语音识别系统中多种特征参数组合的抗噪性

Anti-noise Research of the Various Feature Coefficient on the Speech Recognition System

构建了基于连续隐马尔可夫模型(CHMM)的汉语数字串语音识别系统,为了提高系统在噪音环境下的鲁棒性,抑制平稳噪声及去除信道卷积噪声的影响,引入了动态参数,实验仿真表明采用MFCC参数及一阶、二阶差分及倒谱化明显提高了噪声环境下语音识别系统的识别性能。

Mandarin digital string speech recognition system is designed. Dynamic feature coefficient is introduced in order to enhance noise robust of the recognition system, reduce calm noise and decrease influence from the channel cumulus. The experiment results show that speech recognition rate in noise distinctly improves when choosing MFCC coefficient, one order, second order and cepstrum.