Mandarin tone error detection

Mandarin is a tonal language.Syllables having the same initials and finals,yet different tones may carry rather different meanings.When someone learns Mandarin,pronunciation of the tone is one of the most difficult parts.Therefore,computer assistant language learning systems need to accurately detect tone errors in speech.This paper describes a context dependent tone model to capture the variations of tones in different contexts with a Kullback-Leibler divergence(KLD) to relate an expected tone model for a speech segment with the most fitting tone model as a measure of the correctness of the tone pronunciation.The results show that the method achieves a 6.7% false acceptance rate when the false alarm rate is equal to the false accept rate.