<?xml version="1.0" encoding="utf-8"?>
<TEI xmlns="http://www.tei-c.org/ns/1.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:hal="http://hal.archives-ouvertes.fr/" xmlns:gml="http://www.opengis.net/gml/3.3/" xmlns:gmlce="http://www.opengis.net/gml/3.3/ce" version="1.1" xsi:schemaLocation="http://www.tei-c.org/ns/1.0 http://api.archives-ouvertes.fr/documents/aofr-sword.xsd">
  <teiHeader>
    <fileDesc>
      <titleStmt>
        <title>HAL TEI export of hal-03631210</title>
      </titleStmt>
      <publicationStmt>
        <distributor>CCSD</distributor>
        <availability status="restricted">
          <licence target="https://creativecommons.org/publicdomain/zero/1.0/">CC0 1.0 - Universal</licence>
        </availability>
        <date when="2026-05-19T03:05:22+02:00"/>
      </publicationStmt>
      <sourceDesc>
        <p part="N">HAL API Platform</p>
      </sourceDesc>
    </fileDesc>
  </teiHeader>
  <text>
    <body>
      <listBibl>
        <biblFull>
          <titleStmt>
            <title xml:lang="en">Cosine-similarity penalty to discriminate sound classes in weakly-supervised sound event detection.</title>
            <author role="aut">
              <persName>
                <forename type="first">Thomas</forename>
                <surname>Pellegrini</surname>
              </persName>
              <email type="md5">cf3d88b6af1d106e594c0cb087012952</email>
              <email type="domain">irit.fr</email>
              <idno type="idhal" notation="string">thomas-pellegrini</idno>
              <idno type="idhal" notation="numeric">741962</idno>
              <idno type="halauthorid" notation="string">13197-741962</idno>
              <idno type="ARXIV">https://arxiv.org/a/pellegrini_t_1</idno>
              <idno type="ORCID">https://orcid.org/0000-0001-8984-1399</idno>
              <idno type="IDREF">https://www.idref.fr/127577955</idno>
              <affiliation ref="#struct-393117"/>
            </author>
            <author role="aut">
              <persName>
                <forename type="first">Léo</forename>
                <surname>Cances</surname>
              </persName>
              <idno type="halauthorid">2028272-0</idno>
              <affiliation ref="#struct-393117"/>
            </author>
            <editor role="depositor">
              <persName>
                <forename>Open Archive Toulouse Archive Ouverte</forename>
                <surname>(OATAO)</surname>
              </persName>
              <email type="md5">ac3f71840d50da799c1bb19c7b808bac</email>
              <email type="domain">listes-diff.inp-toulouse.fr</email>
            </editor>
          </titleStmt>
          <editionStmt>
            <edition n="v1" type="current">
              <date type="whenSubmitted">2022-04-05 14:54:35</date>
              <date type="whenModified">2025-10-22 18:04:09</date>
              <date type="whenReleased"/>
              <date type="whenProduced">2019-07-14</date>
              <date type="whenEndEmbargoed">2022-04-05</date>
              <ref type="file" target="https://hal.science/hal-03631210v1/document">
                <date notBefore="2022-04-05"/>
              </ref>
              <ref type="file" subtype="author" n="1" target="https://hal.science/hal-03631210v1/file/pellegrini_24872.pdf" id="file-3631210-3172937">
                <date notBefore="2022-04-05"/>
              </ref>
            </edition>
            <respStmt>
              <resp>contributor</resp>
              <name key="146910">
                <persName>
                  <forename>Open Archive Toulouse Archive Ouverte</forename>
                  <surname>(OATAO)</surname>
                </persName>
                <email type="md5">ac3f71840d50da799c1bb19c7b808bac</email>
                <email type="domain">listes-diff.inp-toulouse.fr</email>
              </name>
            </respStmt>
          </editionStmt>
          <publicationStmt>
            <distributor>CCSD</distributor>
            <idno type="halId">hal-03631210</idno>
            <idno type="halUri">https://hal.science/hal-03631210</idno>
            <idno type="halBibtex">pellegrini:hal-03631210</idno>
            <idno type="halRefHtml">&lt;i&gt;International Joint Conference on Neural Networks (IJCNN 2019)&lt;/i&gt;, Jul 2019, Budapest, Hungary. pp.1-8</idno>
            <idno type="halRef">International Joint Conference on Neural Networks (IJCNN 2019), Jul 2019, Budapest, Hungary. pp.1-8</idno>
            <availability status="restricted">
              <licence target="https://about.hal.science/hal-authorisation-v1/">HAL Authorization<ref corresp="#file-3631210-3172937"/></licence>
            </availability>
          </publicationStmt>
          <seriesStmt>
            <idno type="stamp" n="UNIV-TLSE2">Université Toulouse 2</idno>
            <idno type="stamp" n="UNIV-TLSE3">Université de Toulouse</idno>
            <idno type="stamp" n="CNRS">CNRS - Centre national de la recherche scientifique</idno>
            <idno type="stamp" n="UT1-CAPITOLE">Université Toulouse 1 Capitole</idno>
            <idno type="stamp" n="IRIT">IRIT - Institut de Recherche en Informatique de Toulouse</idno>
            <idno type="stamp" n="IRIT-SAMOVA" corresp="IRIT">Structuration, Analyse et Modélisation de documents Vidéo et Audio</idno>
            <idno type="stamp" n="IRIT-SI" corresp="IRIT">Signaux et Images</idno>
            <idno type="stamp" n="TOULOUSE-INP">Toulouse - INP</idno>
            <idno type="stamp" n="UNIV-UT3">Université Toulouse 3</idno>
            <idno type="stamp" n="UT3-INP">Université de Toulouse / Toulouse INP</idno>
            <idno type="stamp" n="UT3-TOULOUSEINP">Université de Toulouse / Toulouse INP</idno>
          </seriesStmt>
          <notesStmt>
            <note type="audience" n="2">International</note>
            <note type="invited" n="0">No</note>
            <note type="popular" n="0">No</note>
            <note type="peer" n="1">Yes</note>
            <note type="proceedings" n="1">Yes</note>
          </notesStmt>
          <sourceDesc>
            <biblStruct>
              <analytic>
                <title xml:lang="en">Cosine-similarity penalty to discriminate sound classes in weakly-supervised sound event detection.</title>
                <author role="aut">
                  <persName>
                    <forename type="first">Thomas</forename>
                    <surname>Pellegrini</surname>
                  </persName>
                  <email type="md5">cf3d88b6af1d106e594c0cb087012952</email>
                  <email type="domain">irit.fr</email>
                  <idno type="idhal" notation="string">thomas-pellegrini</idno>
                  <idno type="idhal" notation="numeric">741962</idno>
                  <idno type="halauthorid" notation="string">13197-741962</idno>
                  <idno type="ARXIV">https://arxiv.org/a/pellegrini_t_1</idno>
                  <idno type="ORCID">https://orcid.org/0000-0001-8984-1399</idno>
                  <idno type="IDREF">https://www.idref.fr/127577955</idno>
                  <affiliation ref="#struct-393117"/>
                </author>
                <author role="aut">
                  <persName>
                    <forename type="first">Léo</forename>
                    <surname>Cances</surname>
                  </persName>
                  <idno type="halauthorid">2028272-0</idno>
                  <affiliation ref="#struct-393117"/>
                </author>
              </analytic>
              <monogr>
                <title level="m">Proceedings of IJCNN 2019</title>
                <meeting>
                  <title>International Joint Conference on Neural Networks (IJCNN 2019)</title>
                  <date type="start">2019-07-14</date>
                  <date type="end">2019-07-19</date>
                  <settlement>Budapest</settlement>
                  <country key="HU">Hungary</country>
                </meeting>
                <imprint>
                  <biblScope unit="pp">1-8</biblScope>
                  <date type="datePub">2019</date>
                </imprint>
              </monogr>
            </biblStruct>
          </sourceDesc>
          <profileDesc>
            <langUsage>
              <language ident="en">English</language>
            </langUsage>
            <textClass>
              <keywords scheme="author">
                <term xml:lang="en">Recurrent convolutional neural networks</term>
                <term xml:lang="en">Multiple instance learning</term>
                <term xml:lang="en">Weakly supervised learning</term>
                <term xml:lang="en">Sound event detection</term>
              </keywords>
              <classCode scheme="halDomain" n="info.info-sd">Computer Science [cs]/Sound [cs.SD]</classCode>
              <classCode scheme="halTypology" n="COMM">Conference papers</classCode>
              <classCode scheme="halOldTypology" n="COMM">Conference papers</classCode>
              <classCode scheme="halTreeTypology" n="COMM">Conference papers</classCode>
            </textClass>
            <abstract xml:lang="en">
              <p>The design of new methods and models when only weakly-labeled data are available is of paramount importance in order to reduce the costs of manual annotation and the considerable human effort associated with it. In this work, we address Sound Event Detection in the case where a weakly annotated dataset is available for training. The weak annotations provide tags of audio events but do not provide temporal boundaries. The objective is twofold: 1) audio tagging, i.e. multi-label classification at recording level, 2) sound event detection, i.e. localization of the event boundaries within the recordings. This work focuses mainly on the second objective. We explore an approach inspired by Multiple Instance Learning, in which we train a convolutional recurrent neural network to give predictions at frame-level, using a custom loss function based on the weak labels and the statistics of the frame-based predictions. Since some sound classes cannot be distinguished with this approach, we improve the method by penalizing similarity between the predictions of the positive classes during training. On the test set used in the DCASE 2018 challenge, consisting of 288 recordings and 10 sound classes, the addition of a penalty resulted in a localization F-score of 34.75%, and brought 10% relative improvement compared to not using the penalty. Our best model achieved a 26.20% F-score on the DCASE-2018 official Eval subset close to the 10-system ensemble approach that ranked second in the challenge with a 29.9% F-score.</p>
            </abstract>
          </profileDesc>
        </biblFull>
      </listBibl>
    </body>
    <back>
      <listOrg type="structures">
        <org type="researchteam" xml:id="struct-393117" status="OLD">
          <orgName>Équipe Structuration, Analyse et MOdélisation de documents Vidéo et Audio</orgName>
          <orgName type="acronym">IRIT-SAMoVA</orgName>
          <date type="end">2025-01-01</date>
          <desc>
            <address>
              <addrLine>IRIT118 Route de Narbonne31062 Toulouse Cedex 9</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.irit.fr/SAMOVA/site/</ref>
          </desc>
          <listRelation>
            <relation active="#struct-34499" type="direct"/>
            <relation active="#struct-81148" type="indirect"/>
            <relation name="443875" active="#struct-443875" type="indirect"/>
            <relation active="#struct-116256" type="indirect"/>
            <relation active="#struct-443875" type="indirect"/>
            <relation active="#struct-217752" type="indirect"/>
            <relation name="UMR5505 / URA1399" active="#struct-441569" type="indirect"/>
            <relation active="#struct-448187" type="indirect"/>
            <relation active="#struct-1149079" type="indirect"/>
            <relation name="FED4171" active="#struct-217752" type="indirect"/>
          </listRelation>
        </org>
        <org type="laboratory" xml:id="struct-34499" status="OLD">
          <idno type="IdRef">030947448</idno>
          <idno type="ISNI">0000000121927225</idno>
          <idno type="RNSR">199511949P</idno>
          <idno type="ROR">https://ror.org/01rx4qw44</idno>
          <orgName>Institut de recherche en informatique de Toulouse</orgName>
          <orgName type="acronym">IRIT</orgName>
          <date type="start">1990-01-01</date>
          <date type="end">2025-01-01</date>
          <desc>
            <address>
              <addrLine>118 Route de Narbonne, F-31062 Toulouse Cedex 9</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.irit.fr/</ref>
          </desc>
          <listRelation>
            <relation active="#struct-81148" type="direct"/>
            <relation name="443875" active="#struct-443875" type="indirect"/>
            <relation active="#struct-116256" type="direct"/>
            <relation active="#struct-443875" type="indirect"/>
            <relation active="#struct-217752" type="direct"/>
            <relation name="UMR5505 / URA1399" active="#struct-441569" type="direct"/>
            <relation active="#struct-448187" type="direct"/>
            <relation active="#struct-1149079" type="direct"/>
            <relation name="FED4171" active="#struct-217752" type="direct"/>
          </listRelation>
        </org>
        <org type="institution" xml:id="struct-81148" status="VALID">
          <idno type="IdRef">026404354</idno>
          <idno type="ISNI">0000000121902837</idno>
          <idno type="ROR">https://ror.org/0443n9e75</idno>
          <orgName>Université Toulouse Capitole</orgName>
          <orgName type="acronym">UT Capitole</orgName>
          <date type="start">1970-01-01</date>
          <desc>
            <address>
              <addrLine>2 rue du Doyen-Gabriel-Marty - 31042 Toulouse Cedex 9</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.ut-capitole.fr/</ref>
          </desc>
          <listRelation>
            <relation name="443875" active="#struct-443875" type="direct"/>
          </listRelation>
        </org>
        <org type="regroupinstitution" xml:id="struct-443875" status="VALID">
          <idno type="ROR">https://ror.org/017tgbk05</idno>
          <orgName>Communauté d'universités et établissements de Toulouse</orgName>
          <orgName type="acronym">Comue de Toulouse</orgName>
          <desc>
            <address>
              <addrLine>41 Allée Jules Guesde, 31000 Toulouse</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.univ-toulouse.fr/</ref>
          </desc>
        </org>
        <org type="institution" xml:id="struct-116256" status="VALID">
          <idno type="IdRef">026403994</idno>
          <idno type="ROR">https://ror.org/04ezk3x31</idno>
          <orgName>Université Toulouse - Jean Jaurès</orgName>
          <orgName type="acronym">UT2J</orgName>
          <date type="start">1971-01-01</date>
          <desc>
            <address>
              <addrLine>5 allées Antonio Machado - 31058 Toulouse Cedex 9</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.univ-tlse2.fr</ref>
          </desc>
          <listRelation>
            <relation active="#struct-443875" type="direct"/>
          </listRelation>
        </org>
        <org type="institution" xml:id="struct-217752" status="OLD">
          <idno type="IdRef">026404672</idno>
          <idno type="ISNI">0000000121617331</idno>
          <idno type="ROR">https://ror.org/02v6kpv12</idno>
          <idno type="Wikidata">Q1273188</idno>
          <orgName>Université Toulouse III - Paul Sabatier</orgName>
          <orgName type="acronym">UT3</orgName>
          <date type="end">2025-01-01</date>
          <desc>
            <address>
              <addrLine>118 route de Narbonne - 31062 Toulouse</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">http://www.univ-tlse3.fr/</ref>
          </desc>
          <listRelation>
            <relation active="#struct-443875" type="direct"/>
          </listRelation>
        </org>
        <org type="regroupinstitution" xml:id="struct-441569" status="VALID">
          <idno type="IdRef">02636817X</idno>
          <idno type="ISNI">0000000122597504</idno>
          <idno type="ROR">https://ror.org/02feahw73</idno>
          <orgName>Centre National de la Recherche Scientifique</orgName>
          <orgName type="acronym">CNRS</orgName>
          <date type="start">1939-10-19</date>
          <desc>
            <address>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.cnrs.fr/</ref>
          </desc>
        </org>
        <org type="institution" xml:id="struct-448187" status="VALID">
          <idno type="IdRef">026388820</idno>
          <idno type="ISNI">0000000121766169</idno>
          <idno type="ROR">https://ror.org/033p9g875</idno>
          <orgName>Institut National Polytechnique (Toulouse)</orgName>
          <orgName type="acronym">Toulouse INP</orgName>
          <desc>
            <address>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.inp-toulouse.fr/</ref>
          </desc>
          <listRelation>
            <relation active="#struct-443875" type="direct"/>
          </listRelation>
        </org>
        <org type="regrouplaboratory" xml:id="struct-1149079" status="OLD">
          <orgName>Toulouse Mind &amp; Brain Institut</orgName>
          <orgName type="acronym">TMBI</orgName>
          <date type="start">2021-01-01</date>
          <date type="end">2025-01-01</date>
          <desc>
            <address>
              <addrLine>CHU Purpan, Pavillon Baudot, 31059 Toulouse cedex 9</addrLine>
              <country key="FR"/>
            </address>
            <ref type="url">https://www.tmbi.fr/</ref>
          </desc>
          <listRelation>
            <relation active="#struct-116256" type="direct"/>
            <relation active="#struct-443875" type="indirect"/>
            <relation name="FED4171" active="#struct-217752" type="direct"/>
          </listRelation>
        </org>
      </listOrg>
    </back>
  </text>
</TEI>