论文部分内容阅读
针对说话人识别系统的鲁棒性问题,文章将语音增强方法应用于说话人识别并提出了一种基于信号子空间与EMD方法联合的语音增强算法。该算法首先用信号子空间增强方法对含噪语音进行降噪,再用EMD方法进一步消除语音中的残留噪声和语音畸变,提高语音增强效果。将上述方法应用于说话人识别系统,进行与文本无关的说话人识别实验。实验结果表明文中方法要优于传统子空间增强算法,对噪声有很好的鲁棒性,在80人的说话人识别实验中,系统识别率达到97.5%。
Aiming at the robustness of speaker recognition system, this paper applies speech enhancement method to speaker recognition and proposes a speech enhancement algorithm based on signal subspace combined with EMD method. Firstly, this algorithm uses the signal subspace enhancement method to denoise the noisy speech, and then uses the EMD method to further eliminate the residual noise and speech distortion in the speech to improve the speech enhancement effect. The above method is applied to the speaker recognition system to perform text-independent speaker recognition experiments. The experimental results show that the proposed method is superior to the traditional subspace enhancement algorithm and robust to noise. In 80 speaker recognition experiments, the system recognition rate reaches 97.5%.