A Hybrid Method of Tone Assessment for Mandarin CALL System

Abstract

Mandarin is known as a tonal language, so the tone is a distinctive discriminative feature in Mandarin. However, accurate segmentation of utterances has a great effect on tone recognition. In this chapter, an approach based on forced alignment of HMM (Hidden Markov Model) is employed to train utterances for obtaining model of getting accurate syllable boundary of utterances. Moreover, for the purpose of getting more objective tone evaluation , a competing models based measure is used to get tonal syllable and tone assessments. We combine two scoring functions to acquire the overall tone scoring results. The experimental results demonstrate that this proposed hybrid method, using forced alignment based tone model and competing models based tone assessment, gives a relative accurate tone assessment.