Reproduce “Noisy Target Training Meets Drifting for Weakly-Supervised Speech Enhancement” 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
Yoshiki Masuyama, Dominik Klement, Christoph Boeddeker, Julius Richter, Kohei Saijo, Gordon Wichern, Takahiro Edo, Jonathan Le Roux
The method combines noisy-target training with Sinkhorn drifting, a distribution-matching signal that uses unpaired clean speech to guide enhanced-speech features. On VoiceBank-DEMAND, it raises DNSMOS P.808 from 3.53 to 3.59, UTMOS from 3.81 to 3.93, and wide-band PESQ from 2.63 to 2.69, while reported SI-SDR means remain 18.58 dB. On CHiME-3, perceptual scores improve over NyTT+, but word error rate exceeds noisy speech.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.