Sixth International Conference on Spoken Language Processing
In the pronunciation component, a dynamic low-pass filtered power (DLP) was used as an acoustic feature to discriminate between aspirated sounds and unaspirated sounds. With the sounds uttered by Chinese native speak- ers, we obtained a discrimination accuracy of 98.37% and 96.71% for aspirated consonants and unaspirated conso- nants respectively. Then we applied the pronunciation sys- tem to the sounds uttered by Japanese speakers who are learning Chinese as a second language and obtained an ac- curacy of discrimination. At last, a score is fed back to trainee according to the accuracy of his pronunciation and at the same time, the system instructs the trainee to im- prove his/her pronunciation.
In the listening component, in contrast to traditional methods, our system can not only let the trainee listen to the sound, but can also give him a feedback which will show whether or not his judgment about what he has heard is correct. The listening training system was constructed in Java language so that it can be used on any platform.
We will introduce the background and basic principles of the system, and give some examples of its use in language education.
Bibliographic reference. Tian, Jiapeng / Miwa, Jouji (2000): "Internet training system for listening and pronunciation of Chinese stop consonants", In ICSLP-2000, vol.1, 782-785.