Learning Word Embeddings: Unsupervised Methods for Fixed-size Representations of Variable-length Speech Segments

Nils Holzenberger (1) , Mingxing Du (2) , Julien Karadayi (3, 4) , Rachid Riad (2, 4) , Emmanuel Dupoux (2, 4)

1 JHU - Johns Hopkins University
2 LSCP - Laboratoire de sciences cognitives et psycholinguistique
3 EHESS - École des hautes études en sciences sociales
4 CoML - Apprentissage machine et développement cognitif

Fixed-length embeddings of words are very useful for a variety of tasks in speech and language processing. Here we systematically explore two methods of computing fixed-length embeddings for variable-length sequences. We evaluate their susceptibility to phonetic and speaker-specific variability on English, a high resource language and Xitsonga, a low resource language, using two evaluation metrics: ABX word discrimination and ROC-AUC on same-different phoneme n-grams. We show that a simple downsampling method supplemented with length information can outperform the variable-length input feature representation on both evaluations. Recurrent autoencoders, trained without supervision, can yield even better results at the expense of increased computational complexity.

Mots clés: unsupervised speech processing; audio word embeddings; ABX discrimination; same-different classification; representation learning

Domaines: Linguistique; Sciences cognitives

Nils Holzenberger, Mingxing Du, Julien Karadayi, Rachid Riad, Emmanuel Dupoux. Learning Word Embeddings: Unsupervised Methods for Fixed-size Representations of Variable-length Speech Segments. Interspeech 2018, Sep 2018, Hyderabad, India. ⟨10.21437/Interspeech.2018-2364⟩. ⟨hal-01888708⟩