Corrective and reinforcement learning for speaker-independent continuous speech recognition
This paper addresses the issue of learning hidden Markov model (HMM) parameters for speaker-independent continuous speech recognition. Bahl et al. (IEEE conference on acoustics, speech and signal processing, April 1988 a) introduced the corrective training algorithm for speaker-dependent isolated-wo...
Gespeichert in:
| Veröffentlicht in: | Computer speech & language 1990-07, Vol.4 (3), p.231-245 |
|---|---|
| Hauptverfasser: | , |
| Format: | Artikel |
| Sprache: | eng |
| Online-Zugang: | Volltext |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| container_end_page | 245 |
|---|---|
| container_issue | 3 |
| container_start_page | 231 |
| container_title | Computer speech & language |
| container_volume | 4 |
| creator | Lee, Kai-Fu Mahajan, Sanjoy |
| description | This paper addresses the issue of learning hidden Markov model (HMM) parameters for speaker-independent continuous speech recognition. Bahl
et al. (IEEE conference on acoustics, speech and signal processing, April 1988
a) introduced the corrective training algorithm for speaker-dependent isolated-word recognition. Their algorithm attempted to improve the recognition accuracy on the training data. In this work, we extend this algorithm to speaker-independent continjous speech recognition. We use cross-validation to increase the effective training size. We also introduce a near-miss sentence hypothesization algorithm for continuous speech training. The combination of these two approaches resulted in over 20% error reductions both with and without grammar. |
| doi_str_mv | 10.1016/0885-2308(90)90006-R |
| format | Article |
| fullrecord | <record><control><sourceid>proquest_cross</sourceid><recordid>TN_cdi_proquest_miscellaneous_85524012</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><els_id>088523089090006R</els_id><sourcerecordid>85347217</sourcerecordid><originalsourceid>FETCH-LOGICAL-c355r-cbe83ebbe333d3638c51c193a2a711cdb9c667fb106442ce4a291a7377d72e3a3</originalsourceid><addsrcrecordid>eNqNkU1LxDAQhoMouK7-Aw8FQfRQzUebtBdBFr9gQVj05CGk6XTN2k1q0i74701d8eBBvGQg88zLzPsidEzwBcGEX-KiyFPKcHFW4vMSY8zTxQ6aEFzmacE420WTH2QfHYSwGpk8ExP0MnPeg-7NBhJl68SDsY3zGtZg-6QF5a2xyyR-JaED9QY-NbaGDuITAe1sb-zghjC2Qb9GAe2W1vTG2UO016g2wNF3naLn25un2X06f7x7mF3PU83y3Ke6goJBVQFjrI7bFjonmpRMUSUI0XVVas5FUxHMs4xqyBQtiRJMiFpQYIpN0elWt_PufYDQy7UJGtpWWYibySLPaYYJ_QfIMkGJiODJL3DlBm_jEZIwzDmnhOJIZVtKexeCh0Z23qyV_5AEyzEYObouR9dlieVXMHIRx662YxA92RjwMmgDVkNtxiRk7czfAp9PlJXT</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>1306662120</pqid></control><display><type>article</type><title>Corrective and reinforcement learning for speaker-independent continuous speech recognition</title><source>Elsevier ScienceDirect Journals</source><source>Periodicals Index Online</source><creator>Lee, Kai-Fu ; Mahajan, Sanjoy</creator><creatorcontrib>Lee, Kai-Fu ; Mahajan, Sanjoy</creatorcontrib><description>This paper addresses the issue of learning hidden Markov model (HMM) parameters for speaker-independent continuous speech recognition. Bahl
et al. (IEEE conference on acoustics, speech and signal processing, April 1988
a) introduced the corrective training algorithm for speaker-dependent isolated-word recognition. Their algorithm attempted to improve the recognition accuracy on the training data. In this work, we extend this algorithm to speaker-independent continjous speech recognition. We use cross-validation to increase the effective training size. We also introduce a near-miss sentence hypothesization algorithm for continuous speech training. The combination of these two approaches resulted in over 20% error reductions both with and without grammar.</description><identifier>ISSN: 0885-2308</identifier><identifier>EISSN: 1095-8363</identifier><identifier>DOI: 10.1016/0885-2308(90)90006-R</identifier><identifier>CODEN: CSPLEO</identifier><language>eng</language><publisher>London: Elsevier Ltd</publisher><ispartof>Computer speech & language, 1990-07, Vol.4 (3), p.231-245</ispartof><rights>1990</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-c355r-cbe83ebbe333d3638c51c193a2a711cdb9c667fb106442ce4a291a7377d72e3a3</citedby><cites>FETCH-LOGICAL-c355r-cbe83ebbe333d3638c51c193a2a711cdb9c667fb106442ce4a291a7377d72e3a3</cites></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://www.sciencedirect.com/science/article/pii/088523089090006R$$EHTML$$P50$$Gelsevier$$H</linktohtml><link.rule.ids>314,776,780,3537,27846,27901,27902,65306</link.rule.ids></links><search><creatorcontrib>Lee, Kai-Fu</creatorcontrib><creatorcontrib>Mahajan, Sanjoy</creatorcontrib><title>Corrective and reinforcement learning for speaker-independent continuous speech recognition</title><title>Computer speech & language</title><description>This paper addresses the issue of learning hidden Markov model (HMM) parameters for speaker-independent continuous speech recognition. Bahl
et al. (IEEE conference on acoustics, speech and signal processing, April 1988
a) introduced the corrective training algorithm for speaker-dependent isolated-word recognition. Their algorithm attempted to improve the recognition accuracy on the training data. In this work, we extend this algorithm to speaker-independent continjous speech recognition. We use cross-validation to increase the effective training size. We also introduce a near-miss sentence hypothesization algorithm for continuous speech training. The combination of these two approaches resulted in over 20% error reductions both with and without grammar.</description><issn>0885-2308</issn><issn>1095-8363</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>1990</creationdate><recordtype>article</recordtype><sourceid>K30</sourceid><recordid>eNqNkU1LxDAQhoMouK7-Aw8FQfRQzUebtBdBFr9gQVj05CGk6XTN2k1q0i74701d8eBBvGQg88zLzPsidEzwBcGEX-KiyFPKcHFW4vMSY8zTxQ6aEFzmacE420WTH2QfHYSwGpk8ExP0MnPeg-7NBhJl68SDsY3zGtZg-6QF5a2xyyR-JaED9QY-NbaGDuITAe1sb-zghjC2Qb9GAe2W1vTG2UO016g2wNF3naLn25un2X06f7x7mF3PU83y3Ke6goJBVQFjrI7bFjonmpRMUSUI0XVVas5FUxHMs4xqyBQtiRJMiFpQYIpN0elWt_PufYDQy7UJGtpWWYibySLPaYYJ_QfIMkGJiODJL3DlBm_jEZIwzDmnhOJIZVtKexeCh0Z23qyV_5AEyzEYObouR9dlieVXMHIRx662YxA92RjwMmgDVkNtxiRk7czfAp9PlJXT</recordid><startdate>19900701</startdate><enddate>19900701</enddate><creator>Lee, Kai-Fu</creator><creator>Mahajan, Sanjoy</creator><general>Elsevier Ltd</general><general>Academic Press</general><scope>AAYXX</scope><scope>CITATION</scope><scope>HVZBN</scope><scope>K30</scope><scope>PAAUG</scope><scope>PAWHS</scope><scope>PAWZZ</scope><scope>PAXOH</scope><scope>PBHAV</scope><scope>PBQSW</scope><scope>PBYQZ</scope><scope>PCIWU</scope><scope>PCMID</scope><scope>PCZJX</scope><scope>PDGRG</scope><scope>PDWWI</scope><scope>PETMR</scope><scope>PFVGT</scope><scope>PGXDX</scope><scope>PIHIL</scope><scope>PISVA</scope><scope>PJCTQ</scope><scope>PJTMS</scope><scope>PLCHJ</scope><scope>PMHAD</scope><scope>PNQDJ</scope><scope>POUND</scope><scope>PPLAD</scope><scope>PQAPC</scope><scope>PQCAN</scope><scope>PQCMW</scope><scope>PQEME</scope><scope>PQHKH</scope><scope>PQMID</scope><scope>PQNCT</scope><scope>PQNET</scope><scope>PQSCT</scope><scope>PQSET</scope><scope>PSVJG</scope><scope>PVMQY</scope><scope>PZGFC</scope><scope>8BM</scope><scope>7T9</scope></search><sort><creationdate>19900701</creationdate><title>Corrective and reinforcement learning for speaker-independent continuous speech recognition</title><author>Lee, Kai-Fu ; Mahajan, Sanjoy</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c355r-cbe83ebbe333d3638c51c193a2a711cdb9c667fb106442ce4a291a7377d72e3a3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>1990</creationdate><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Lee, Kai-Fu</creatorcontrib><creatorcontrib>Mahajan, Sanjoy</creatorcontrib><collection>CrossRef</collection><collection>Periodicals Index Online Segment 24</collection><collection>Periodicals Index Online</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - West</collection><collection>Primary Sources Access (Plan D) - International</collection><collection>Primary Sources Access & Build (Plan A) - MEA</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - Midwest</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - Northeast</collection><collection>Primary Sources Access (Plan D) - Southeast</collection><collection>Primary Sources Access (Plan D) - North Central</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - Southeast</collection><collection>Primary Sources Access (Plan D) - South Central</collection><collection>Primary Sources Access & Build (Plan A) - UK / I</collection><collection>Primary Sources Access (Plan D) - Canada</collection><collection>Primary Sources Access (Plan D) - EMEALA</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - North Central</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - South Central</collection><collection>Primary Sources Access & Build (Plan A) - International</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - International</collection><collection>Primary Sources Access (Plan D) - West</collection><collection>Periodicals Index Online Segments 1-50</collection><collection>Primary Sources Access (Plan D) - APAC</collection><collection>Primary Sources Access (Plan D) - Midwest</collection><collection>Primary Sources Access (Plan D) - MEA</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - Canada</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - UK / I</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - EMEALA</collection><collection>Primary Sources Access & Build (Plan A) - APAC</collection><collection>Primary Sources Access & Build (Plan A) - Canada</collection><collection>Primary Sources Access & Build (Plan A) - West</collection><collection>Primary Sources Access & Build (Plan A) - EMEALA</collection><collection>Primary Sources Access (Plan D) - Northeast</collection><collection>Primary Sources Access & Build (Plan A) - Midwest</collection><collection>Primary Sources Access & Build (Plan A) - North Central</collection><collection>Primary Sources Access & Build (Plan A) - Northeast</collection><collection>Primary Sources Access & Build (Plan A) - South Central</collection><collection>Primary Sources Access & Build (Plan A) - Southeast</collection><collection>Primary Sources Access (Plan D) - UK / I</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - APAC</collection><collection>Primary Sources Access—Foundation Edition (Plan E) - MEA</collection><collection>ComDisDome</collection><collection>Linguistics and Language Behavior Abstracts (LLBA)</collection><jtitle>Computer speech & language</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Lee, Kai-Fu</au><au>Mahajan, Sanjoy</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Corrective and reinforcement learning for speaker-independent continuous speech recognition</atitle><jtitle>Computer speech & language</jtitle><date>1990-07-01</date><risdate>1990</risdate><volume>4</volume><issue>3</issue><spage>231</spage><epage>245</epage><pages>231-245</pages><issn>0885-2308</issn><eissn>1095-8363</eissn><coden>CSPLEO</coden><abstract>This paper addresses the issue of learning hidden Markov model (HMM) parameters for speaker-independent continuous speech recognition. Bahl
et al. (IEEE conference on acoustics, speech and signal processing, April 1988
a) introduced the corrective training algorithm for speaker-dependent isolated-word recognition. Their algorithm attempted to improve the recognition accuracy on the training data. In this work, we extend this algorithm to speaker-independent continjous speech recognition. We use cross-validation to increase the effective training size. We also introduce a near-miss sentence hypothesization algorithm for continuous speech training. The combination of these two approaches resulted in over 20% error reductions both with and without grammar.</abstract><cop>London</cop><pub>Elsevier Ltd</pub><doi>10.1016/0885-2308(90)90006-R</doi><tpages>15</tpages><oa>free_for_read</oa></addata></record> |
| fulltext | fulltext |
| identifier | ISSN: 0885-2308 |
| ispartof | Computer speech & language, 1990-07, Vol.4 (3), p.231-245 |
| issn | 0885-2308 1095-8363 |
| language | eng |
| recordid | cdi_proquest_miscellaneous_85524012 |
| source | Elsevier ScienceDirect Journals; Periodicals Index Online |
| title | Corrective and reinforcement learning for speaker-independent continuous speech recognition |
| url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-03T21%3A10%3A24IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_cross&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Corrective%20and%20reinforcement%20learning%20for%20speaker-independent%20continuous%20speech%20recognition&rft.jtitle=Computer%20speech%20&%20language&rft.au=Lee,%20Kai-Fu&rft.date=1990-07-01&rft.volume=4&rft.issue=3&rft.spage=231&rft.epage=245&rft.pages=231-245&rft.issn=0885-2308&rft.eissn=1095-8363&rft.coden=CSPLEO&rft_id=info:doi/10.1016/0885-2308(90)90006-R&rft_dat=%3Cproquest_cross%3E85347217%3C/proquest_cross%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=1306662120&rft_id=info:pmid/&rft_els_id=088523089090006R&rfr_iscdi=true |