Music identification with audio fingerprinting an industrial perspective

Music identification is a mature and well-studied field in the Music Information Retrieval community. In the music industry, it ensures fair distribution of royalties, which are allocated based on usage, such as plays in live venues or airtime in broadcasts. This thesis has been conducted as part of...

ver descrição completa

Detalhes bibliográficos
Autor: Cortès Sebastià, Guillem
Formato: tesis doctoral
Estado:Versión publicada
Fecha de publicación:2025
País:España
Recursos:CBUC, CESCA
Repositorio:TDR. Tesis Doctorales en Red
OAI Identifier:oai:www.tdx.cat:10803/694234
Acesso em linha:http://hdl.handle.net/10803/694234
Access Level:acceso abierto
Palavra-chave:Audio fingerprinting
Music identification
Self-supervised learning
Identificació de música
Aprenentatge auto-supervisat
Empremtes digitals d’àudio
62
id ES_fb2f2edbda3154adb8573c5cc6b58140
oai_identifier_str oai:www.tdx.cat:10803/694234
network_acronym_str ES
network_name_str España
repository_id_str
dc.title.none.fl_str_mv Music identification with audio fingerprinting an industrial perspective
title Music identification with audio fingerprinting an industrial perspective
spellingShingle Music identification with audio fingerprinting an industrial perspective
Cortès Sebastià, Guillem
Audio fingerprinting
Music identification
Self-supervised learning
Identificació de música
Aprenentatge auto-supervisat
Empremtes digitals d’àudio
62
title_short Music identification with audio fingerprinting an industrial perspective
title_full Music identification with audio fingerprinting an industrial perspective
title_fullStr Music identification with audio fingerprinting an industrial perspective
title_full_unstemmed Music identification with audio fingerprinting an industrial perspective
title_sort Music identification with audio fingerprinting an industrial perspective
dc.creator.none.fl_str_mv Cortès Sebastià, Guillem
author Cortès Sebastià, Guillem
author_facet Cortès Sebastià, Guillem
author_role author
dc.contributor.none.fl_str_mv Molina Martínez, Emilio
Serra Casals, Xavier
Universitat Pompeu Fabra. Departament de Tecnologies de la Informació i les Comunicacions
dc.subject.none.fl_str_mv Audio fingerprinting
Music identification
Self-supervised learning
Identificació de música
Aprenentatge auto-supervisat
Empremtes digitals d’àudio
62
topic Audio fingerprinting
Music identification
Self-supervised learning
Identificació de música
Aprenentatge auto-supervisat
Empremtes digitals d’àudio
62
description Music identification is a mature and well-studied field in the Music Information Retrieval community. In the music industry, it ensures fair distribution of royalties, which are allocated based on usage, such as plays in live venues or airtime in broadcasts. This thesis has been conducted as part of an industrial PhD at BMAT, a company specializing in music monitoring and identification services. This thesis explores advancements in Audio Fingerprinting (AFP), a core technology for music identification that identifies audio by matching compact signatures extracted from audio signals. From their early development in the 2000s, AFP systems have evolved to address challenges such as robustness to time-frequency modifications, or noise and speech overlays, for instance. However, scenarios like background music identification or extreme time-frequency modifications remain challenging for these systems. To address these gaps, this thesis first introduces a self-contained dataset specifically designed for broadcast monitoring, featuring TV recordings with a high prevalence of background music and reference tracks of production music. Alongside this dataset, it proposes \emph{PeakFP}, a new baseline method tailored for background music identification. To improve the AFP performance, this thesis explores a two-step approach combining source separation algorithms with AFP systems. This approach demonstrates substantial performance improvements in background music identification, albeit at the cost of computational overhead. Finally, this thesis presents PeakNetFP, the first hybrid AFP system that integrates the simplicity and scalability of spectral peaks with the abstraction capabilities of neural networks. PeakNetFP achieves comparable performance to state-of-the-art models while being 100 times smaller, offering a scalable and efficient solution for AFP tasks, including severe time-stretched audio. Despite being conducted in an industrial setting, this work adheres to the principles of open science, with all datasets, code, and evaluations made publicly available. This thesis aims to foster further research in AFP, particularly in underexplored scenarios, and to contribute to the development of more robust and versatile AFP systems.
publishDate 2025
dc.date.none.fl_str_mv 2025
2025
2025
dc.type.none.fl_str_mv info:eu-repo/semantics/doctoralThesis
info:eu-repo/semantics/publishedVersion
format doctoralThesis
status_str publishedVersion
dc.identifier.none.fl_str_mv http://hdl.handle.net/10803/694234
url http://hdl.handle.net/10803/694234
dc.language.none.fl_str_mv Inglés
language_invalid_str_mv Inglés
dc.rights.none.fl_str_mv http://creativecommons.org/licenses/by-nc-sa/4.0/
info:eu-repo/semantics/openAccess
rights_invalid_str_mv http://creativecommons.org/licenses/by-nc-sa/4.0/
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv 165 p.
application/pdf
dc.publisher.none.fl_str_mv Universitat Pompeu Fabra
publisher.none.fl_str_mv Universitat Pompeu Fabra
dc.source.none.fl_str_mv TDX (Tesis Doctorals en Xarxa)
reponame:TDR. Tesis Doctorales en Red
instname:CBUC, CESCA
instname_str CBUC, CESCA
reponame_str TDR. Tesis Doctorales en Red
collection TDR. Tesis Doctorales en Red
repository.name.fl_str_mv
repository.mail.fl_str_mv
_version_ 1869425306475954176
spelling Music identification with audio fingerprinting an industrial perspectiveCortès Sebastià, GuillemAudio fingerprintingMusic identificationSelf-supervised learningIdentificació de músicaAprenentatge auto-supervisatEmpremtes digitals d’àudio62Music identification is a mature and well-studied field in the Music Information Retrieval community. In the music industry, it ensures fair distribution of royalties, which are allocated based on usage, such as plays in live venues or airtime in broadcasts. This thesis has been conducted as part of an industrial PhD at BMAT, a company specializing in music monitoring and identification services. This thesis explores advancements in Audio Fingerprinting (AFP), a core technology for music identification that identifies audio by matching compact signatures extracted from audio signals. From their early development in the 2000s, AFP systems have evolved to address challenges such as robustness to time-frequency modifications, or noise and speech overlays, for instance. However, scenarios like background music identification or extreme time-frequency modifications remain challenging for these systems. To address these gaps, this thesis first introduces a self-contained dataset specifically designed for broadcast monitoring, featuring TV recordings with a high prevalence of background music and reference tracks of production music. Alongside this dataset, it proposes \emph{PeakFP}, a new baseline method tailored for background music identification. To improve the AFP performance, this thesis explores a two-step approach combining source separation algorithms with AFP systems. This approach demonstrates substantial performance improvements in background music identification, albeit at the cost of computational overhead. Finally, this thesis presents PeakNetFP, the first hybrid AFP system that integrates the simplicity and scalability of spectral peaks with the abstraction capabilities of neural networks. PeakNetFP achieves comparable performance to state-of-the-art models while being 100 times smaller, offering a scalable and efficient solution for AFP tasks, including severe time-stretched audio. Despite being conducted in an industrial setting, this work adheres to the principles of open science, with all datasets, code, and evaluations made publicly available. This thesis aims to foster further research in AFP, particularly in underexplored scenarios, and to contribute to the development of more robust and versatile AFP systems.La identificació musical és un àmbit madur i àmpliament estudiat dins la comunitat de Recuperació d'Informació Musical (MIR) des de fa molts anys. En la indústria musical, garanteix una distribució justa de les regalies, que es reparteixen segons l'ús, com ara reproduccions en esdeveniments en directe o temps d'emissió en retransmissions. Aquesta tesi s'ha dut a terme en el marc d'un doctorat industrial a BMAT, una empresa especialitzada en serveis de monitoratge i identificació musical, i explora els avenços en el camp de l'\emph{Audio Fingerprinting (AFP)}, una tecnologia clau per a la identificació musical que reconeix àudios mitjançant la comparació de signatures compactes extretes dels senyals d'àudio. Des del seu desenvolupament inicial als anys 2000, els sistemes AFP han evolucionat per afrontar reptes com la robustesa davant modificacions tempo-freqüencials, soroll i superposicions de veu. No obstant això, escenaris com la identificació de música de fons o els àudios amb estiraments temporals extrems continuen sent un desafiament per a aquests sistemes. Per abordar aquestes limitacions, aquesta tesi presenta primer un conjunt de dades autocontingut dissenyat específicament per al monitoratge de retransmissions, amb enregistraments de televisió amb alta prevalença de música de fons i pistes de música de producció com a referència. Altrament, també proposa un nou mètode base adaptat per a la identificació de música de fons que serveix com a sistema de referència. Per millorar el rendiment dels sistemes AFP, s'explora si les tecnologies existents poden ajudar. En aquest sentit, s'avalua un enfocament en dues fases que combina algoritmes de separació de fonts amb sistemes AFP. Aquest mètode mostra millores substancials en la identificació de música de fons, tot i que en alguns casos implica un cost computacional significatiu. Finalment, aquesta tesi presenta PeakNetFP, el primer sistema AFP híbrid que integra la simplicitat i escalabilitat dels pics espectrals amb les capacitats d'abstracció de les xarxes neuronals. PeakNetFP aconsegueix un rendiment comparable als models més avançats del moment, amb una mida 100 vegades menor, oferint una solució escalable i eficient per a tasques AFP, incloent-hi àudios amb distorsions temporals extremes. Tot i que la tesi s'ha desenvolupat en un entorn industrial, segueix els principis de la ciència oberta, amb tots els conjunts de dades, codi i avaluacions disponibles públicament. Aquesta tesi té com a objectiu fomentar la investigació futura en AFP, especialment en escenaris poc explorats, i contribuir al desenvolupament de sistemes AFP més robusts i versàtils.Programa de Doctorat en Tecnologies de la Informació i les ComunicacionsUniversitat Pompeu FabraMolina Martínez, EmilioSerra Casals, XavierUniversitat Pompeu Fabra. Departament de Tecnologies de la Informació i les Comunicacions202520252025info:eu-repo/semantics/doctoralThesisinfo:eu-repo/semantics/publishedVersion165 p.application/pdfhttp://hdl.handle.net/10803/694234TDX (Tesis Doctorals en Xarxa)reponame:TDR. Tesis Doctorales en Redinstname:CBUC, CESCAInglésL'accés als continguts d'aquesta tesi queda condicionat a l'acceptació de les condicions d'ús establertes per la següent llicència Creative Commons: http://creativecommons.org/licenses/by-nc-sa/4.0/http://creativecommons.org/licenses/by-nc-sa/4.0/info:eu-repo/semantics/openAccessoai:www.tdx.cat:10803/6942342026-06-14T12:46:07Z
score 15.812429

Registos relacionados