基于群模型改进的说话人确认系统

THE SPEAKER VERIFICATION BASED ON THE IMPROVED COHORT MODEL

在研究说话人识别系统时,分别从特征参数的选取和识别训练两种不同角度分析了加权特征向量和群模型在增强系统性能方面的可行性,并采用群模型与加权特征向量相结合的方式建立与文本无关的说话人确认系统。试验结果表明,含加权特征向量的群模型比传统的矢量量化有更高的辨识率,而且错误拒绝率在一定的错误接受率下也有显著降低。

The feasibility of system performance enhancement by the weighed feature vector and the cohort model is analyzed from the angle of feature parameter selection and recognition in the study of speaker recognition system. Text-lndependent verification system is built based on the combination of the cohort model and the weighed feature vector. The experimental results show that the new method provides significantly higher identification accuracy than the traditional vector quantization and the errors of false rejection are significantly reduced.