Segmentación automática de letras de canciones populares iberoamericanas para una plataforma de análisis musical
Loading...
Official URL
Full text at PDC
Publication date
2025
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
Ante los avances tecnológicos actuales, surge la necesidad de automatizar el guardado y análisis del estudio de obras musicales. El material que existe hoy en día es completamente artesanal, y estructurarlo a mano es muy costoso. En este contexto surge el proyecto EA-DIGIFOLK, encargado del análisis, la preservación y la divulgación de cultura, sobre todo en el ámbito musical. Como parte del proyecto, aparece la idea de analizar la estructura lírica de las canciones populares de distintas culturas. Por ello, en este estudio se ha investigado y aplicado el uso de LLMs para segmentar estas canciones por versos. Tras realizar una serie de pruebas y analizar las respuestas obtenidas, se puede concluir que esta solución genera resultados favorables, aunque hace falta más experimentación para que el procedimiento sea robusto.
Given current technological advances, there is a growing need to automate the storage and analysis of musical works. The existing material is completely handcrafted and structuring it by hand is very expensive. The EA-DIGIFOLK project was created in this context, tasked with the analysis, preservation, and dissemination of culture, particularly in the musical field. As part of the project, the idea of analyzing the lyrical structure of popular songs from different cultures appeared. We investigated and used LLMs to segment these songs by verse. After testing and analyzing the responses obtained, we can conclude that this solution can be useful, although further experimentation is needed to ensure the procedure is correct.
Given current technological advances, there is a growing need to automate the storage and analysis of musical works. The existing material is completely handcrafted and structuring it by hand is very expensive. The EA-DIGIFOLK project was created in this context, tasked with the analysis, preservation, and dissemination of culture, particularly in the musical field. As part of the project, the idea of analyzing the lyrical structure of popular songs from different cultures appeared. We investigated and used LLMs to segment these songs by verse. After testing and analyzing the responses obtained, we can conclude that this solution can be useful, although further experimentation is needed to ensure the procedure is correct.
Description
Trabajo de Fin de Grado en Desarrollo de Videojuegos, Facultad Informática UCM, Dpto. de Ingeniería de Software e Inteligencia Artificial (ISIA), Curso 2024/2025












