Evaluation of English pronunciation based on the static and dynamic spectral characteristics of words spoken by Japanese
Hiroshi Hamada, Ryohei Nakatsu · The Journal of the Acoustical Society of America · 1988
To develop an English pronunciation training system, a method is proposed for evaluating the pronunciation ability of Japanese speakers. The quality of English pronunciation is assumed to be determined by the static characteristics of phonetic spectra, the dynamic structure of spectrum sequences, and prosodic characteristics of utterances. Since it is difficult to evaluate these factors directly, evaluation is achieved by comparing English words uttered by a Japanese speaker with those uttered by a native speaker using speech recognition techniques. The static characteristics of phonemes are evaluated by measuring the stability of mapping functions that adapt phonetic spectra of Japanese speakers to those of native speakers. The mapping functions are obtained by speaker adaptation through vector quantization. Evaluation values for the dynamic spectral structure are defined by the DTW matching distance between words spoken by Japanese and those spoken by native speakers. Although prosodic characteristics are not considered, preliminary experiments show that the evaluation results obtained by the proposed method have a good correspondence with human judgments of pronunciation quality.