Loading…

Learning Music Audio Representations Via Weak Language Supervision

Audio representations for music information retrieval are typically learned via supervised learning in a task-specific fashion. Although effective at producing state-of-the-art results, this scheme lacks flexibility with respect to the range of applications a model can have and requires extensively...

Full description

Saved in:
Bibliographic Details
Main Authors: Manco, Ilaria, Benetos, Emmanouil, Quinton, Elio, Fazekas, Gyorgy
Format: Conference Proceeding
Language:English
Subjects:
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
cited_by
cites
container_end_page 460
container_issue
container_start_page 456
container_title
container_volume
creator Manco, Ilaria
Benetos, Emmanouil
Quinton, Elio
Fazekas, Gyorgy
description Audio representations for music information retrieval are typically learned via supervised learning in a task-specific fashion. Although effective at producing state-of-the-art results, this scheme lacks flexibility with respect to the range of applications a model can have and requires extensively annotated datasets. In this work, we pose the question of whether it may be possible to exploit weakly aligned text as the only supervisory signal to learn general-purpose music audio representations. To address this question, we design a multimodal architecture for music and language pre-training (MuLaP) optimised via a set of proxy tasks. Weak supervision is provided in the form of noisy natural language descriptions conveying the overall musical content of the track. After pre-training, we transfer the audio backbone of the model to a set of music audio classification and regression tasks. We demonstrate the usefulness of our approach by comparing the performance of audio representations produced by the same audio backbone with different training strategies and show that our pre-training method consistently achieves comparable or higher scores on all tasks and datasets considered. Our experiments also confirm that MuLaP effectively leverages audio-caption pairs to learn representations that are competitive with audio-only and cross-modal self-supervised methods in the literature.
doi_str_mv 10.1109/ICASSP43922.2022.9746996
format conference_proceeding
fullrecord <record><control><sourceid>ieee_CHZPO</sourceid><recordid>TN_cdi_ieee_primary_9746996</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>9746996</ieee_id><sourcerecordid>9746996</sourcerecordid><originalsourceid>FETCH-LOGICAL-i203t-d344bee6e35aee02813bd314c389d831b6ae890cd1c09ee96015dd7f3c8b85d83</originalsourceid><addsrcrecordid>eNotT9tKxDAUjILg7uoX-JIf6HqSk6bJ47qoK1QU6-1tSZuzJV6ypWkF_96CC8PMwwzDDGNcwFIIsJd361VVPSq0Ui4lTGQLpa3VR2wutM4VTNDHbCaxsJmw8H7K5il9AIAplJmxq5JcH0Ns-f2YQsNXow97_kRdT4ni4Iawj4m_BsffyH3y0sV2dC3xauyo_wlpss_Yyc59JTo_6IK93Fw_rzdZ-XA7zSuzIAGHzKNSNZEmzB0RSCOw9ihUg8Z6g6LWjoyFxosGLJHVIHLvix02pjb5lFiwi__eQETbrg_frv_dHv7iH9kzS9o</addsrcrecordid><sourcetype>Publisher</sourcetype><iscdi>true</iscdi><recordtype>conference_proceeding</recordtype></control><display><type>conference_proceeding</type><title>Learning Music Audio Representations Via Weak Language Supervision</title><source>IEEE Xplore All Conference Series</source><creator>Manco, Ilaria ; Benetos, Emmanouil ; Quinton, Elio ; Fazekas, Gyorgy</creator><creatorcontrib>Manco, Ilaria ; Benetos, Emmanouil ; Quinton, Elio ; Fazekas, Gyorgy</creatorcontrib><description>Audio representations for music information retrieval are typically learned via supervised learning in a task-specific fashion. Although effective at producing state-of-the-art results, this scheme lacks flexibility with respect to the range of applications a model can have and requires extensively annotated datasets. In this work, we pose the question of whether it may be possible to exploit weakly aligned text as the only supervisory signal to learn general-purpose music audio representations. To address this question, we design a multimodal architecture for music and language pre-training (MuLaP) optimised via a set of proxy tasks. Weak supervision is provided in the form of noisy natural language descriptions conveying the overall musical content of the track. After pre-training, we transfer the audio backbone of the model to a set of music audio classification and regression tasks. We demonstrate the usefulness of our approach by comparing the performance of audio representations produced by the same audio backbone with different training strategies and show that our pre-training method consistently achieves comparable or higher scores on all tasks and datasets considered. Our experiments also confirm that MuLaP effectively leverages audio-caption pairs to learn representations that are competitive with audio-only and cross-modal self-supervised methods in the literature.</description><identifier>EISSN: 2379-190X</identifier><identifier>EISBN: 1665405406</identifier><identifier>EISBN: 9781665405409</identifier><identifier>DOI: 10.1109/ICASSP43922.2022.9746996</identifier><language>eng</language><publisher>IEEE</publisher><subject>audio and language ; audio representations ; multimodal learning ; Multiple signal classification ; Music ; music information retrieval ; Natural languages ; Signal processing ; Supervised learning ; Task analysis ; Training</subject><ispartof>ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022, p.456-460</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/9746996$$EHTML$$P50$$Gieee$$H</linktohtml><link.rule.ids>309,310,780,784,789,790,23929,23930,25139,27924,54554,54931</link.rule.ids><linktorsrc>$$Uhttps://ieeexplore.ieee.org/document/9746996$$EView_record_in_IEEE$$FView_record_in_$$GIEEE</linktorsrc></links><search><creatorcontrib>Manco, Ilaria</creatorcontrib><creatorcontrib>Benetos, Emmanouil</creatorcontrib><creatorcontrib>Quinton, Elio</creatorcontrib><creatorcontrib>Fazekas, Gyorgy</creatorcontrib><title>Learning Music Audio Representations Via Weak Language Supervision</title><title>ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)</title><addtitle>ICASSP</addtitle><description>Audio representations for music information retrieval are typically learned via supervised learning in a task-specific fashion. Although effective at producing state-of-the-art results, this scheme lacks flexibility with respect to the range of applications a model can have and requires extensively annotated datasets. In this work, we pose the question of whether it may be possible to exploit weakly aligned text as the only supervisory signal to learn general-purpose music audio representations. To address this question, we design a multimodal architecture for music and language pre-training (MuLaP) optimised via a set of proxy tasks. Weak supervision is provided in the form of noisy natural language descriptions conveying the overall musical content of the track. After pre-training, we transfer the audio backbone of the model to a set of music audio classification and regression tasks. We demonstrate the usefulness of our approach by comparing the performance of audio representations produced by the same audio backbone with different training strategies and show that our pre-training method consistently achieves comparable or higher scores on all tasks and datasets considered. Our experiments also confirm that MuLaP effectively leverages audio-caption pairs to learn representations that are competitive with audio-only and cross-modal self-supervised methods in the literature.</description><subject>audio and language</subject><subject>audio representations</subject><subject>multimodal learning</subject><subject>Multiple signal classification</subject><subject>Music</subject><subject>music information retrieval</subject><subject>Natural languages</subject><subject>Signal processing</subject><subject>Supervised learning</subject><subject>Task analysis</subject><subject>Training</subject><issn>2379-190X</issn><isbn>1665405406</isbn><isbn>9781665405409</isbn><fulltext>true</fulltext><rsrctype>conference_proceeding</rsrctype><creationdate>2022</creationdate><recordtype>conference_proceeding</recordtype><sourceid>6IE</sourceid><recordid>eNotT9tKxDAUjILg7uoX-JIf6HqSk6bJ47qoK1QU6-1tSZuzJV6ypWkF_96CC8PMwwzDDGNcwFIIsJd361VVPSq0Ui4lTGQLpa3VR2wutM4VTNDHbCaxsJmw8H7K5il9AIAplJmxq5JcH0Ns-f2YQsNXow97_kRdT4ni4Iawj4m_BsffyH3y0sV2dC3xauyo_wlpss_Yyc59JTo_6IK93Fw_rzdZ-XA7zSuzIAGHzKNSNZEmzB0RSCOw9ihUg8Z6g6LWjoyFxosGLJHVIHLvix02pjb5lFiwi__eQETbrg_frv_dHv7iH9kzS9o</recordid><startdate>20220523</startdate><enddate>20220523</enddate><creator>Manco, Ilaria</creator><creator>Benetos, Emmanouil</creator><creator>Quinton, Elio</creator><creator>Fazekas, Gyorgy</creator><general>IEEE</general><scope>6IE</scope><scope>6IH</scope><scope>CBEJK</scope><scope>RIE</scope><scope>RIO</scope></search><sort><creationdate>20220523</creationdate><title>Learning Music Audio Representations Via Weak Language Supervision</title><author>Manco, Ilaria ; Benetos, Emmanouil ; Quinton, Elio ; Fazekas, Gyorgy</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-i203t-d344bee6e35aee02813bd314c389d831b6ae890cd1c09ee96015dd7f3c8b85d83</frbrgroupid><rsrctype>conference_proceedings</rsrctype><prefilter>conference_proceedings</prefilter><language>eng</language><creationdate>2022</creationdate><topic>audio and language</topic><topic>audio representations</topic><topic>multimodal learning</topic><topic>Multiple signal classification</topic><topic>Music</topic><topic>music information retrieval</topic><topic>Natural languages</topic><topic>Signal processing</topic><topic>Supervised learning</topic><topic>Task analysis</topic><topic>Training</topic><toplevel>online_resources</toplevel><creatorcontrib>Manco, Ilaria</creatorcontrib><creatorcontrib>Benetos, Emmanouil</creatorcontrib><creatorcontrib>Quinton, Elio</creatorcontrib><creatorcontrib>Fazekas, Gyorgy</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/IET 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>Manco, Ilaria</au><au>Benetos, Emmanouil</au><au>Quinton, Elio</au><au>Fazekas, Gyorgy</au><format>book</format><genre>proceeding</genre><ristype>CONF</ristype><atitle>Learning Music Audio Representations Via Weak Language Supervision</atitle><btitle>ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)</btitle><stitle>ICASSP</stitle><date>2022-05-23</date><risdate>2022</risdate><spage>456</spage><epage>460</epage><pages>456-460</pages><eissn>2379-190X</eissn><eisbn>1665405406</eisbn><eisbn>9781665405409</eisbn><abstract>Audio representations for music information retrieval are typically learned via supervised learning in a task-specific fashion. Although effective at producing state-of-the-art results, this scheme lacks flexibility with respect to the range of applications a model can have and requires extensively annotated datasets. In this work, we pose the question of whether it may be possible to exploit weakly aligned text as the only supervisory signal to learn general-purpose music audio representations. To address this question, we design a multimodal architecture for music and language pre-training (MuLaP) optimised via a set of proxy tasks. Weak supervision is provided in the form of noisy natural language descriptions conveying the overall musical content of the track. After pre-training, we transfer the audio backbone of the model to a set of music audio classification and regression tasks. We demonstrate the usefulness of our approach by comparing the performance of audio representations produced by the same audio backbone with different training strategies and show that our pre-training method consistently achieves comparable or higher scores on all tasks and datasets considered. Our experiments also confirm that MuLaP effectively leverages audio-caption pairs to learn representations that are competitive with audio-only and cross-modal self-supervised methods in the literature.</abstract><pub>IEEE</pub><doi>10.1109/ICASSP43922.2022.9746996</doi><tpages>5</tpages></addata></record>
fulltext fulltext_linktorsrc
identifier EISSN: 2379-190X
ispartof ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022, p.456-460
issn 2379-190X
language eng
recordid cdi_ieee_primary_9746996
source IEEE Xplore All Conference Series
subjects audio and language
audio representations
multimodal learning
Multiple signal classification
Music
music information retrieval
Natural languages
Signal processing
Supervised learning
Task analysis
Training
title Learning Music Audio Representations Via Weak Language Supervision
url http://sfxeu10.hosted.exlibrisgroup.com/loughborough?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-12T06%3A26%3A24IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-ieee_CHZPO&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=proceeding&rft.atitle=Learning%20Music%20Audio%20Representations%20Via%20Weak%20Language%20Supervision&rft.btitle=ICASSP%202022%20-%202022%20IEEE%20International%20Conference%20on%20Acoustics,%20Speech%20and%20Signal%20Processing%20(ICASSP)&rft.au=Manco,%20Ilaria&rft.date=2022-05-23&rft.spage=456&rft.epage=460&rft.pages=456-460&rft.eissn=2379-190X&rft_id=info:doi/10.1109/ICASSP43922.2022.9746996&rft.eisbn=1665405406&rft.eisbn_list=9781665405409&rft_dat=%3Cieee_CHZPO%3E9746996%3C/ieee_CHZPO%3E%3Cgrp_id%3Ecdi_FETCH-LOGICAL-i203t-d344bee6e35aee02813bd314c389d831b6ae890cd1c09ee96015dd7f3c8b85d83%3C/grp_id%3E%3Coa%3E%3C/oa%3E%3Curl%3E%3C/url%3E&rft_id=info:oai/&rft_id=info:pmid/&rft_ieee_id=9746996&rfr_iscdi=true