A Comparative Study in Machine Learning and Audio Features for Kitchen Sounds Recognition



Document title: A Comparative Study in Machine Learning and Audio Features for Kitchen Sounds Recognition
Journal: Computación y sistemas
Database:
System number: 000560678
ISSN: 1405-5546
Authors: 1
1
1
1
Institutions: 1Universidad Autónoma de Chihuahua, Facultad de Ingeniería, México
Year:
Season: Abr-Jun
Volumen: 26
Number: 2
Pages: 603-621
Country: México
Language: Inglés
English abstract For the last decades the work on audio recognition has been directed to speech and music, however, an increasing interest for the classification and recognition of acoustic events is observed for the last years. This poses the challenge to determine the identity of sounds, their sources, and the importance of analysing the context of the scenario where they act. The aim of this paper is focused on evaluating the robustness to retain the characteristic information of an acoustic event against the background noise using audio features in the task of identifying acoustic events from a mixture of sounds that are produced in a kitchen environment. A new database of kitchen sounds was built by us, since in the reviewed literature there is no similar benchmark that allows us to evaluate this issue in conditions of 3 decibels for the signal to noise ratio. In our study, we compared two methods of audio features, Multiband Spectral Entropy Signature (MSES) and Mel Frequency Cepstral Coefficients (MFCC). To evaluate the performance of both MSES and MFCC, we used different classifiers such as Similarity Distance, k-Nearest Neighbors, Support Vector Machines and Artificial Neural Networks (ANN). The results showed that MSES supported with an ANN outperforms any other combination of classifiers with MSES or MFCC for getting a better score.
Keyword: Entropy,
Neural networks,
Mixture of sounds,
MFCC
Full text: Texto completo (Ver HTML) Texto completo (Ver PDF)