Sound event detection in real life audio using perceptual linear predictive feature with neural network

Abstract

Over the last few years, sound event detection problem has received increasing interest due to its applications in home security, listening through walls and video retrieval etc. [1, 2, 3]. Event detection from sound is a challenging task due to the change in channel behavior and the number of sound sources over time. In this paper, a multi-label classifier is proposed for sound event detection in a polyphonic environment. The proposed classifier uses spectral coefficients of perceptual linear predictive (PLP) feature with convolution neural network. Results show that the proposed classifier using PLP reduces error rate for individual classes as compared to classifiers using other features such as loudness and Mel-Frequency Cepstral coefficients.