Environmental sound classification using central auditory representations and deep neural networks

Kean Chen, Lixue Yang, Zhiming Sang

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This study presents an effective environmental sound classification method that combines central auditory representations and deep neural networks. The central auditory representation characterizes the responses of sound in primary auditory cortex and is related directly to human auditory perception. Meanwhile, the deep neural network is composed by no less than two hidden layers and is good at detecting high-level invariant features associated with the objects. An environmental sound classification task with 11 distinct classes is used for evaluation, and the timbre features and deep neural networks are combined and used as a baseline. The results show that the proposed method outperforms the baseline, and the highest accuracy of 89.94% is achieved by a two-layer network with 300 hidden nodes each layer.

Original languageEnglish
Title of host publicationICSV 2016 - 23rd International Congress on Sound and Vibration
Subtitle of host publicationFrom Ancient to Modern Acoustics
PublisherInternational Institute of Acoustics and Vibrations
ISBN (Electronic)9789609922623
StatePublished - 2016
Event23rd International Congress on Sound and Vibration, ICSV 2016 - Athens, Greece
Duration: 10 Jul 201614 Jul 2016

Publication series

NameICSV 2016 - 23rd International Congress on Sound and Vibration: From Ancient to Modern Acoustics

Conference

Conference23rd International Congress on Sound and Vibration, ICSV 2016
Country/TerritoryGreece
CityAthens
Period10/07/1614/07/16

Fingerprint

Dive into the research topics of 'Environmental sound classification using central auditory representations and deep neural networks'. Together they form a unique fingerprint.

Cite this