摘要
提出了一种基于RASTA滤波技术的多维语音特征和支持向量机分类的VAD算法,适用于低信噪比情况下的话音检测。算法在语音特征选择上使用了RASTA-PLP滤波方法,提取出了多维倒谱参数,并将其作为特征向量输入给支持向量机进行分类检测。算法所提取的特征是基于人类听觉感知系统特性的,因此具有普遍意义和稳定性,多维特征与支持向量机的结合则提高了语音分类判决的可靠性。实验结果表明,算法在低信噪比环境下对话音和噪声均具有较高的检出率。
A VAD algorithm based on RASTA-filter multi-dimensional speech feature and Support Vector Machine is presented. It applies to the speech detection under the low SNR conditions. In the selection of speech, it introduces the RASTA-PLP method, which abstracts muhi-dimensional cepstral parameters, and then those parameters are given to SVM as feature vectors to make decision. Because the feature is based on the character of human hearing perception, it is general and robust. Multi-dimensional feature combines with SVM, the algorithm improves the reliability of speech classification and decision. The experimental results indicate that the algorithm has high correct detection rate of speech and noise under the low SNR conditions.
出处
《微计算机信息》
2009年第18期231-232,227,共3页
Control & Automation
