Thesis Exploración de modelos de aprendizaje para analizar señales de audio
Loading...
Date
2024
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Informática
Departament
Campus
Campus Santiago San Joaquín
Abstract
Este trabajo explora modelos de aprendizaje supervisado y no supervisado para analizar señales de audio, con el objetivo de desarrollar una herramienta que permita cuantificar objetivamente la calidad de los archivos de audio. Para abordar este problema se recopilaron más de 2.000 muestras de audio, variando características como la compresión, profundidad de bits y tasa de muestreo para entrenar y validar modelos de aprendizaje supervisados y no supervisados. Los resultados destacan la efectividad del modelo no supervisado K-means para clasificar la calidad de las señales, entregando agrupaciones coherentes y distinguibles entre ellas, sin la necesidad de requerir una etiqueta que cuantifique la calidad de la señal de los archivos de audio. Esta investigación es relevante para usuarios y plataformas que buscan una evaluación precisa y transparente de la calidad de audio en servicios de streaming y descarga digital.
This work explores supervised and unsupervised learning models to analyze audio signals, aiming to develop a tool that objectively quantifies the quality of audio files. To approach this problem, over 2,000 audio samples were collected, varying in characteristics such as compression, bit depth, and sample rate, to train and validate supervised and unsupervised learning models. The results highlight the effectiveness of the unsupervised K-means model in classifying signal quality, delivering coherent and distinguishable clusters without the need for a label that quantifies the signal quality of audio files. This research is relevant for users and platforms seeking accurate and transparent audio quality evaluation in streaming and digital download services.
This work explores supervised and unsupervised learning models to analyze audio signals, aiming to develop a tool that objectively quantifies the quality of audio files. To approach this problem, over 2,000 audio samples were collected, varying in characteristics such as compression, bit depth, and sample rate, to train and validate supervised and unsupervised learning models. The results highlight the effectiveness of the unsupervised K-means model in classifying signal quality, delivering coherent and distinguishable clusters without the need for a label that quantifies the signal quality of audio files. This research is relevant for users and platforms seeking accurate and transparent audio quality evaluation in streaming and digital download services.
Description
Keywords
Aprendizaje automático, Machine learning, Señales de audio, Audio signals, Calidad de audio, Audio quality
