Reproduce “Speaker Adaptation Using Spectro-Temporal Deep Features for Dysarthric and Elderly Speech Recognition” 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

Mengzhe Geng, Xurong Xie, Zi Ye, Tianzi Wang, Guinan Li, Shujie Hu, Xunying Liu, Helen Meng

The authors propose spectral-basis embeddings for adapting speech recognition to dysarthric and elderly speakers. They derive features from spectrograms using singular value decomposition and classifiers, then combine them with acoustic features. On UASpeech’s 16-speaker test set, the best system achieves 25.05% WER, versus 26.63% with matched i-vector adaptation. Temporal features do not consistently help.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.