鲁棒性话者辨识中的一种改进的马尔科夫模型

为了提高话者识别系统的噪声鲁棒性,本文对CHMM 进行了改进,将每帧特征参数之间的差分参数来对应状态之间的转移,从而使帧间信息在模型中得到了体现.利用改进后的CHMM模型对不同的特征参数携带的信息进行信息融合.使得在强噪环境下,鲁棒性好的特征参数起主导作用,而在噪声比较小的环境下,精细度高的特征参数起主导作用.实验证明,这种改进的马尔可夫模型明显提高语音识别系统的鲁棒性能,这种技术具有良好的发展和应用前景.

An Improved HMM for Robust Speaker Recognition

A modified CHMM model is proposed to improve the robustness of the speaker recognition system.In the modified model,the difference of the speech feature parameter was the observed vectors of the state transitions of CHMM.Because of the modification,the dependence between frames was used in the modified CHMM.By the fusion of different features,the robust feature will dominate under the noisy environment and the delicate feature will dominate under clean environment.The experiments indicate that the modified CHMM have effectively improved the robustness of the speaker recognition system.