Home // MMEDIA 2015, The Seventh International Conferences on Advances in Multimedia // View article

Audio Event Detection Using Adaptive Feature Extraction Scheme

Authors:
Selver Ezgi Kucukbay
Mustafa Sert

Keywords: Audio event detection; Audio content analysis; Environmental sound detection; MFCC; SVM;

Abstract:
Audio event detection is one of the important tasks of multimedia content analysis. The noise like characteristics and the diversity of audio events make the recognition task difficult when compared with music and speech sounds. Therefore, proper application of feature extraction methods is very crucial, as well as feature selection and machine learning algorithms. Here, we propose a novel adaptive feature extraction scheme along with Support Vector Machine (SVM) learner in recognizing audio events. In our scheme, we propose to apply the widely used Mel frequency cepstral coefficients (MFCCs) feature to the problem in an adaptive way. To this end, we analyze each audio event in its frequency space to obtain a dominant frequency and then make use of the determined dominant frequency in the feature extraction phase. Extensive experiments have been conducted on sixteen (16) different audio events namely alert, clear throat, cough, door slam, drawer, keyboard, keys, knock, laughter, mouse, page turn, pen drop, phone, printer, speech, and switch using the IEEE AASP CASA Challenge Dataset to demonstrate the performance of the proposed scheme. The results show that our adaptive feature extraction scheme achieves significantly higher recognition accuracy than traditional feature extraction method with an average F-measure value of 72%.

Pages: 44 to 49

Copyright: Copyright (c) IARIA, 2015

Publication date: April 19, 2015

Published in: conference

ISSN: 2308-4448

ISBN: 978-1-61208-400-8

Location: Barcelona, Spain

Dates: from April 19, 2015 to April 24, 2015