基于矢量量化方法的说话人识别技术

Speaker Recognition Technology Based on VQ

说话人识别是一项通过语音来识别说话人身份的技术,它在保安、司法、军事、财经和信息服务等领域都具有广泛的应用前景。该文采用线性预测倒谱系数和美尔倒谱系数特征相结合,基于矢量量化聚类方法建立了一个与文本无关的、连续语音发音的说话人识别系统。只要矢量量化聚类法码本大小选择合适,该说话人识别系统就可以获得较好的识别效果。当阈值恰当选取时,该系统具备拒绝识别集外人的功能。

Speaker recognition is a kind of technology to judge the speaker＇s identify according to his voice. It has good prospect in many areas such as security, judicatory, and military. One speaker identification system by extracting MFCC as feature vector and using VQ in match phase is constructed. The results of the experiment indicate that, the speaker recognition model based on VQ is effective; the advantage is correct classifying, small memory need and rapid judging.