| 非齐次语音识别HMM模型和THED语音识别与理解系统 |
| |
| 引用本文： | 王作英. 非齐次语音识别HMM模型和THED语音识别与理解系统[J]. 电信科学, 1993, 0(4) |
| |
| 作者姓名： | 王作英 |
| |
| 作者单位： | 清华大学 北京 |
| |
| 摘 要： | 本文从语音状态驻留长度分布出发,建立了一个非齐次隐含马尔可夫(Markov)语音识别模型。这个模型更接近语音信号物理实际,训练和识别的时间、空间复杂性比经典的HMM模型有很大的改进。文中描述了新模型的训练和识别算法,介绍了根据这一模型所设计的一个汉语孤立字全字表的实时识别和理解系统。
|
| 关 键 词： | 识别 训练 理解 状态长度分布 复杂性 |
HMM Model of Inhomogeneous Speech Recognition and THED Speech Recognition and Understanding Svstem |
| |
| Wang Zuoying Tsinghua University,Beijing. HMM Model of Inhomogeneous Speech Recognition and THED Speech Recognition and Understanding Svstem[J]. Telecommunications Science, 1993, 0(4) |
| |
| Authors: | Wang Zuoying Tsinghua University Beijing |
| |
| Affiliation: | Wang Zuoying Tsinghua University,Beijing 100084 |
| |
| Abstract: | In this paper a new inhomogeneous hidden Markov model based on state duration distribution (BDDHMM) of speech has been proposed. The BDDHMM is more suitable to describe the practical speech signal. In this model not only the time complexity but also the space complexity have been greatly reduced in training and recognition phase. The training algorithm and the recognition algorithm of this new model are described. A real-time Chinese recognition and understanding system,which is designed by BDDHMM for all Chinese words,is presented in the paper. |
| |
| Keywords: | recognition training understanding state duration distribution complexity |
| 本文献已被 CNKI 等数据库收录！ |
