摘要： 隐马尔可夫模型工具包(HTK)的HParse命令根据用户以正则表达式形式定义的任务语法来生成HTK可用的底层表示的语音识别网络，但不是每个语句都能用正则表达式表示出来。针对该问题，提出基于HTK的语音识别网络算法用于识别网络的优化问题，给出该算法的具体实现过程。实验结果表明，在保证识别率的前提下，优化后的语音识别网络在语音识别系统中所用的时间比较短，算法是有效的。

关键词: 连续语音识别, 自动机, 隐马尔可夫模型工具包, 语音识别网络

Abstract: For speech recognition network of Hidden Markov Model ToolKit(HTK) bottom representation is generated by the HParse command module of the HTK according to the form of regular expressions to define the task grammar, but not every language can use regular expressions to express. Aiming at the problem, this paper presents a HTK-based speech recognition network algorithm used to identify the network optimization problem, gives the detailed realization of the algorithm. Experimental results show that the optimized speech recognition network costs less time in speech recognition than the original un-optimized one, while the recognition rate of the two recognition system configurations are almost the same, and verifies the validity of the proposed algorithm.

Key words: continuous speech recognition, automata, Hidden Markov Model ToolKit(HTK), speech recognition network

中图分类号: 

• TP18