摘要
动态HMM模型通过描述语音特征与动态特性之间的关系来改善语音识别的性能 ,本文提出以扩展的基音周期曲线作为动态特性的描述 ,以及在识别过程中采用动态特性自动提取算法以改善识别性能。实验结果表明 ,采用扩展基音周期曲线作为动态特性的动态HMM模型 。
The dynamic HMM describe the relationship between dynamic property and speech to improve the performance of speech recognition. The extraction of dynamic property is important to the dynamic HMM model. In this paper, we use extended pitch contour as the dynamic property. An automatic dynamic property extraction method has been introduced. The result shows that the dynamic HMM model using extended pitch contour can improve the performance of speech recognition.
出处
《电子测量与仪器学报》
CSCD
2004年第1期23-28,共6页
Journal of Electronic Measurement and Instrumentation
基金
教育部重点科技项目-高速移动多媒体通信的理论研究 (A80 82 )
关键词
扩展基音周期曲线
动态特性
自动提取
动态HMM模型
语音识别
Dynamic HMM model, extended pitch contour, dynamic property, automatic dynamic property extraction.