基于加权特征值补偿的说话人识别

背景噪声的存在,使得说话人识别系统的训练环境和测试环境发生失配,导致系统性能发生急剧下降。本论文提出一种加权特征值补偿算法,把由噪声引起的使带噪语音信号特征值与纯净语音特征值发生偏差的部分去除,从而使进入识别器的特征值接近纯净语音的特征值。在特征值补偿过程中引入了信噪比加权的方法。实验表明,这种方法能够有效的提高说话人识别系统的性能。

Robust Speaker Identification Based on Weighted Feature Compensation

The background noise leads a mismatch between the training environment and testing environment, and degrades the performance of speaker identification system. A weighted feature compensation method for noisy speaker recognition is proposed, which aims at removing the additional term introduced by noise in the feature domain. The SNR of noisy speech is used to weight the feature in the compensation. The experiment results indicate that the proposed method can significantly improve the speaker identification accuracy.