Interspeech'2005 - Eurospeech
In this paper, we propose a statistical method of evaluating the pronunciation proficiency of Japanese words. We analyze statistically the utterances to note a combination that has a high correlation between a Japanese teacher's score and certain acoustic features. We found that the syllable recognition rates (accuracy) was the best measure of pronunciation proficiency. The effective measure which was highly correlated with Japanese teacher's score was the combination of the posteriori probability, the substitution/accuracy rates and the standard deviation of mora lengths. We obtained a correlation coefficient of 0.712 with closed data and 0.591 with open data for speaker at the five words set level, respectively. The coefficient was near the correlation between humans' scores, 0.600.
Bibliographic reference. Ohta, Kei / Nakagawa, Seiichi (2005): "A statistical method of evaluating pronunciation proficiency for Japanese words", In INTERSPEECH-2005, 2233-2236.