Mandarin tone recognition considering context information

The primary goal of this research is to improve the robustness of tone models and tone recognition rate in continuous Mandarin speech. The procedure was as follows: firstly, four lexical tone models, which are used in isolated Mandarin speech, are employed in clustering tones of continuous Mandarin speech. Then, for reference tone data, variable tone patterns are predicted by using context information. Thirdly, tone dictionary is obtained as combinations of all prediction patterns for tone recognition. The studies we carried out demonstrated that the proposed method outperforms the traditional method with 20 context-dependent tone models and that with eight left-context-dependent tone models for tone recognition.