汉语讲话者辨别方法研究  

STUDY ON THE METHOD FOR CHINESE SPEAKER-RECOGNITION

作　　者：林宝成[1] 陈永彬[1] 

机构地区：[1]东南大学无线电工程系,南京210096

出　　处：《模式识别与人工智能》1998年第1期62-66,共5页Pattern Recognition and Artificial Intelligence

基　　金：全国博士后基金资助;江苏省自然科学基金;国家自然科学

摘　　要：根据讲话者在讲话时鼻腔相对固定以及汉语鼻音声母少(只有m-和n-两种)的特点,使用极零模型,并以所有汉语鼻声母音节的零点谱为参数,实现汉语讲话者辨别.所选择的参数稳定、有效,系统模板存储量少.系统训练和识别的运算量小,其规模在20个人时的性能为:用各个人所有单个声母测试时.首选总正识率为67.02%;随机地用3,4,5个声母平均后测试时,则首选正识率分别可达93.33%、96.7%、98.3%.Recording to the properties of speaker's fixed nasal cavity, and a few quantity of Chinese nesal initials, a method of Chinese speaker recognition is presented, by using ARMA model and the spectrum parameters of zero-point of all Chinese nasal initials. The parameters are stable and effectual. The memory for system templates is small. The cost ofrcomputation of training and recognition is low. The correct recognition rate (CRR) are follows: The CRR. of the first chosen is 67.0% for each speaker to test all initials; when randy choosing 3,4,3 initials for each speaker and than averaging them, the average CRR is 93.33%,96.67%,98.33% respectively.

关 键 词：讲话者识别 鼻音声母 极零模型 语音信号处理 

分 类 号：TN912.34[电子电信—通信与信息系统] TP391.42[电子电信—信息与通信工程]