Recognition of acoustic events using deep neural networks

Author

Gencoglu, Oguzhan ; Virtanen, Tuomas ; Huttunen, Heikki

Author_Institution

Dept. of Signal Process., Tampere Univ. of Technol., Tampere, Finland

fYear

2014

fDate

1-5 Sept. 2014

Firstpage

506

Lastpage

510

Abstract

This paper proposes the use of a deep neural network for the recognition of isolated acoustic events such as footsteps, baby crying, motorcycle, rain etc. For an acoustic event classification task containing 61 distinct classes, classification accuracy of the neural network classifier (60.3%) excels that of the conventional Gaussian mixture model based hidden Markov model classifier (54.8%). In addition, an unsupervised layer-wise pretraining followed by standard backpropagation training of a deep network (known as a deep belief network) results in further increase of 2-4% in classification accuracy. Effects of implementation parameters such as types of features and number of adjacent frames as additional features are found to be significant on classification accuracy.

Keywords

Gaussian processes; acoustic signal processing; backpropagation; hidden Markov models; mixture models; neural nets; signal classification; unsupervised learning; Gaussian mixture model; acoustic event classification task; acoustic event recognition; adjacent frames; deep belief network; deep neural networks; hidden Markov model classifier; neural network classifier; standard backpropagation training; unsupervised layer-wise pretraining; Accuracy; Acoustics; Artificial neural networks; Feature extraction; Hidden Markov models; Training; acoustic event classification; artificial neural networks; deep belief networks; deep neural networks; pattern classification;

fLanguage

English

Publisher

ieee

Conference_Titel

Signal Processing Conference (EUSIPCO), 2014 Proceedings of the 22nd European

Conference_Location

Lisbon

Type

conf

Filename

6952140