Maximum Entropy Confidence Estimation for Speech Recognition

For many automatic speech recognition (ASR) applications, it is useful to predict the likelihood that the recognized string contains an error. This paper explores two modifications of a classic design. First, it replaces the standard maximum likelihood classifier with a maximum entropy classifier. T...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: White, C., Droppo, J., Acero, A., Odell, J.
Format: Tagungsbericht
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page IV-812
container_issue
container_start_page IV-809
container_title
container_volume 4
creator White, C.
Droppo, J.
Acero, A.
Odell, J.
description For many automatic speech recognition (ASR) applications, it is useful to predict the likelihood that the recognized string contains an error. This paper explores two modifications of a classic design. First, it replaces the standard maximum likelihood classifier with a maximum entropy classifier. The maximum entropy framework carries the dual advantages discriminative training and reasonable generalization. Second, it includes a number of alternative features. Our ASR system is heavily pruned, and often produces recognition lattices with only a single path. These alternate features are meant to serve as a surrogate for the typical features that can be computed from a rich lattice. We show that the maximum entropy classifier easily outperforms the standard baseline system, and the alternative features provide consistent gains for all of our test sets.
doi_str_mv 10.1109/ICASSP.2007.367036
format Conference Proceeding
fullrecord <record><control><sourceid>ieee_6IE</sourceid><recordid>TN_cdi_ieee_primary_4218224</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>4218224</ieee_id><sourcerecordid>4218224</sourcerecordid><originalsourceid>FETCH-LOGICAL-i175t-b298cdb77066b4cc054caff3f0b054c012ba0983b0497734b726873b6e7b74b93</originalsourceid><addsrcrecordid>eNpVjttKxDAUReMNrOP8gL7kB1pPktOcBHyRoV5gRLEKvg1NJtWIvdBWcP7eGfTFp73ZCzaLsTMBmRBgL-4WV2X5mEkAypQmUHqPzS0ZgRIRSBq9zxKpyKbCwuvBP0b2kCUil5BqgfaYnYzjBwAYQpOwy_vqOzZfDS_aaej6DV90bR3XofWBF-MUm2qKXcvrbuBlH4J_50_Bd29t3M2n7KiuPscw_8sZe7kunhe36fLhZmu8TKOgfEqdtMavHRFo7dB7yNFXda1qcLsKQroKrFEO0BIpdCS1IeV0IEforJqx89_fGEJY9cPWatisUAojJaofSJ5M5g</addsrcrecordid><sourcetype>Publisher</sourcetype><iscdi>true</iscdi><recordtype>conference_proceeding</recordtype></control><display><type>conference_proceeding</type><title>Maximum Entropy Confidence Estimation for Speech Recognition</title><source>IEEE Electronic Library (IEL) Conference Proceedings</source><creator>White, C. ; Droppo, J. ; Acero, A. ; Odell, J.</creator><creatorcontrib>White, C. ; Droppo, J. ; Acero, A. ; Odell, J.</creatorcontrib><description>For many automatic speech recognition (ASR) applications, it is useful to predict the likelihood that the recognized string contains an error. This paper explores two modifications of a classic design. First, it replaces the standard maximum likelihood classifier with a maximum entropy classifier. The maximum entropy framework carries the dual advantages discriminative training and reasonable generalization. Second, it includes a number of alternative features. Our ASR system is heavily pruned, and often produces recognition lattices with only a single path. These alternate features are meant to serve as a surrogate for the typical features that can be computed from a rich lattice. We show that the maximum entropy classifier easily outperforms the standard baseline system, and the alternative features provide consistent gains for all of our test sets.</description><identifier>ISSN: 1520-6149</identifier><identifier>ISBN: 9781424407279</identifier><identifier>ISBN: 1424407273</identifier><identifier>EISSN: 2379-190X</identifier><identifier>EISBN: 9781424407286</identifier><identifier>EISBN: 1424407281</identifier><identifier>DOI: 10.1109/ICASSP.2007.367036</identifier><language>eng</language><publisher>IEEE</publisher><subject>Automatic speech recognition ; Engines ; Entropy ; Lattices ; Maximum entropy methods ; Maximum likelihood decoding ; Maximum likelihood estimation ; Natural languages ; Speech processing ; Speech recognition ; System testing</subject><ispartof>2007 IEEE International Conference on Acoustics, Speech and Signal Processing - ICASSP '07, 2007, Vol.4, p.IV-809-IV-812</ispartof><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://ieeexplore.ieee.org/document/4218224$$EHTML$$P50$$Gieee$$H</linktohtml><link.rule.ids>309,310,776,780,785,786,2052,27902,54895</link.rule.ids><linktorsrc>$$Uhttps://ieeexplore.ieee.org/document/4218224$$EView_record_in_IEEE$$FView_record_in_$$GIEEE</linktorsrc></links><search><creatorcontrib>White, C.</creatorcontrib><creatorcontrib>Droppo, J.</creatorcontrib><creatorcontrib>Acero, A.</creatorcontrib><creatorcontrib>Odell, J.</creatorcontrib><title>Maximum Entropy Confidence Estimation for Speech Recognition</title><title>2007 IEEE International Conference on Acoustics, Speech and Signal Processing - ICASSP '07</title><addtitle>ICASSP</addtitle><description>For many automatic speech recognition (ASR) applications, it is useful to predict the likelihood that the recognized string contains an error. This paper explores two modifications of a classic design. First, it replaces the standard maximum likelihood classifier with a maximum entropy classifier. The maximum entropy framework carries the dual advantages discriminative training and reasonable generalization. Second, it includes a number of alternative features. Our ASR system is heavily pruned, and often produces recognition lattices with only a single path. These alternate features are meant to serve as a surrogate for the typical features that can be computed from a rich lattice. We show that the maximum entropy classifier easily outperforms the standard baseline system, and the alternative features provide consistent gains for all of our test sets.</description><subject>Automatic speech recognition</subject><subject>Engines</subject><subject>Entropy</subject><subject>Lattices</subject><subject>Maximum entropy methods</subject><subject>Maximum likelihood decoding</subject><subject>Maximum likelihood estimation</subject><subject>Natural languages</subject><subject>Speech processing</subject><subject>Speech recognition</subject><subject>System testing</subject><issn>1520-6149</issn><issn>2379-190X</issn><isbn>9781424407279</isbn><isbn>1424407273</isbn><isbn>9781424407286</isbn><isbn>1424407281</isbn><fulltext>true</fulltext><rsrctype>conference_proceeding</rsrctype><creationdate>2007</creationdate><recordtype>conference_proceeding</recordtype><sourceid>6IE</sourceid><sourceid>RIE</sourceid><recordid>eNpVjttKxDAUReMNrOP8gL7kB1pPktOcBHyRoV5gRLEKvg1NJtWIvdBWcP7eGfTFp73ZCzaLsTMBmRBgL-4WV2X5mEkAypQmUHqPzS0ZgRIRSBq9zxKpyKbCwuvBP0b2kCUil5BqgfaYnYzjBwAYQpOwy_vqOzZfDS_aaej6DV90bR3XofWBF-MUm2qKXcvrbuBlH4J_50_Bd29t3M2n7KiuPscw_8sZe7kunhe36fLhZmu8TKOgfEqdtMavHRFo7dB7yNFXda1qcLsKQroKrFEO0BIpdCS1IeV0IEforJqx89_fGEJY9cPWatisUAojJaofSJ5M5g</recordid><startdate>200704</startdate><enddate>200704</enddate><creator>White, C.</creator><creator>Droppo, J.</creator><creator>Acero, A.</creator><creator>Odell, J.</creator><general>IEEE</general><scope>6IE</scope><scope>6IH</scope><scope>CBEJK</scope><scope>RIE</scope><scope>RIO</scope></search><sort><creationdate>200704</creationdate><title>Maximum Entropy Confidence Estimation for Speech Recognition</title><author>White, C. ; Droppo, J. ; Acero, A. ; Odell, J.</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-i175t-b298cdb77066b4cc054caff3f0b054c012ba0983b0497734b726873b6e7b74b93</frbrgroupid><rsrctype>conference_proceedings</rsrctype><prefilter>conference_proceedings</prefilter><language>eng</language><creationdate>2007</creationdate><topic>Automatic speech recognition</topic><topic>Engines</topic><topic>Entropy</topic><topic>Lattices</topic><topic>Maximum entropy methods</topic><topic>Maximum likelihood decoding</topic><topic>Maximum likelihood estimation</topic><topic>Natural languages</topic><topic>Speech processing</topic><topic>Speech recognition</topic><topic>System testing</topic><toplevel>online_resources</toplevel><creatorcontrib>White, C.</creatorcontrib><creatorcontrib>Droppo, J.</creatorcontrib><creatorcontrib>Acero, A.</creatorcontrib><creatorcontrib>Odell, J.</creatorcontrib><collection>IEEE Electronic Library (IEL) Conference Proceedings</collection><collection>IEEE Proceedings Order Plan (POP) 1998-present by volume</collection><collection>IEEE Xplore All Conference Proceedings</collection><collection>IEEE Electronic Library (IEL)</collection><collection>IEEE Proceedings Order Plans (POP) 1998-present</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>White, C.</au><au>Droppo, J.</au><au>Acero, A.</au><au>Odell, J.</au><format>book</format><genre>proceeding</genre><ristype>CONF</ristype><atitle>Maximum Entropy Confidence Estimation for Speech Recognition</atitle><btitle>2007 IEEE International Conference on Acoustics, Speech and Signal Processing - ICASSP '07</btitle><stitle>ICASSP</stitle><date>2007-04</date><risdate>2007</risdate><volume>4</volume><spage>IV-809</spage><epage>IV-812</epage><pages>IV-809-IV-812</pages><issn>1520-6149</issn><eissn>2379-190X</eissn><isbn>9781424407279</isbn><isbn>1424407273</isbn><eisbn>9781424407286</eisbn><eisbn>1424407281</eisbn><abstract>For many automatic speech recognition (ASR) applications, it is useful to predict the likelihood that the recognized string contains an error. This paper explores two modifications of a classic design. First, it replaces the standard maximum likelihood classifier with a maximum entropy classifier. The maximum entropy framework carries the dual advantages discriminative training and reasonable generalization. Second, it includes a number of alternative features. Our ASR system is heavily pruned, and often produces recognition lattices with only a single path. These alternate features are meant to serve as a surrogate for the typical features that can be computed from a rich lattice. We show that the maximum entropy classifier easily outperforms the standard baseline system, and the alternative features provide consistent gains for all of our test sets.</abstract><pub>IEEE</pub><doi>10.1109/ICASSP.2007.367036</doi></addata></record>
fulltext fulltext_linktorsrc
identifier ISSN: 1520-6149
ispartof 2007 IEEE International Conference on Acoustics, Speech and Signal Processing - ICASSP '07, 2007, Vol.4, p.IV-809-IV-812
issn 1520-6149
2379-190X
language eng
recordid cdi_ieee_primary_4218224
source IEEE Electronic Library (IEL) Conference Proceedings
subjects Automatic speech recognition
Engines
Entropy
Lattices
Maximum entropy methods
Maximum likelihood decoding
Maximum likelihood estimation
Natural languages
Speech processing
Speech recognition
System testing
title Maximum Entropy Confidence Estimation for Speech Recognition
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-28T17%3A59%3A05IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-ieee_6IE&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=proceeding&rft.atitle=Maximum%20Entropy%20Confidence%20Estimation%20for%20Speech%20Recognition&rft.btitle=2007%20IEEE%20International%20Conference%20on%20Acoustics,%20Speech%20and%20Signal%20Processing%20-%20ICASSP%20'07&rft.au=White,%20C.&rft.date=2007-04&rft.volume=4&rft.spage=IV-809&rft.epage=IV-812&rft.pages=IV-809-IV-812&rft.issn=1520-6149&rft.eissn=2379-190X&rft.isbn=9781424407279&rft.isbn_list=1424407273&rft_id=info:doi/10.1109/ICASSP.2007.367036&rft_dat=%3Cieee_6IE%3E4218224%3C/ieee_6IE%3E%3Curl%3E%3C/url%3E&rft.eisbn=9781424407286&rft.eisbn_list=1424407281&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rft_ieee_id=4218224&rfr_iscdi=true