Reproduce “Noise-Robust Forced Alignment for Spontaneous Speech” 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
Laurin Wagner, Mario Zusag
The authors adapt a classical HMM-based forced aligner for spontaneous speech by replacing MFCCs with frozen WavLM features and adding transcript units for fillers and vocal events. Trained on 149.1 hours of verbatim speech, the system reaches 20.6 ms word-boundary MAE on Buckeye and 76.9 ms on FluencyBank, outperforming tested alternatives there, but not on read-speech TIMIT.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.