To improve tone recognition accuracy,discriminative training in both feature and model parameters for hidden Markov model based tone modeling is proposed.

• 提出从特征提取参数、模型参数对隐马尔可夫声调模型进行区分型训练,来提高声调识别率;