中文摘要：

在说话人识别系统中，训练和测试环境的不匹配会造成识别性能的显著下降。把小波变换和维纳滤波结合起来对语音进行去噪。对于说话人识别，设计了一个由传统方法（如GMM、MLP和DTW）作为前识别器和通过小波分析（加权求和法）检测到的基音作为后识别器所组成的混合识别器。传统方法分别由三类特征矢量（LSF、倒谱和滤波器组）组成。通过小波分析获得的基音携带了关于说话人身份的信息。这个系统能在不同噪声环境下分析基音周期。试验结果显示，所提出的系统的鲁棒性和辨识率都有所提高。

英文摘要：

In this paper, wavelet transform and Wiener filtering are combined to de-noise. For speaker recognition, we design a hybrid recognizer, which uses conventional methods （such as GMM, MLP and DTW） as pre-recognizer and uses pitch-detection method through wavelet transform （weighing method） as post-recognizer. Conventional methods are composed of three classes of feature vectors ： LSF, cepstrum and filter bank. The pitch carries the information about speaker identification, which is obtained by wavelet analysis. This system can analyze pitch under various noisy environment. Experiment result shows that the robustness and identification rate of the proposed system are both improved.