基于矢量量化的说话人识别系统的研究

Speaker Recognition is a special mode of speech recognition. The order does not recognize speech signals. It expects who is speaking. That is distilling one’s character. Using VQ not only can avoid difficult speech subsection and time warping, but also it can reduce data store as a constringent method. This text introduces a method of digits text speaker recognition is introduced, which is based on complex cepstrum parameters and improved vector quantization modeling. When the data of training is small, complex cepstrum can recognize stably.