噪声环境下文本相关说话人识别方法改进

Improvement of text dependent speaker recognition in noisy environment

采用传统的自相关检测算法识别说话人语音,在受到较大的背景噪声干扰时,检测输出的信噪比不高.为此,提出一种基于小波语音增强和文本相关特征提取算法,在噪声环境下进行文本说话人语音识别系统的总体设计,通过语音噪点的特征匹配,完成语音信号降噪滤波处理;采用小波自适应特征分解,进行语音增强处理,完成文本相关特征提取;将提取结果输入到BP神经网络分类器中,实现说话人识别.仿真结果表明,采用该说话人语音识别算法进行语音检测和分析,具有较高的识别精度,误检概率较低,对语音的降噪性能较好.

The traditional speaker recognition method uses auto correlation detection algorithm.When it is disturbed by the background noise,the detection output signal to noise ratio is not high.Thus,an algorithm is proposed based on wavelet speech enhancement and text related feature extraction.The text speaker speech recognition system′s overall design is done in noisy environment,by voice noise′s feature matching to complete speech signal noise filtering processing.Wavelet adaptive feature decomposition is used to accomplish speech enhancement processing,and to extract the relevant features.The extracted is put into the BP neural network classifier to realize speaker recognition.The simulation results show that the algorithm for speech detection and analysis has high recognition accuracy,low probability of false detection,good performance of noise reduction,and superior technical indicators.