A Method of Discrimination and Scoring of Aspirated and Unaspirated Sounds for Chinese Spoken Language Learning System

摘要

To support the construction of a Chinese spoken language self-instruction system, we propose a discrimination method that uses the dynamic low-pass power (DLP) as the feature quantity for automatic discrimination between aspirated and unaspirated-sounds, which is an important problem in Chinese pronunciation education. We also propose a scoring method based on the a posteriori probability of the feature quantity. In an open experiment in which the speech data of five Chinese native speakers other than the training speaker were discriminated, the results show that a discrimination rate of 95.4% was obtained for aspirated sounds and 97.5% for unaspirated sounds. For the speech data of six Japanese learners, an average coincidence rate of 95.2% between the perception result and the system discrimination result was achieved. We have also performed the scoring of the pronunciation by the learner and evaluated its quality, demonstrating the possibility of constructing a real system.