Abstract Confidence measures for speech recognition A survey.pdf

下载文档

2
0
约6.59万字
约 16页
2017-04-13 发布于江苏
举报
版权申诉
保障服务

Abstract Confidence measures for speech recognition A survey.pdf

1、本文档共16页，可阅读全部内容。
2、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。

Abstract Confidence measures for speech recognition A survey

Speech Communication 45 (2005) 455–470 /locate/specomConfidence measures for speech recognition: A survey Hui Jiang * Department of Computer Science, York University, 4700 Keele Street, Toronto, Ont., Canada M3J 1P3 Received 3 August 2004; received in revised form 26 November 2004; accepted 27 December 2004Abstract In speech recognition, confidence measures (CM) are used to evaluate reliability of recognition results. A good con- fidence measure can largely benefit speech recognition systems in many practical applications. In this survey, I summa- rize most research works related to confidence measures which have been done during the past 10–12 years. I will present all these approaches as three major categories, namely CM as a combination of predictor features, CM as a posterior probability, and CM as utterance verification. Then, I also introduce some recent advances in the area. More- over, I will discuss capabilities and limitations of the current CM techniques and generally comment on todays CM approaches. Based on the discussion, I will conclude the paper with some clues for future works. 2005 Elsevier B.V. All rights reserved. Keywords: Automatic speech recognition (ASR); Confidence measures (CM); Word posterior probability; Utterance verification; Likelihood ratio testing (LRT); Bayes factors1. Introduction Automatic speech recognition (ASR) has achieved some substantial successes in past few decades mostly attributing to two prevalent tech- nologies in the field, namely hidden Markov mod- eling (HMM) of speech signals and efficient dynamic programming search (also known as decoding) techniques for very-large-scale networks. Today, in many aspects, it has become a standard0167-6393/$ - see front matter 2005 Elsevier B.V. All rights reserv doi:10.1016/j.specom.2004.12.004 * Tel.: +1 416 736 2100x33346; fax: +1 416 736 5872. E-mail address: hj@cs.yorku.caroutine to build a state-of-the-art speech recogni- tion system for any particular task if sufficient