Pronunciation variation analysis based on acoustic and phonemic distance measures with application examples on Mandarin Chinese

In this paper, two conceptually different statistical distance metrics are defined and analyzed. First, the asymmetric acoustic distance measures how the acoustic property of one phoneme is close to that of another, and is defined here based on the Mahalanobis distance between two hidden Markov models. Second, the asymmetric phonemic distance measures how probable a phoneme is realized as another, based on the aligned phonemic canonical and surface transcriptions of speech corpora. These two mutually dependent distance measures are used to construct an abstract acoustic/phonemic distance plane which is helpful in quantitatively analyzing pronunciation variations. Besides, clear distinction and complicated correlation between these two distances were discussed, which is believed to be helpful in many application areas involving lexical design and acoustic modelling as well. Preliminary analysis were performed on LDC Hub-4NE Mandarin Broadcast News database, and possible application in pronunciation modelling was discussed.