Deep Within-Class Covariance Analysis for Acoustic Scene Classification

Within-Class Covariance Normalization (WCCN) is a powerful post-processing method for normalizing the within-class covariance of a set of data points. WCCN projects the observations into a linear sub-space where the within-class variability is reduced. This property has proven to be beneficial in subsequent recognition tasks. The central idea of this paper is to reformulate the classic WCCN as a Deep Neural Network (DNN) compatible version. We propose the Deep WithinClass Covariance Analysis (DWCCA) which can be incorporated in a DNN architecture. This formulation enables us to exploit the beneficial properties of WCCN, and still allows for training with Stochastic Gradient Descent (SGD) in an end-to-end fashion. We investigate the advantages of DWCCA on deep neural networks with convolutional layers for supervised learning. Our results on Acoustic Scene Classification show that via DWCCA we can achieves equal or superior performance in a VGG-style deep neural network.