基于PNCC特征的录音语音自动识别方法  

Automatic Recognition Method of Recorded Speech Based on PNCC Feature

作　　者：肖宜 葛罗 胡凯 严斌俊 邵立政 XIAO Yi;GE Luo;HU Kai;YAN Bin-jun;SHAO Li-zheng(State Grid Hubei Electric Power Company,Wuhan 430077 China)

机构地区：[1]国网湖北省电力有限公司,湖北武汉430077

出　　处：《自动化技术与应用》2024年第5期163-167,共5页Techniques of Automation and Applications

基　　金：国网湖北省电力有限公司科技项目研究成果(521505210004)。

摘　　要：为了提高录音语音自动识别方法的识别精度,提出基于语音特征提取算法(Power-Normalized Cepstral Coefficients,PNCC)特征的录音语音自动识别方法。采用小波变换方法转换录音语音信号,建立语音特征序列;利用滤波器对语音信号进行滤波分析,抑制非对称噪声;采用离散余弦变换方法变换非线性信号序列,提取基于PNCC特征的录音语音特征;利用人工蜂群算法对录音语音特征参数进行矢量量化,获得录音语音最优码书;构建录音语音识别模型,将提取的特征参数与录音语音最优码书输送到模型内,实现录音语音自动识别。实验结果表明,该方法的录音语音误识率低于20%,提高了识别精度,有效性较强。In order to improve the recognition accuracy of automatic recording speech recognition method,an automatic recording speech recognition method based on speech feature extraction algorithm(PNCC)feature is proposed.The wavelet transform method is used to transform the recorded speech signal and establish the recorded speech feature sequence.The sequence is filtered and analyzed by filter to suppress the asymmetric noise in speech.The discrete cosine transform method is used to transform the nonlinear signal sequence and extract the recorded speech features based on PNCC features.The feature parameters of recorded speech are vector quantized by artificial bee colony algorithm to obtain the optimal codebook of recorded speech.A recording speech recognition model is constructed,and the extracted feature parameters and the optimal codebook of recording speech are transmitted to the model to realize the automatic recognition of recording speech.The experimental results show that the error recognition rate of recorded speech is less than 20%,which improves the recognition accuracy and has strong effectiveness.

关 键 词：PNCC特征 录音语音 自动识别 矢量量化 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]