Abstract—Chinese aspirates are usually difficult to pronounce for Japanese students. In particular, discriminating between the utterances of aspirated and unaspirated sounds is the most difficult to learn for them. For self-learning, an automatic judgment system was developed that enabled students to check their pronunciations using a computer. We extracted the features of correctly pronounced single-vowel bilabial aspirated sounds pa[p‘a], pi[p‘i], po[p‘o], and pu[p‘u] and unaspirated sounds of ba[pa], bi[pi], bo[po], and bu[pu] by observing the spectrum evolution of breathing power during both voice onset time (VOT), and the voiced period when uttered by 50 native Chinese speakers. We developed a high performance 35-channel computerized filter bank to analyze the evolution of the breathing power spectrum using MATLAB and automatically evaluated the utterances of 50 Japanese students. Using a high-resolution spectrogram, we closely examined the features in VOT closely and improve the criteria for a proper pronunciation. We applied our developed automatic recognition system with improved criteria to the utterances of the students, which passed the screening of native-Chinese speakers. Although our system rejected several samples passing the native speakers’ screening, the success rates were higher than 95% and 98% for aspirated and unaspirated sounds, respectively.

Index Terms—Automatic recognition, Chinese aspirated sounds, extracted the features, self-learning system.

Akemi Hoshino is with National Institute of Technology, Toyama College, 1-2 Ebie, Neriya, Imizu City, Toyama, 933-0293 Japan.

Cite:Akemi Hoshino, "Automatic Recognition of Chinese Aspirated Sounds Pronounced by Japanese Students," International Journal of Computer and Communication Engineering vol. 6, no. 3, pp. 221-228, 2017.