基于噪声分类与补偿的车载语音识别

摘要

Focusing on the issue that the robustness of the existing vehicular speech recognition system degrades drastically under practical application environments,a noise classification and compensation method based on Support Vector Machine(SVM) is proposed.Firstly,the noise of each application scene is collected to construct the SVM noise classifier which is used to classify the noise in the mute segment of the speech signal,and the corresponding noise training template is selected according to the noise type.The Delta-Spectral Cepstral Coefficients(DSCC) is used as the characteristic parameter,further suppresses the noise in the speech segment for vehicle speech recognition system.Experimental results show that the proposed method can effectively improve the noise robustness of vehicle speech recognition system and has higher speech recognition rate than sparse coded speech enhancement and PNCC feature enhancement methods.%针对现有车载语音识别系统在实际应用环境下噪声鲁棒性较差的问题,提出一种基于支持向量机(SVM)的噪声分类与补偿方法.采集各应用场景下的噪声构建SVM噪声分类器,利用SVM对待测语音静音段中的噪声进行分类,根据噪声类型选择相应的带噪训练模板进行噪声补偿,并将差分频谱倒谱系数作为特征参数进一步抑制语音段中的噪声,从而实现车载语音识别.实验结果表明,该方法可有效增强车载语音识别系统的噪声鲁棒性,并且与稀疏编码语音增强和能量规整倒谱系数特征增强方法相比,具有更高的语音识别率.