基于高斯语音滤波的稳健文本无关说话人识别

基于高斯模型，提出一种去除实际电话录音中噪音、静音等非语音信号的新方法。与传统的语音检测器方法相比，基于高斯语音滤波的方法在不同信道条件下都可以自动进行，更好地保留了与说话人身份有关的信息。实验结果表明，采用该方法的系统的等错误率比传统方法最多下降了21.2％。

Robust Text-independent Speaker Identification Based on Gaussian Speech-filter

In this paper, a novel approach based on Gaussian model is proposed to remove silence and noise in real-life telephone recordings. Compared to the commonly used energy-based speech detector approach, the painful procedure to choose good thresholds in different channel conditions are avoided, and the useful frames containing information about the speaker identities are better retained. Experiment results show that this approach gives reasonably good performance.