Acoustic-to-articulatory inversion using a speaker-normalized HMM-based speech production model

Acoustic-to-articulatory inverse mapping is a difficult problem because of its non-linear and oneto-many characteristics. We have previously developed a speech inversion method using a hidden Markov model (HMM)-based speech production model which takes into account the phonemespecific dynamic constraints of articulatory parameters. We found that the constraint significantly decreases the estimation error of articulatory parameters. However, the model was trained for each speaker and articulatory parameters were estimated in a speaker-dependent manner. In this study, we present a speaker-normalized HMM-based speech production model which is constructed from a multispeaker articulatory-acoustic database, and estimate articulatory parameters from multi-speakers’ speech signals using the model. Result shows that the estimation error of articulatory parameters for vowels is about 1.0 mm.