基于改进VQ算法的说话人识别

Speaker Recognition Based on Modified VQ Algorithm

介绍了基于改进矢量量化(VQ)方法的说话人识别系统。该系统采用了能够反映人对语音感知特性的Mel频率倒谱系数(MFCC)作为特征参数,对VQ训练时码书的形成算法作了一些改进,并提出了一种去空胞腔分裂法的优化算法。实验证明,此优化算法减少了矢量量化失真,同时改善了量化的性能。

The speaker recognition system based on vector quantization method is introduced by this text. The system uses Mel frequency cepstrum coefficient （MFCC） which can reflect persons＇ perception characteristic as feature parameters,then makes some improvements on VQ codebook＇s forming algorithm when training,and proposes an optimized splitting algorithm of taking out vacant cell. Experiments show that this optimized algorithm reduces quantization distortion of vector and improves the performance of quantization.