提高话者识别鲁棒性的信道空间映射方法

Robust speaker recognition: using channel space mapping

该文指出了常用的倒谱均值归一方法在去除信道因素的同时,也去掉了一些说话人的语音特征,因此,在信道失配的环境下鲁棒性较差。提出利用信道间差异,补偿信道失配的信道空间映射方法,并构建了一个与文本无关对随机信道鲁棒的说话人识别系统。实验结果表明:对来自随机信道的说话人语音,第1名和前30名的正确识别率,与实验室基线系统的性能比较,分别提高了5.4%和18.6%。寻找并补偿信道间的差异,是一种提高说话人识别鲁棒性的有效方法。

The cepstrual means normalization for removing the channel effects from audio material has also removed some speech features of speaker that make speaker recognition systems less robust in complex channels conditions. This paper presents a text-independent speaker recognition using channel space mapping method that is more robust. The test results show that the correct recognition rate of the first speaker is 5.4%, and for the top 30 speakers is 18.6% better than the base line system. Therefore , the method provides an effective way to improve the robustness of speaker recognition systems by compensating for differences between channels.