论文部分内容阅读
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。
This paper points out that the commonly used cepstral mean normalization method removes some of the speaker’s speech features while removing channel factors and therefore has poor robustness in the channel mismatch environment. A channel space mapping method using channel differences and channel mismatch compensation is proposed. A speaker independent speech recognition system is constructed, which is robust to random channels. The experimental results show that the correct recognition rate of the first and the first 30 speakers from the random channel is 5.4% and 18.6% higher than that of the laboratory baseline system, respectively. Finding and compensating for the differences between channels is an effective way to improve the speaker recognition robustness.