Automated tagging of music according to mood

Music libraries are constantly growing, often tagged in relation to its instrumentation or artist. An emerging trend is the annotation of music according to its emotionally affective features, but the tools and methods used in annotating music remain the same, making it increasingly difficult to loc...

Full description

Saved in:
Bibliographic Details
Main Authors: Chua, Carina Natalia, Domagas, Michael Lance, Lim, John Raymond C., Partosa, John Raymond
Format: text
Language:English
Published: Animo Repository 2012
Subjects:
Online Access:https://animorepository.dlsu.edu.ph/etd_bachelors/11184
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: De La Salle University
Language: English
id oai:animorepository.dlsu.edu.ph:etd_bachelors-11829
record_format eprints
spelling oai:animorepository.dlsu.edu.ph:etd_bachelors-118292022-11-14T06:40:55Z Automated tagging of music according to mood Chua, Carina Natalia Domagas, Michael Lance Lim, John Raymond C. Partosa, John Raymond Music libraries are constantly growing, often tagged in relation to its instrumentation or artist. An emerging trend is the annotation of music according to its emotionally affective features, but the tools and methods used in annotating music remain the same, making it increasingly difficult to locate or recall a specific song for certain events. The approach presented here extracts musical features from a music file and an emotive classification of the song based on a classification model, which can then be used in conjunction with other components, such as a music recommendation system. A dataset of 546 songs tagged by a group of 4 people using a valence- arousal scale of -1 to +1 was used in training models with different classifier algorithms such as multilayer perception and different implementations of regression. Results for valence classification show a root mean square error of 0.3016 while arousal classification is at 0.3498. Overall error, calculated as the Euclidean distance between valence and arousal on a plane is an average of 0.6164 and a median o 0.5926. Some of the discriminant music features were identified to be the song spectral moments, linear predictive coding coefficients, and zero-crossings rate. These results show that while music mood classification through purely music features is feasible, it proves to be a difficult task for only musical features, and the inclusion of lyrics and establishment of the listeners cultural context in relation to the music are likely key in improving classifier performance. 2012-01-01T08:00:00Z text https://animorepository.dlsu.edu.ph/etd_bachelors/11184 Bachelor's Theses English Animo Repository Automatic musical dictation Computer Science--Information Retrieval, Computer Sciences
institution De La Salle University
building De La Salle University Library
continent Asia
country Philippines
Philippines
content_provider De La Salle University Library
collection DLSU Institutional Repository
language English
topic Automatic musical dictation
Computer Science--Information Retrieval,
Computer Sciences
spellingShingle Automatic musical dictation
Computer Science--Information Retrieval,
Computer Sciences
Chua, Carina Natalia
Domagas, Michael Lance
Lim, John Raymond C.
Partosa, John Raymond
Automated tagging of music according to mood
description Music libraries are constantly growing, often tagged in relation to its instrumentation or artist. An emerging trend is the annotation of music according to its emotionally affective features, but the tools and methods used in annotating music remain the same, making it increasingly difficult to locate or recall a specific song for certain events. The approach presented here extracts musical features from a music file and an emotive classification of the song based on a classification model, which can then be used in conjunction with other components, such as a music recommendation system. A dataset of 546 songs tagged by a group of 4 people using a valence- arousal scale of -1 to +1 was used in training models with different classifier algorithms such as multilayer perception and different implementations of regression. Results for valence classification show a root mean square error of 0.3016 while arousal classification is at 0.3498. Overall error, calculated as the Euclidean distance between valence and arousal on a plane is an average of 0.6164 and a median o 0.5926. Some of the discriminant music features were identified to be the song spectral moments, linear predictive coding coefficients, and zero-crossings rate. These results show that while music mood classification through purely music features is feasible, it proves to be a difficult task for only musical features, and the inclusion of lyrics and establishment of the listeners cultural context in relation to the music are likely key in improving classifier performance.
format text
author Chua, Carina Natalia
Domagas, Michael Lance
Lim, John Raymond C.
Partosa, John Raymond
author_facet Chua, Carina Natalia
Domagas, Michael Lance
Lim, John Raymond C.
Partosa, John Raymond
author_sort Chua, Carina Natalia
title Automated tagging of music according to mood
title_short Automated tagging of music according to mood
title_full Automated tagging of music according to mood
title_fullStr Automated tagging of music according to mood
title_full_unstemmed Automated tagging of music according to mood
title_sort automated tagging of music according to mood
publisher Animo Repository
publishDate 2012
url https://animorepository.dlsu.edu.ph/etd_bachelors/11184
_version_ 1751550389909454848