Sony researchers apply flow matching to four-stem stereo music source separation in FLOMSS
Shuyang Cui · hf · 2026-10-26
Sony Group researchers published FLOMSS (arXiv, Oct 14, 2026), the first systematic study of flow matching for full-band stereo music source separation (MSS).
- Task: separate songs into four stems — Vocals / Bass / Drums / Other — in full-band stereo
- Ablates prediction targets, flow paths, single- vs multi-source modeling, and consistency constraints
- Trained/evaluated on MUSDB18-HQ with spectral and embedding-based metrics from pretrained audio encoders, plus a listening test
- Proposed model is competitive on both objective and subjective evaluations; code open-sourced
The paper fills the gap of generative flow matching for predefined-stem MSS, previously studied mainly for open-domain audio separation.
More from Research
- LessWrong post argues FDT is difficult-to-impossible to specify due to logical counterfactuals — jessi_cata · 2026-10-26
- Bitter Lesson misread: scaling won't brute-force biology, researcher argues — anshulkundaje · 2026-10-26
- Chain Theory: rebuilding physics from quantum information, AI finds 137 fine-structure-constant expressions — skdh · 2026-10-26
- HKU/ZJU/Fudan with Lightspeed launch HiSTEVE: hierarchical VLA agent playing Minecraft via native mouse and keyboard — tencent · 2026-10-26
- Zhejiang University and Tencent propose Easy-3Dfy: decoupling object generation and layout for single-image 3D scene construction — Ziang Zhang · 2026-10-26
- One week: 80-year-old cells reset to under 20, AI designing working proteins — rand_longevity · 2026-10-26