声纹识别的应用与矢量量化算法研究

Speaker Recognition Application in Information Course and VQ Algorithm Research

针对声纹识别算法中的矢量量化方法,给出一种新的初始码本选择策略:超球面极值选择法;对提出的新策略进行性能分析,证明该方法在矢量量化码本训练过程中收敛速度快,不存在振荡点与空胞腔问题,计算量小,易于实现;应用此策略实现了基于矢量量化的说话人识别系统,与传统选择策略的系统识别结果进行统计比较,结果表明由改进超球面极值选择法得到的说话人矢量量化码本,识别性能好,具有可应用价值。

A new original codes selection strategy called hypersphere extremum method for the VQ（Veetor Quantization） method which is one of the popular speaker recognition algorithms is presented. The performance of the strategy is analyzed, and it has been testified that the convergence of the VQ code training which does not have vibrating points and empty cells is quick. The algorithm is so simple that it can be easily applied. A speaker recognition system has been realized based on the VQ method. Compared with the traditional methods by statistical results, the speakers VQ codes by improved hypersphere extremum selection method show the fine performance and the available application value in the recognition.