Acoustic scene classification: An evaluation of an extremely compact feature representation

Gustavo Sena Mafra (1) , Ngoc Q K Duong (2) , Alexey Ozerov (2) , Patrick Pérez (2)

1 UFSC - Universidade Federal de Santa Catarina = Federal University of Santa Catarina [Florianópolis]
2 Technicolor R & I [Cesson Sévigné]

This paper investigates several approaches to address the acoustic scene classification (ASC) task. We start from low-level feature representation for segmented audio frames and investigate different time granularity for feature aggregation. We study the use of support vector machine (SVM), as a well-known classifier, together with two popular neural network (NN) architectures, namely mul-tilayer perceptron (MLP) and convolutional neural network (CNN). We evaluate the performance of these approaches on benchmark datasets provided from the 2013 and 2016 Detection and Classification of Acoustic Scenes and Events (DCASE) challenges. We observe that a simple approach exploiting averaged Mel-log-spectrograms and SVM can obtain even better results than NN-based approaches and comparable performance with the best systems in the DCASE 2013 challenge.

Mots clés: Acoustic scene classification; Audio features; Multilayer Perceptron; Convolutional Neural Network; Support Vector Machine

Domaines: Traitement du signal et de l'image [eess.SP]

Gustavo Sena Mafra, Ngoc Q K Duong, Alexey Ozerov, Patrick Pérez. Acoustic scene classification: An evaluation of an extremely compact feature representation. Detection and Classification of Acoustic Scenes and Events 2016, Sep 2016, Budapest, Hungary. ⟨hal-01400986⟩