Tone recognition of continuous Cantonese speech based on support vector machines

Tone is an essential component for word formation in all tone languages. It plays a very important role in the transmission of information in speech communication. In this paper, we look at using support vector machines (SVMs) for automatic tone recognition in continuously spoken Cantonese, which is well known for its complex tone system. An adaptive log-scale 5-level F0normalization method is proposed to reduce the tone-irrelevant variation of F0values. Furthermore, an extended version of the above normalization method that considers intonation is also presented. A tone recognition accuracy of 71.50% has been obtained in a speaker-independent task. This result compares favorably with the results reported earlier for the same task. Considerable improvement has been achieved by adopting this tone recognition scheme in a speaker-independent Cantonese large vocabulary continuous speech recognition (LVCSR) task.