自动发音人识别中测试文本选择的研究

Password selection on automatic speaker recognition

摘要: 本文研究自动发音人识别中测试文本的选择.提出并证实了结合汉语特点的测试文本选择的重要性和应用潜力,并总结了几条简单规则;应用并改善了解决发音动态变化的时间域规正法.
本系统用12阶LPCC倒谱系数和基音周期构成混合特征矢量,采用三字三模板匹配的识别方法,在一般实验室环境下,以录音机为传输媒介,达到了0.6%的确认错误率。

Abstract: The importance and the potential of password selection associating with Chinese speech characteristics are discussed and several rules are put forward; The method of time domain normalization solving the problem of speech dynamic variation is applied and improved.
Using 12th order LPCC and Pitch Period as Multi-parameter Verification Vector and using multi-passwords and multi-templates recognition method, in a normal laboratory the error probability of the text-dependent speaker verification is 0.6% on tape recorders.