Reproduce “WeSep: A Scalable and Flexible Toolkit Towards Generalizable Target Speaker Extraction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shuai Wang, Ke Zhang, Shaoxiong Lin, Junjie Li, Xuefei Wang, Meng Ge, Jianwei Yu, Yanmin Qian and 1 more
WeSep is a toolkit for target speaker extraction that combines extraction backbones with speaker-cue modeling, online mixture simulation, and training support. It creates mixtures during training from speaker recordings, with optional noise and reverberation. BSRNN trained on VoxCeleb1 scored 10.18 dB SI-SDR on AISHELL2Mix, versus 8.17 dB for Libri2Mix-train-360; on Libri2Mix, scores were 16.18 and 16.57 dB.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.