Megadose AI progress, ranked and analyzed.

Do speech foundation models really learn words?

· ArXiv · AI/CL/LG ·
HuBERT and wav2vec 2.0 appear to encode word identity beyond just phoneme patterns in later layers.

The paper tests whether strong word discrimination in speech foundation models is only a byproduct of encoding word form. Huo and Dunbar use residualization to partial out phoneme information, then check what remains. They report that later-layer representations still preserve words with reasonable fidelity, independent of local phonetic content. The same disentangling step can improve word discovery tasks by making higher-order linguistic information clearer. ArXiv · AI/CL/LG's note

score 4

Categories: Research