CHMM语音识别初值选择方法的研究

Study of Initial Value Selection Method for Speech Recognition Based on Continuous Hidden Markov Models

针对隐马尔科夫模型用于语音识别时传统的参数初始化方法(随机分布之值、K均值算法)可能导致模型参数收敛于局部最优而非全局最优的问题,提出了先按最大距离选择初值中心,再按最小距离将原始数据分割成小类后去除类内干扰点,使类内相似性更强的K均值方法.实验结果表明,改进后的方法与传统方法相比,更好地平滑逼近语音特征,提高语音的识别率.

Given that traditional method of hidden Markov models parameter initialization for speech recognition （random method, k-means） can lead to convergence in local optimization of model parameters rather than global optimization problems. A new approach was proposed with three steps. First, the initial center was selected according to the maximum distance; second, the original data was split into small kinds by the minimum distance; finally, the interference point in the small kind was eliminated. The method resulted in much more similarity k-means than traditional method in the kind. Experimental results show that the improved method has the better approximation of smooth voice characteristics and improves the speech recognition rates which comparing with traditional methods.