几种开源英语识别工具包的对比分析  

Comparison and Analysis of Several Open Source English Identification Toolkits

作　　者：刘琼[1] LIU Qiong(Xi'an Aeronautical Polytechnic Institute,Xi'an,Shaanxi 710089,China)

机构地区：[1]西安航空职业技术学院通识教育学院,陕西西安710089

出　　处：《计算技术与自动化》2018年第4期123-127,共5页Computing Technology and Automation

摘　　要：对开源英语语音识别工具包在可用性和识别准确性方面进行了对比评价。所对比的语音工具包为HTK语音工具包、CMU Sphinx系列语音处理系统和Kaldi语音工具包。通过对比分析发现,Kaldi语音工具包提供了最先进和全面的声学模型训练技术支持,具有最出色的识别正确率结果;CMU Sphinx系列语音工具包在提供较为全面的声学模型训练基础上,具有最好的识别效率;而HTK语音识别工具包所提供的训练技术支持最少,并且需要自行开发训练脚本,因此使用难度最大。This article makes a comparative evaluation of the availability and recognition accuracy of the open source English speech recognition toolkit.The contrast voice kits are the HTK Voice Kit,CMU Sphinx Series Voice Processing System and Kaldi Voice Toolkit.Through comparative analysis,the Kaldi Voice Toolkit provides state-of-the-art and comprehensive acoustic model training technical support with the best recognition accuracy results;the CMU Sphinx Series Voice Kit provides the most comprehensive acoustic model training.Good recognition efficiency;and HTK voice recognition toolkit provides the least training technical support,and needs to develop training scripts,so it is the most difficult to use.

关 键 词：开源语音识别工具 HTK Spinx Kaldi 语言模型 声学模型 

分 类 号：TM769[电气工程—电力系统及自动化]