International Journal of
Information and Education Technology

Abstract

Chinese retroflex aspirates are generally difficult for Japanese students learning Chinese pronunciation. In particular, discriminating between utterances of aspirated dental and retroflex affricates is the most difficult for Japanese-speakers to learn. For the students' self-learning, the automatic judgment system was developed using a computer, so as to check the pronunciations by themselves. We extracted the features of correctly pronounced aspirated dental affricates ca[ʦ‘a], ci[ʦ‘i], ce[ʦ‘ɤ], and cu[ʦ‘u] and aspirated retroflex affricates cha[tʂ‘a], chi[tʂ‘i], che[tʂ‘ɤ], and chu[tʂ‘u] by observing the spectrum evolution of breathing power during both voice onset time and voiced period of sounds uttered by 40 Chinese native speakers. We developed a 35-channel computerized filter bank to analyze the evolution of the breathing power spectrum by using MATLAB and then automatically evaluated the utterances of the 50 Japanese students. At the same time, we asked to native Chinese speakers to evaluate the same students’ utterances for the reference. Our system rejected several samples that passed the screening by the native speakers. The success rates of the system were higher than 91% and 95% for aspirated retroflex and dental affricates, respectively.