一种基于特征提取技术的语音识别算法

语音识别是通过对说话人语音信号的分析和提取．自动确定说话人是否在所登记的说话人集合中，以及说话的人是谁的过程．本文提出了一种基于特征提取技术的模板匹配语音识别算法．该算法中，采用与文本有关（主要用于时域分析）和与文本无关（主要用于频域分析）相结合的三种识别方法：快速傅立叶变换（FFT）、平均过零率分析（RBZ）和高斯混合模型（GMM），最后通过三种方法的加权平均来生成判决门限值．通过比较门限值误差的大小，根据“最大似然准则”给出待识别语音的身份．

Algorithm for Voice Recognition Based on Technique of Feature Extractions

Voice Recognition （VR） is defined to confirm whether the speaker is in the registered union automatically, and who the speaker is, by analyzing the voice of speakers. In this paper, a new algorithm based on feature extractions is proposed, which runs on the way of template-matching and takes the way of combining three different ways ： FFT, RBZ and GMM. What the most crucial is getting the variances of every sample with the priority assigned to the other methods. We can compare the variances between every sample and confirmed templates, the visitor identity can be recognized by the minimum variance.