基于GA/VQ的说话人辨认的研究与实现

Research and implementation of speaker identification based on GA/VQ

为了改善在矢量量化说话人识别中,采用模板(码书)表征说话人,模板的质量对识别系统的性能。采用遗传算法改进模板的生成方式,构建了一种GA/VQ说话人辨认系统,给出了一种GA/VQ识别算法,通过遗传操作获得全局优化的说话人模板。实验证明,GA/VQ方法提高了码书的质量,比经典矢量量化识别系统识别率高。

In speech signal processing, the vector qua ntization(VQ) is an important method. The speaker recognition method based on VQ is researched in this paper. In VQ speaker recognition, codebook was adopted t o represent the speaker, and the performance of system is mainly influenced by the qua lity of codebook. Genetic algorithm was adopted in this paper to get global opti m ized speaker codebook and a GA/VQ speaker identification algorithm was proposed. E xperiment results show that this method improves the quality of codebook, the sy s tem based on GA/VQ performs higher correct identification rates compared with th at based on conventional VQ method.