利用语音倒谱参数及基音信息辨认说话人

本文介绍了一种基于矢量量化技术的说话人识别系统，在特征参数上，选用动、静态倒谱互相补偿，充分描述说话人声道模型，同时选用能描述说话人声带激励源特点的基音参数，以全面反映说话人特征。识别时，对三个参数进行优化组合，构成最佳的归一化联合失真进行判决。实验表明，对16人进行测试，当语音长度为2秒时，识别率高达96．8％。

On the Use of Cepstrum and Pitch Information in Speaker Recognition

This paper proposes a vector quantization based speaker recognition system. Instantaneous and transitional cepstral representations carrying complementary information of the vocal tract and pitch period revealing theproperty of the vocal cords are chosen,so as to provide parameters to describe, sufficiently, t he speaker's characteristics. With opitimized combination of three parameters,the best normalized combination distance can be used to make arecognition decision. A 16-speaker database has indicated a 96. 8% identification rate for a 2-second test speech.