Bayesian Subspace Hidden Markov Model for Acoustic Unit Discovery

Bayesian Subspace Hidden Markov Model for Acoustic Unit Discovery

Paper in proceedings (conference paper)

This work tackles the problem of learning a set of language specificacoustic units from unlabeled speech recordings given aset of labeled recordings from other languages. Our approachmay be described by the following two steps procedure: firstthe model learns the notion of acoustic units from the labelleddata and then the model uses its knowledge to find new acousticunits on the target language. We implement this processwith the Bayesian Subspace Hidden Markov Model (SHMM), amodel akin to the Subspace Gaussian Mixture Model (SGMM)where each low dimensional embedding represents an acousticunit rather than just a HMMs state. The subspace is trainedon 3 languages from the GlobalPhone corpus (German, Polishand Spanish) and the AUs are discovered on the TIMIT corpus.Results, measured in equivalent Phone Error Rate, show thatthis approach significantly outperforms previous HMM basedacoustic units discovery systems and compares favorably withthe Variational Auto Encoder-HMM.

This work tackles the problem of learning a set of language specificacoustic units from unlabeled speech recordings given aset of labeled recordings from other languages. Our approachmay be described by the following two steps procedure: firstthe model learns the notion of acoustic units from the labelleddata and then the model uses its knowledge to find new acousticunits on the target language. We implement this processwith the Bayesian Subspace Hidden Markov Model (SHMM), amodel akin to the Subspace Gaussian Mixture Model (SGMM)where each low dimensional embedding represents an acousticunit rather than just a HMMs state. The subspace is trainedon 3 languages from the GlobalPhone corpus (German, Polishand Spanish) and the AUs are discovered on the TIMIT corpus.Results, measured in equivalent Phone Error Rate, show thatthis approach significantly outperforms previous HMM basedacoustic units discovery systems and compares favorably withthe Variational Auto Encoder-HMM.

Bayesian Inference, Hidden Markov Model,Subspace Model, Variational Bayes, Low-resource languages,Acoustic Unit Discovery

Bayesian Inference, Hidden Markov Model,Subspace Model, Variational Bayes, Low-resource languages,Acoustic Unit Discovery

ONDEL YANG, L.; VYDANA, H.; BURGET, L.; ČERNOCKÝ, J.

2020

15.09.2019

International Speech Communication Association

Graz

Proceedings of Interspeech 2019

1990-9772

Proceedings of Interspeech

2019

9

French Republic

261

265

5

https://www.isca-speech.org/archive/Interspeech_2019/pdfs/2224.pdf

http://hdl.handle.net/

ondel_is2019_192224