摘要
通过分析当前话者识别系统中常用的一些特征 ,提出了新的特征向量组合 .试验表明 ,特征组合后能很好地区分话者 ;在建模的过程中 ,提出了矢量量化模型与隐含马尔柯夫模型的混合识别策略 ,使正识率有所提高 ;同时 ,对文本有关和文本无关作了对比试验 ,试验证明 ,选择 4~
A text dependent speaker verification system is implemented. In the process of feature extraction, new feature combination are presented. The results of identification prove that the above features are valid in speaker identification system. In the process of speech modeling, the combined identification policy using VQ and HMM are presented, the identifying rate has been improved by using two models. Some other experiments are also done in order to compare the text dependent speaker verification with text independent speaker verification. The experiment proved that the identifying rate is better if 4~6 digital string size is chosen.
出处
《哈尔滨工业大学学报》
EI
CAS
CSCD
北大核心
2000年第4期16-18,共3页
Journal of Harbin Institute of Technology
关键词
说话人识别
文本有关
矢量量化
话者确认系统
speaker identification
text dependent
hidden Markov Model
Vertor quantization
MFCC\