Reproduce “Word-Free Spoken Language Understanding for Mandarin-Chinese” 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

Zhiyuan Guo, Yuexin Li, Guo Chen, Xingyu Chen, Akshat Gupta

The authors test whether Mandarin intent can be classified from natural speech without a word transcript. Allosaurus converts audio into phonetic units, which a Transformer classifies; the model can also be pretrained on phone sequences. On the restructured CATSLU test set, it scores 0.53 macro-F1 and 0.61 accuracy, below a CNN+LSTM baseline at 0.59 and 0.63, respectively.

Summary by alphaXiv.

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