology":{"n_ran":7,"n_unverified":29,"n_samples":36,"n_pointer_only_licence":16}},{"rank_in_archive_order":15,"model":"UMXL","metrics":{"SDR (avg)":"6.316","SDR (bass)":"6.015","SDR (drums)":"7.148","SDR (other)":"4.889","SDR (vocals)":"7.213"},"uses_additional_data":true,"paper_date":"2019-09-08","paper":"/paper/open-unmix-a-reference-implementation-for","paper_url":"https://joss.theoj.org/papers/10.21105/joss.01667","paper_title":"Open-Unmix - A Reference Implementation for Music Source Separation","code":"https://github.com/sigsep/open-unmix-pytorch","n_code_links":1,"syntology":null},{"rank_in_archive_order":16,"model":"DEMUCS","metrics":{"SDR (avg)":"6.28","SDR (bass)":"7.01","SDR (drums)":"6.86","SDR (other)":"4.42","SDR (vocals)":"6.84"},"uses_additional_data":false,"paper_date":"2019-11-27","paper":"/paper/music-source-separation-in-the-waveform-1","paper_url":"https://arxiv.org/abs/1911.13254v2","paper_title":"Music Source Separation in the Waveform Domain","code":"https://github.com/facebookresearch/demucs","n_code_links":1,"syntology":null},{"rank_in_archive_order":17,"model":"TAK2","metrics":{"SDR (avg)":"6.04","SDR (bass)":"5.40","SDR (drums)":"6.81","SDR (other)":"4.80","SDR (vocals)":"7.16"},"uses_additional_data":true,"paper_date":"2018-05-07","paper":"/paper/mmdenselstm-an-efficient-combination-of","paper_url":"https://arxiv.org/abs/1805.02410v2","paper_title":"MMDenseLSTM: An efficient combination of convolutional and recurrent neural networks for audio source separation","code":"https://github.com/tsurumeso/vocal-remover","n_code_links":1,"syntology":{"n_ran":0,"n_unverified":7,"n_samples":7,"n_pointer_only_licence":0}},{"rank_in_archive_order":18,"model":"D3Net","metrics":{"SDR (avg)":"6.01","SDR (bass)":"5.25","SDR (drums)":"7.01","SDR (other)":"4.53","SDR (vocals)":"7.24"},"uses_additional_data":false,"paper_date":"2020-10-05","paper":"/paper/d3net-densely-connected-multidilated-densenet","paper_url":"https://arxiv.org/abs/2010.01733v4","paper_title":"D3Net: Densely connected multidilated DenseNet for music source separation","code":"https://github.com/sony/ai-research-code/tree/master/d3net/music-source-separation","n_code_links":1,"syntology":null},{"rank_in_archive_order":19,"model":"Spleeter (MWF)","metrics":{"SDR (avg)":"5.91","SDR (bass)":"5.51","SDR (drums)":"6.71","SDR (other)":"4.02","SDR (vocals)":"6.86"},"uses_additional_data":true,"paper_date":"2019-11-04","paper":"/paper/spleeter-a-fast-and-state-of-the-art-music","paper_url":"https://archives.ismir.net/ismir2019/latebreaking/","paper_title":"Spleeter: A Fast And State-of-the Art Music Source Separation Tool With Pre-trained Models","code":"https://github.com/deezer/spleeter","n_code_links":3,"syntology":null},{"rank_in_archive_order":20,"model":"LaSAFT+GPoCM","metrics":{"SDR (avg)":"5.88","SDR (bass)":"5.63","SDR (drums)":"5.68","SDR (other)":"4.87","SDR (vocals)":"7.33"},"uses_additional_data":false,"paper_date":"2020-10-22","paper":"/paper/lasaft-latent-source-attentive-frequency","paper_url":"https://arxiv.org/abs/2010.11631v2","paper_title":"LaSAFT: Latent Source Attentive Frequency Transformation for Conditioned Source Separation","code":"https://github.com/ws-choi/Conditioned-Source-Separation-LaSAFT","n_code_links":1,"syntology":{"n_ran":1,"n_unverified":2,"n_samples":3,"n_pointer_only_licence":0}},{"rank_in_archive_order":21,"model":"X-UMX","metrics":{"SDR (avg)":"5.79","SDR (bass)":"5.43","SDR (drums)":"6.47","SDR (other)":"4.64","SDR (vocals)":"6.61"},"uses_additional_data":false,"paper_date":"2020-10-08","paper":"/paper/all-for-one-and-one-for-all-improving-music","paper_url":"https://arxiv.org/abs/2010.04228v4","paper_title":"All for One and One for All: Improving Music Separation by Bridging Networks","code":"https://github.com/asteroid-team/asteroid","n_code_links":5,"syntology":{"n_ran":1,"n_unverified":9,"n_samples":10,"n_pointer_only_licence":0}},{"rank_in_archive_order":22,"model":"Conv-TasNet","metrics":{"SDR (avg)":"5.73","SDR (bass)":"5.66","SDR (drums)":"6.08","SDR (other)":"4.37","SDR (vocals)":"6.81"},"uses_additional_data":false,"paper_date":"2018-09-20","paper":"/paper/tasnet-surpassing-ideal-time-frequency","paper_url":"https://arxiv.org/abs/1809.07454v3","paper_title":"Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Speech Separation","code":"https://github.com/facebookresearch/demucs","n_code_links":17,"syntology":{"n_ran":7,"n_unverified":29,"n_samples":36,"n_pointer_only_licence":16}},{"rank_in_archive_order":23,"model":"Sams-Net","metrics":{"SDR (avg)":"5.65","SDR (bass)":"5.25","SDR (drums)":"6.63","SDR (other)":"4.09","SDR (vocals)":"6.61"},"uses_additional_data":false,"paper_date":"2019-09-12","paper":"/paper/tf-attention-net-an-end-to-end-neural-network","paper_url":"https://arxiv.org/abs/1909.05746v4","paper_title":"Sams-Net: A Sliced Attention-based Neural Network for Music Source Separation","code":"https://github.com/chenjiawei5/Contribute_Paper","n_code_links":1,"syntology":null},{"rank_in_archive_order":24,"model":"Meta-TasNet","metrics":{"SDR (avg)":"5.52","SDR (bass)":"5.58","SDR (drums)":"5.91","SDR (other)":"4.19","SDR (vocals)":"6.40"},"uses_additional_data":false,"paper_date":"2020-02-17","paper":"/paper/meta-learning-extractors-for-music-source","paper_url":"https://arxiv.org/abs/2002.07016v1","paper_title":"Meta-learning Extractors for Music Source Separation","code":"https://github.com/pfnet-research/meta-tasnet","n_code_links":1,"syntology":null},{"rank_in_archive_order":25,"model":"UMX","metrics":{"SDR (avg)":"5.33","SDR (bass)":"5.23","SDR (drums)":"5.73","SDR (other)":"4.02","SDR (vocals)":"6.32"},"uses_additional_data":false,"paper_date":"2019-09-08","paper":"/paper/open-unmix-a-reference-implementation-for","paper_url":"https://joss.theoj.org/papers/10.21105/joss.01667","paper_title":"Open-Unmix - A Reference Implementation for Music Source Separation","code":"https://github.com/sigsep/open-unmix-pytorch","n_code_links":1,"syntology":null},{"rank_in_archive_order":26,"model":"Wavenet","metrics":{"SDR (avg)":"3.5","SDR (bass)":"2.49","SDR (drums)":"4.60","SDR (other)":"0.54","SDR (vocals)":"3.46"},"uses_additional_data":false,"paper_date":"2018-10-29","paper":"/paper/end-to-end-music-source-separation-is-it","paper_url":"https://arxiv.org/abs/1810.12187v2","paper_title":"End-to-end music source separation: is it possible in the waveform domain?","code":"https://github.com/francesclluis/source-separation-wavenet","n_code_links":2,"syntology":{"n_ran":1,"n_unverified":0,"n_samples":1,"n_pointer_only_licence":1}},{"rank_in_archive_order":27,"model":"STL2","metrics":{"SDR (avg)":"3.23","SDR (bass)":"3.21","SDR (drums)":"4.22","SDR (other)":"2.25","SDR (vocals)":"3.25"},"uses_additional_data":false,"paper_date":"2018-06-08","paper":"/paper/wave-u-net-a-multi-scale-neural-network-for","paper_url":"http://arxiv.org/abs/1806.03185v1","paper_title":"Wave-U-Net: A Multi-Scale Neural Network for End-to-End Audio Source Separation","code":"https://github.com/f90/Wave-U-Net","n_code_links":10,"syntology":{"n_ran":1,"n_unverified":18,"n_samples":19,"n_pointer_only_licence":1}}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,264 of the 9,581 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9581,"papers_checked":6264,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":3316},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-24T18:15:14+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":11,"rows_with_any_sample_ran":6,"distinct_papers_with_graph_line":8,"distinct_papers_with_any_sample_ran":5,"samples_over_distinct_papers":{"n_ran":11,"n_unverified":85,"n_samples":96,"n_pointer_only_licence":28,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":18,"n_unverified":134,"n_samples":152,"n_pointer_only_licence":54,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}