摘要
人的声音会随着时间的变化而变化,例如中年男性的声音比青年时要低沉。因此一般情况下,时间的跨度与说话人识别的正确率成反比。基于模型更新的方式可以提高说话人识别系统的识别率。由于数字均衡的模型语音的发音覆盖率广,所以该文重点研究数字均衡的时变数字语音的模型更新,采用用数字均衡的模型更新方式和非数字均衡的模型更新方式和不更新的方式做对比实验,实验结果显示不固定大小特征的数字均衡的模型效果较好。
A person′s voice changes over time, for example, middle-aged men have a lower voice than they are young. In general, the time span is inversely proportional to the correct rate of the speaker recognition. So Model updating can improve the recognition rate. In the training model, in order to ensure the pronunciation of the digital equalization of the coverage rate is wide, so the paper study of digital balanced digital voice model updating. The paper compare a kind of digital balanced model updating mode with a non digital balance updating mode and non updating model, the conclusion is unsure size characteristics of digital balanced model is better.
作者
苏力
尹琦
SU Li;YIN Qi(School of Computer and Software,Huaiyin Institute of Technology,Huaian 223001,China;School of Computer Engineering,Jiangsu Ocean University,Lianyungang 222005,China)
出处
《电脑知识与技术》
2020年第7期269-271,276,共4页
Computer Knowledge and Technology
关键词
说话人识别
时变
数字均衡
speaker identification
time variant
digital balanced
