基于ARMA模型的汉语讲话者识别

Chinese speaker recognition based on ARMA model

摘要: 实现了一个仅用鼻音声母且与文本无关的汉语讲话者识别系统,根据讲话者在讲话时鼻腔相对固定、发鼻音时咽腔稳定,以及汉语鼻音声母(只有m-和n-两种)少(全部音节分别只有53和48个)的特点,使用极零(ARMA)模型获得所有汉语鼻声母音节的极点和零点系数的谱参数。系统在对20个讲话者识别时,其性能为:各个人所有单个声母测试时,总正识率为87.92%;分别随机地选用各人的人3、4、5个声母平均后测试时,则平均正识率可达91.67%、95.00%、96.67%、99.97%。

Abstract: According to the properties of speaker's fixed nasal cavity,stable pharynx cavity and a few Chinese nasal initials,A text-independellt Chinese speaker recognition system is presented.By using ARMA model we have gotten the spectrum parameters of all Chinese nasal initials.The correct recognition rate (CRR) for 20 speakers are as follows:The CCR is 87.92% for each speaker in the test with all initials;when random choosing of 2,3,4 and 5 initials in each speaker's utterance and then averaging thier spectrum to test individual template,the average CRRs are 91.67%,95.00%,96.67% and 99.97%,respectively.