与文本无关的话者识别

The Text-Independent Speaker Recognition

本文提出一种采用数字信号迭加处理技术从语声中分离话者个人身份特征信息的实用方法,实验验证了这种方法的可行性,并以此建立了与文本无关的话者辨识和话者确认实验系统。通过大量实验,比较了从连续语音和从断续语音中提取话者个人特征信息的效果;比较了线性预测系数、反射系数、倒谱系数和对数面积比系数用于话者识别的效果。

This paper proposes a new utility method to acquire speaKaer-l(?)en(?)y(?)g (?) from voices. The digital signal overlap processing is used in this method, which is testified by the spec tral analysis. Based on this methad, the text -independent speaker identigication and verification testing system is developed. The recognitional effects of the LPC parameter and its tramsforms are com pared. And shortes utterance is used to get the more rapid recognitional speed.