Benchmarks and leaderboards for sound demixing tasks

TL;DR

This work introduces two new benchmarks, Synth MVSep and Multisong MVSep, to assess music sound demixing beyond traditional datasets and mitigate overfitting. It surveys popular models (e.g., Demucs, MDX-Net, UVR) and demonstrates that ensembles tailored to vocal versus instrumental stems yield the best performance. Using SDR-based evaluation on dynamic leaderboards, the authors show that UVR-MDX variants excel at vocals while Demucs HT variants dominate instrumental separation, and they validate these findings through SDX23, including cinematic and music tracks with data-constraint leaderboards A–C. The proposed ensembling framework achieved top results in SDX23 and is openly available on GitHub, enhancing reproducibility and practical impact for real-world audio separation and remixing applications.

Abstract

Music demixing is the task of separating different tracks from the given single audio signal into components, such as drums, bass, and vocals from the rest of the accompaniment. Separation of sources is useful for a range of areas, including entertainment and hearing aids. In this paper, we introduce two new benchmarks for the sound source separation tasks and compare popular models for sound demixing, as well as their ensembles, on these benchmarks. For the models' assessments, we provide the leaderboard at https://mvsep.com/quality_checker/, giving a comparison for a range of models. The new benchmark datasets are available for download. We also develop a novel approach for audio separation, based on the ensembling of different models that are suited best for the particular stem. The proposed solution was evaluated in the context of the Music Demixing Challenge 2023 and achieved top results in different tracks of the challenge. The code and the approach are open-sourced on GitHub.