| 摘要: |
| 超音段信息主要由声调、语调和轻重缓急等信息组成,反映到特征参数就是基频及其时序信息、能量时序信息和音节长度等,这些信息是话者识别系统中的重要特征参数.在分析已有的提取这些特征参数的方法的基础上,提出了先进行语音切分,再提取声调、能量和音节长度等特征,并生成基于Bigram模型的超音段信息参数模型的方法.将此特征参数形成的模型作为使用MFCC为参数的话者识别主系统的辅助决策模型,实验结果表明：融合超音段信息模型的文本无关话者识别系统的EER相对下降10.5%. |
| 关键词: 话者识别 语音处理 超音段信息 声调 文本无关 |
| DOI： |
| 分类号:TN912.3 |
| 基金项目:广东省教育部产学研项目(2011B090400120);江门市科技项目资金资助(江科［2010］211号 |
|
| Study on the influence of supra-segmental information in text-independent speaker recognition |
|
TANG Lin1,YIN Jun-xun21,2
|
|
1. Department of Electronic & Information Technology, Jiangmen Polytechnic, Jiangmen 529090, China;2. College of Electronics & Information Engineering, South China Univ. of Tech., Guangzhou 510640, China
|
| Abstract: |
| Supra-segmental information mainly consists of pitch, intonation, accent and duration. In character domain , they are fundamental frequency, it’s sequence information, energy sequence information and syllable length. they are all important features in speaker recognition system. After analyzing previous methods of extracting these characters, the method was proposed that speech is divided into single syllable firstly, then tone, energy, syllable length and other features were extracted next, thirdly Bigram model was generated as the Supra-Segmental Information model. The Bigram model was used as the character in an assistant speaker recognition system which assists the main speaker recognition system which uses MFCC parameters to make decisions. The experimental results show that the EER decreased 10.5% in text-independent speaker recognition system after joining the Supra-Segmental Information. |
| Key words: speaker recognition speech processing supra-segmental information pitch text-independent |
