基于噪声分类与补偿的车载语音识别

发布时间：2018-01-17 18:33

本文关键词：基于噪声分类与补偿的车载语音识别　出处：《计算机工程》2017年03期 　论文类型：期刊论文

【摘要】：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。
[Abstract]:Aiming at the problem of poor noise robustness of existing vehicle speech recognition system in practical application environment. A method of noise classification and compensation based on support vector machine (SVM) is proposed. The noise of each application scenario is collected to construct SVM noise classifier. SVM is used to classify the noise in the mute segment of the measured speech, and the corresponding training template with noise is selected according to the noise type to compensate the noise. The difference spectrum cepstrum coefficient is used as the characteristic parameter to further suppress the noise in the speech segment so as to realize the vehicle speech recognition. The experimental results show that the proposed method can effectively enhance the noise robustness of the vehicle speech recognition system. And the speech recognition rate is higher than that of sparse speech enhancement and energy regularization cepstrum coefficient enhancement.
【作者单位】：桂林电子科技大学信息与通信学院;
【基金】：广西自然科学基金(2012GXNSFAA053221) 广西千亿元产业产学研用合作项目(信科院0168)
【分类号】：TN912.34
【正文快照】： 数(PNCC)特征,利用子带能量规整算法进行特征增强,去除背景噪声。文献[3]提出基于子带主频车载语音是智能语音识别技术的一个重要应用率信息的提取算法,也取得了优于梅尔频率倒谱系方向,然而现有的语音识别系统在实际噪声环境下数(Mel Frequency Cepstrum Coefficient,MFCC)

【参考文献】