Denoising applied to sound event detection using weighting non-negative matrix factorization

Some sound events (screams, gunshots ...) are often associated with situations of danger and violence. Nevertheless, the noise interference that appears with them decreases the detection performance to a great extent. Most existing methods that employ a trained noise classifier have the problem that...

Descripción completa

Detalles Bibliográficos
Autor: Romero de los Ríos, Andrés
Tipo de recurso: tesis de maestría
Fecha de publicación:2021
País:España
Institución:Universidad de Jaén (UJA)
Repositorio:CREA. Colección de recursos educativos abiertos
OAI Identifier:oai:crea.ujaen.es:10953.1/13604
Acceso en línea:https://hdl.handle.net/10953.1/13604
Access Level:acceso abierto
Palabra clave:3325
Tecnología de las telecomunicaciones
Telecommunications Technology
Descripción
Sumario:Some sound events (screams, gunshots ...) are often associated with situations of danger and violence. Nevertheless, the noise interference that appears with them decreases the detection performance to a great extent. Most existing methods that employ a trained noise classifier have the problem that they cannot provide an optimal result when such noise does not resemble the trained noise or is highly variable in time. Therefore, the task of acoustic noise reduction in this acoustic scenario remains a great challenge for sound event detec-tion. For this Master Thesis, we propose the design and development of a system imple-mented in MATLAB capable of reducing noise for the detection of sound events using a Weighted Non-Negative Matrix factorization (WNMF) approach.