Sensitivity analysis of noise robustness methods
The paper addresses the problem of noise robustness from the standpoint of the sensitivity to noise estimation errors. Since the noise is usually estimated in the power-spectral domain, we show that the implied error in the cepstral domain has interesting properties. These properties allow us to com...
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Tagungsbericht |
| Sprache: | eng |
| Schlagworte: | |
| Online-Zugang: | Volltext bestellen |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| container_end_page | 1037 |
|---|---|
| container_issue | |
| container_start_page | I |
| container_title | |
| container_volume | 1 |
| creator | Brayda, L. Rigazio, L. Boman, R. Junqua, J.C. |
| description | The paper addresses the problem of noise robustness from the standpoint of the sensitivity to noise estimation errors. Since the noise is usually estimated in the power-spectral domain, we show that the implied error in the cepstral domain has interesting properties. These properties allow us to compare two key methods used in noise robust speech recognition: spectral subtraction and parallel model combination. We show that parallel model combination has an advantage over spectral subtraction because it is less sensitive to noise estimation errors. Experimental results on the Aurora2 database confirm our theoretical findings, with parallel model combination clearly outperforming spectral subtraction and other well-known signal-based robustness methods. Our Aurora2 results, with parallel model combination, a basic MFCC front-end and a simple noise estimation, are close to the best results obtained on this database with very complex signal processing schemes. |
| doi_str_mv | 10.1109/ICASSP.2004.1326166 |
| format | Conference Proceeding |
| fullrecord | <record><control><sourceid>pascalfrancis_6IE</sourceid><recordid>TN_cdi_ieee_primary_1326166</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>1326166</ieee_id><sourcerecordid>17566100</sourcerecordid><originalsourceid>FETCH-LOGICAL-i500-e3c05fa848ac6c3bdb1554c3b85db1e407f36894b4216c776c0764fb5ed2ae013</originalsourceid><addsrcrecordid>eNpFkM1Lw0AQxRc_wFj7F_SSi8fEmf3OUYpWoaCQHryVzWaDK2lSMlHIf28ggvBgHrwfM8NjbIOQI0Lx8Lp9LMv3nAPIHAXXqPUFS7gwRYYFfFyydWEszBJWWsmvWIKKQ6ZRFjfslugLAKyRNmFQho7iGH_iOKWuc-1EkdK-Sbs-UkiHvvqmsQtE6SmMn31Nd-y6cS2F9d9cscPz02H7ku3fdvNb-ywqgCwID6px83XntRdVXaFScjZWzTZIMI3QtpCV5Ki9MdqD0bKpVKi5C4Bixe6XtWdH3rXN4Dof6Xge4skN0xGN0hoBZm6zcDGE8B8vnYhfqyNTQw</addsrcrecordid><sourcetype>Index Database</sourcetype><iscdi>true</iscdi><recordtype>conference_proceeding</recordtype></control><display><type>conference_proceeding</type><title>Sensitivity analysis of noise robustness methods</title><source>IEEE Electronic Library (IEL) Conference Proceedings</source><creator>Brayda, L. ; Rigazio, L. ; Boman, R. ; Junqua, J.C.</creator><creatorcontrib>Brayda, L. ; Rigazio, L. ; Boman, R. ; Junqua, J.C.</creatorcontrib><description>The paper addresses the problem of noise robustness from the standpoint of the sensitivity to noise estimation errors. Since the noise is usually estimated in the power-spectral domain, we show that the implied error in the cepstral domain has interesting properties. These properties allow us to compare two key methods used in noise robust speech recognition: spectral subtraction and parallel model combination. We show that parallel model combination has an advantage over spectral subtraction because it is less sensitive to noise estimation errors. Experimental results on the Aurora2 database confirm our theoretical findings, with parallel model combination clearly outperforming spectral subtraction and other well-known signal-based robustness methods. Our Aurora2 results, with parallel model combination, a basic MFCC front-end and a simple noise estimation, are close to the best results obtained on this database with very complex signal processing schemes.</description><identifier>ISSN: 1520-6149</identifier><identifier>ISBN: 9780780384842</identifier><identifier>ISBN: 0780384849</identifier><identifier>EISSN: 2379-190X</identifier><identifier>DOI: 10.1109/ICASSP.2004.1326166</identifier><language>eng</language><publisher>Piscataway, N.J: IEEE</publisher><subject>Acoustic noise ; Additive noise ; Applied sciences ; Automatic speech recognition ; Cepstral analysis ; Estimation error ; Exact sciences and technology ; Information, signal and communications theory ; Maximum likelihood estimation ; Noise robustness ; Sensitivity analysis ; Signal processing ; Speech enhancement ; Speech processing ; Telecommunications and information theory</subject><ispartof>2004 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2004, Vol.1, p.I-1037</ispartof><rights>2006 INIST-CNRS</rights><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://ieeexplore.ieee.org/document/1326166$$EHTML$$P50$$Gieee$$H</linktohtml><link.rule.ids>309,310,778,782,787,788,2054,4038,4039,27908,54903</link.rule.ids><linktorsrc>$$Uhttps://ieeexplore.ieee.org/document/1326166$$EView_record_in_IEEE$$FView_record_in_$$GIEEE</linktorsrc><backlink>$$Uhttp://pascal-francis.inist.fr/vibad/index.php?action=getRecordDetail&idt=17566100$$DView record in Pascal Francis$$Hfree_for_read</backlink></links><search><creatorcontrib>Brayda, L.</creatorcontrib><creatorcontrib>Rigazio, L.</creatorcontrib><creatorcontrib>Boman, R.</creatorcontrib><creatorcontrib>Junqua, J.C.</creatorcontrib><title>Sensitivity analysis of noise robustness methods</title><title>2004 IEEE International Conference on Acoustics, Speech, and Signal Processing</title><addtitle>ICASSP</addtitle><description>The paper addresses the problem of noise robustness from the standpoint of the sensitivity to noise estimation errors. Since the noise is usually estimated in the power-spectral domain, we show that the implied error in the cepstral domain has interesting properties. These properties allow us to compare two key methods used in noise robust speech recognition: spectral subtraction and parallel model combination. We show that parallel model combination has an advantage over spectral subtraction because it is less sensitive to noise estimation errors. Experimental results on the Aurora2 database confirm our theoretical findings, with parallel model combination clearly outperforming spectral subtraction and other well-known signal-based robustness methods. Our Aurora2 results, with parallel model combination, a basic MFCC front-end and a simple noise estimation, are close to the best results obtained on this database with very complex signal processing schemes.</description><subject>Acoustic noise</subject><subject>Additive noise</subject><subject>Applied sciences</subject><subject>Automatic speech recognition</subject><subject>Cepstral analysis</subject><subject>Estimation error</subject><subject>Exact sciences and technology</subject><subject>Information, signal and communications theory</subject><subject>Maximum likelihood estimation</subject><subject>Noise robustness</subject><subject>Sensitivity analysis</subject><subject>Signal processing</subject><subject>Speech enhancement</subject><subject>Speech processing</subject><subject>Telecommunications and information theory</subject><issn>1520-6149</issn><issn>2379-190X</issn><isbn>9780780384842</isbn><isbn>0780384849</isbn><fulltext>true</fulltext><rsrctype>conference_proceeding</rsrctype><creationdate>2004</creationdate><recordtype>conference_proceeding</recordtype><sourceid>6IE</sourceid><sourceid>RIE</sourceid><recordid>eNpFkM1Lw0AQxRc_wFj7F_SSi8fEmf3OUYpWoaCQHryVzWaDK2lSMlHIf28ggvBgHrwfM8NjbIOQI0Lx8Lp9LMv3nAPIHAXXqPUFS7gwRYYFfFyydWEszBJWWsmvWIKKQ6ZRFjfslugLAKyRNmFQho7iGH_iOKWuc-1EkdK-Sbs-UkiHvvqmsQtE6SmMn31Nd-y6cS2F9d9cscPz02H7ku3fdvNb-ywqgCwID6px83XntRdVXaFScjZWzTZIMI3QtpCV5Ki9MdqD0bKpVKi5C4Bixe6XtWdH3rXN4Dof6Xge4skN0xGN0hoBZm6zcDGE8B8vnYhfqyNTQw</recordid><startdate>2004</startdate><enddate>2004</enddate><creator>Brayda, L.</creator><creator>Rigazio, L.</creator><creator>Boman, R.</creator><creator>Junqua, J.C.</creator><general>IEEE</general><scope>6IE</scope><scope>6IH</scope><scope>CBEJK</scope><scope>RIE</scope><scope>RIO</scope><scope>IQODW</scope></search><sort><creationdate>2004</creationdate><title>Sensitivity analysis of noise robustness methods</title><author>Brayda, L. ; Rigazio, L. ; Boman, R. ; Junqua, J.C.</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-i500-e3c05fa848ac6c3bdb1554c3b85db1e407f36894b4216c776c0764fb5ed2ae013</frbrgroupid><rsrctype>conference_proceedings</rsrctype><prefilter>conference_proceedings</prefilter><language>eng</language><creationdate>2004</creationdate><topic>Acoustic noise</topic><topic>Additive noise</topic><topic>Applied sciences</topic><topic>Automatic speech recognition</topic><topic>Cepstral analysis</topic><topic>Estimation error</topic><topic>Exact sciences and technology</topic><topic>Information, signal and communications theory</topic><topic>Maximum likelihood estimation</topic><topic>Noise robustness</topic><topic>Sensitivity analysis</topic><topic>Signal processing</topic><topic>Speech enhancement</topic><topic>Speech processing</topic><topic>Telecommunications and information theory</topic><toplevel>online_resources</toplevel><creatorcontrib>Brayda, L.</creatorcontrib><creatorcontrib>Rigazio, L.</creatorcontrib><creatorcontrib>Boman, R.</creatorcontrib><creatorcontrib>Junqua, J.C.</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><collection>Pascal-Francis</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Brayda, L.</au><au>Rigazio, L.</au><au>Boman, R.</au><au>Junqua, J.C.</au><format>book</format><genre>proceeding</genre><ristype>CONF</ristype><atitle>Sensitivity analysis of noise robustness methods</atitle><btitle>2004 IEEE International Conference on Acoustics, Speech, and Signal Processing</btitle><stitle>ICASSP</stitle><date>2004</date><risdate>2004</risdate><volume>1</volume><spage>I</spage><epage>1037</epage><pages>I-1037</pages><issn>1520-6149</issn><eissn>2379-190X</eissn><isbn>9780780384842</isbn><isbn>0780384849</isbn><abstract>The paper addresses the problem of noise robustness from the standpoint of the sensitivity to noise estimation errors. Since the noise is usually estimated in the power-spectral domain, we show that the implied error in the cepstral domain has interesting properties. These properties allow us to compare two key methods used in noise robust speech recognition: spectral subtraction and parallel model combination. We show that parallel model combination has an advantage over spectral subtraction because it is less sensitive to noise estimation errors. Experimental results on the Aurora2 database confirm our theoretical findings, with parallel model combination clearly outperforming spectral subtraction and other well-known signal-based robustness methods. Our Aurora2 results, with parallel model combination, a basic MFCC front-end and a simple noise estimation, are close to the best results obtained on this database with very complex signal processing schemes.</abstract><cop>Piscataway, N.J</cop><pub>IEEE</pub><doi>10.1109/ICASSP.2004.1326166</doi></addata></record> |
| fulltext | fulltext_linktorsrc |
| identifier | ISSN: 1520-6149 |
| ispartof | 2004 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2004, Vol.1, p.I-1037 |
| issn | 1520-6149 2379-190X |
| language | eng |
| recordid | cdi_ieee_primary_1326166 |
| source | IEEE Electronic Library (IEL) Conference Proceedings |
| subjects | Acoustic noise Additive noise Applied sciences Automatic speech recognition Cepstral analysis Estimation error Exact sciences and technology Information, signal and communications theory Maximum likelihood estimation Noise robustness Sensitivity analysis Signal processing Speech enhancement Speech processing Telecommunications and information theory |
| title | Sensitivity analysis of noise robustness methods |
| url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-16T08%3A01%3A56IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-pascalfrancis_6IE&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=proceeding&rft.atitle=Sensitivity%20analysis%20of%20noise%20robustness%20methods&rft.btitle=2004%20IEEE%20International%20Conference%20on%20Acoustics,%20Speech,%20and%20Signal%20Processing&rft.au=Brayda,%20L.&rft.date=2004&rft.volume=1&rft.spage=I&rft.epage=1037&rft.pages=I-1037&rft.issn=1520-6149&rft.eissn=2379-190X&rft.isbn=9780780384842&rft.isbn_list=0780384849&rft_id=info:doi/10.1109/ICASSP.2004.1326166&rft_dat=%3Cpascalfrancis_6IE%3E17566100%3C/pascalfrancis_6IE%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rft_ieee_id=1326166&rfr_iscdi=true |