An NN based tone classifier for Cantonese

Chinese language is a typical monosyllabic tonal language. Tone identification is undoubtedly an essential component in the speech recognition problem of Chinese, specifically for the Cantonese dialect which is well known of being very rich in tones. This paper presents an efficient method for tone classification of isolated Cantonese syllables. Several suprasegmental feature parameters for tone identification are extracted from the voiced portion of each recorded utterance and then fed into a multilayer neural network classifier. Using a large vocabulary containing 234 distinct syllables, the system performance for single-speaker and multispeaker cases are found to be 89% and 87% respectively.