基于段级特征的快速说话人识别算法及DSP实现

Utterance level feature based fast speaker recognition algorithm with realization by DSP

本文在TMS320C6701EVM板的基础上实现一种快速的说话人识别系统。提出一种基于段级语音特征的说话人识别的快速算法,该算法在传统的GMM算法的基础上使用段级语音特征对测试语音进行数据量压缩,以减少计算时间。并基于车比雪夫和不等式提出了基于协方差模型的段级特征的失真测度描述。根据实验选择了段级特征语音段长度,实验表明该算法在不显著影响识别率的基础上有效地减少了算法延迟,提高了识别速度。

this paper realize a fast speaker recognize system based on TMS320C6701 EVM board.The main algorithm is based on(ULF) utterance level feature.The GMM(Gaussian Mixture Model) is applied in the system.It also proposes utterance level covariance distortion measure by using Chebyshev's summation inequalities.The best pieces of utterance are chosen according to the experiments.Experiments results show that the algorithm efficiently reduce computation delay and fasten the recognition speed.