Automatic discrimination of pronunciations of Chinese retroflex and dental affricates

Abstract

Retroflex aspirates in Chinese are generally difficult for Japanese students learning pronunciation. In particular, discriminating between utterances of aspirated dental and retroflex syllables is the most difficult to learn. We extracted the features of correctly pronouncing the aspirated dental syllables ca[ts'a], ci[ts'i], and ce[ts'γ] and aspirated retroflex ones cha[tş'a], chi[tş'i], and che[tş'γ] by observing the spectrum evolution of breathing power during both voice onset time and voiced period of sounds uttered by nine Chinese native speakers. We developed a 35-channel filter bank on a personal computer to analyze the evolution of breathing power spectrum by using MATLAB. We then automatically evaluated the utterances of 20 students judged to be correct by native Chinese speakers and obtained a success rate of higher than 90% and 95% for aspirated retroflex and dental syllables, respectively. © Springer-Verlag 2013.