Evaluation of linear regression for speaker adaptation in HMM-based articulatory movements estimation

Hao Li, Jianhua Tao, Yang Wang
2015 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)  
Articulatory movements and acoustic features are important descriptions of speech and both of them contain speakerspecific components. This paper is to evaluate if the combination of the two features can improve the performance of speaker adaptation for articulatory movements estimation. HMM-based systems are built with single-stream and multistream, independent clustering and shared clustering structures. The speaker adaptation is realized in stream-independent structure and shared adaptation
more » ... shared adaptation structure. Constrained maximum likelihood linear regression method is used for the speaker-adaptive transformation. The experimental results indicate that the sharing of the speaker-adaptive transformation of the articulatory feature stream and acoustic feature stream can improve the estimation accuracy of articulatory movements in the adaptation system. The multi-stream system with shared clustering and shared adaptation yields the best result among all the tested systems. Index Terms: speaker adaptation, acoustic-to-articulatory inversion, hidden Markov models,
doi:10.1109/icassp.2015.7178911 dblp:conf/icassp/LiTW15 fatcat:xh75a3iz7zbdbkixzrfwhimkfq