Research map / Label-efficient and robust learning / Recognition and long-tail
UNSSOR: Unsupervised Neural Speech Separation by Leveraging Over-determined Training Mixtures
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add ISMS loss to the mixture-constraint objective (ablation) | UNSSOR · SMS-WSJ test, 2-speaker separation, 6-channel input and loss | SDR (dB): 11.9397 -> 15.557143978642484 | Helped |
| Add ISMS loss to the mixture-constraint objective (ablation) | UNSSOR · SMS-WSJ test, 2-speaker separation, 3-channel input and loss | SDR (dB): 9.39789025 -> 15.384367245 | Helped |
| Apply correlation-based frequency alignment after training with mixture-constraint loss (ablation) | UNSSOR · SMS-WSJ test, 2-speaker separation, 6-channel input and loss | SDR (dB): 11.9397 -> 15.700158920758607 | Helped |
| Use oracle frequency alignment instead of unaligned outputs (ablation) | UNSSOR trained with mixture-constraint loss · SMS-WSJ test, 2-speaker separation, 6-channel input and loss | SDR (dB): 11.9397 -> 15.83159 | Helped |
| Use six microphones rather than three for the multi-microphone training loss (ablation) | Monaural-input UNSSOR · SMS-WSJ test, 2-speaker separation | SDR (dB): 12.029486123579598 -> 12.469372936138235 | Helped |
Closest papers by meaning
- Separate and Reconstruct: Asymmetric Encoder-Decoder for Speech Separation (NeurIPS 2024 · similarity 0.89)
- Unsupervised Sound Separation Using Mixture Invariant Training (NeurIPS 2020 · similarity 0.89)
- Voice Separation with an Unknown Number of Multiple Speakers (ICML 2020 · similarity 0.89)
- ArrayDPS: Unsupervised Blind Speech Separation with a Diffusion Prior (ICML 2025 · similarity 0.87)
- ZeroSep: Separate Anything in Audio with Zero Training (NeurIPS 2025 · similarity 0.84)
- Weakly-supervised Audio Separation via Bi-modal Semantic Similarity (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.