首页 | 本学科首页   官方微博 | 高级检索  
     

基于HTK的语音识别的并行化研究与实现
引用本文:刘勇进,史晓东.基于HTK的语音识别的并行化研究与实现[J].计算机应用,2009,29(4):1052-1055.
作者姓名:刘勇进  史晓东
作者单位:1. 厦门大学,计算机科学系,福建,厦门,361005
2. 厦门大学,智能科学与技术系,福建,厦门,361005
基金项目:国家自然科学基金,国家高技术研究发展计划(863计划),福建省重点科技项目,福建省自然科学基金 
摘    要:详细地分析了语音识别的过程,给出了相应的算法描述,并分析了语音识别并行化的可能性。将并行计算的思想应用于语音识别的算法中,使用多线程技术,并引入避免竞争条件的机制,在多核计算机上并行地计算HMM模型节点的似然率,从而得到语音识别的并行化算法。分析了该并行化算法的性能,同时在语音识别工具包HTK 3.4上实现了这种并行化算法。基于WSJ0语料库的实验结果表明该并行化算法在不影响识别结果的前提下能够有效地提高语音识别的实时性能。

关 键 词:多核  HTK  并行计算  语音识别
收稿时间:2008-10-10
修稿时间:2008-12-08

Research and implementation of parallel speech recognition based on HTK
LIU Yong-jin,SHI Xiao-dong.Research and implementation of parallel speech recognition based on HTK[J].journal of Computer Applications,2009,29(4):1052-1055.
Authors:LIU Yong-jin  SHI Xiao-dong
Affiliation:1.Department of Computer Science;Xiamen University;Xiamen Fujian 361005;China;2.Department of Cognitive Science;China
Abstract:After comprehensively analyzing the process of speech recognition and depicting its corresponding algorithm, the feasibility of the parallelism in the algorithm was analyzed. The parallel computing concept and the multi-threading technology were applied in the algorithm of speech recognition, and a protection mechanism was introduced to avoid the occurring race condition during the computing of likelihood of Hidden Markov Model (HMM) on multi-core computer. Then the parallel algorithm for speech recognition was proposed and its performance was also evaluated. The experiments on WSJ0 corpora demonstrate that the implementation of parallel algorithm which was realized in HTK 3.4 toolkit can greatly improve the real-time performance of speech recognition without affecting the results of recognition.
Keywords:HTK
本文献已被 CNKI 维普 万方数据 等数据库收录!
点击此处可从《计算机应用》浏览原始摘要信息
点击此处可从《计算机应用》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号