Mixup-Based Acoustic Scene Classification Using Multi-channel Convolutional Neural Network

Abstract

Audio scene classification, the problem of predicting class labels of audio scenes, has drawn lots of attention during the last several years. However, it remains challenging and falls short of accuracy and efficiency. Recently, Convolutional Neural Network (CNN)-based methods have achieved better performance with comparison to the traditional methods. Nevertheless, conventional single channel CNN may fail to consider the fact that additional cues may be embedded in the multi-channel recordings. In this paper, we explore the use of Multi-channel CNN for the classification task, which aims to extract features from different channels in an end-to-end manner. We conduct the evaluation compared with the conventional CNN and traditional Gaussian Mixture Model-based methods. Moreover, to improve the classification accuracy further, this paper explores the using of mixup method. In brief, mixup trains the neural network on linear combinations of pairs of the representation of audio scene examples and their labels. By employing the mixup approach for data augmentation, the novel model can provide higher prediction accuracy and robustness in contrast with previous models, while the generalization error can also be reduced on the evaluation data.

Multiclass audio segmentation based on recurrent neural networks for broadcast domain data

“…It is important to note that the use of mixup augmentation leads to no addition of more external datasets, so the proposed system is still under the conditions imposed by the Albayzín 2010 evaluation. Mixup augmentation has been shown to improve model generalisation capabilities in different domains, including some audio classification tasks [64]. In our set of experiments, mixup augmentation is applied directly in the feature space.…”

Spatio-Temporal Attention Pooling for Audio Scene Classification

“…In deep learning, data augmentation, which is to increase the data variation by altering the property of the genuine data, is an important method to improve performance of the task at hand. Techniques like adding background noise [28,29], pitch shifting [29], and sample mixing [18,30], have been proven to be useful for environmental sound recognition in general. Motivated by the work of Tokozume et al [18], we pursue a between-class (BC) data augmentation approach that mixes two samples of different classes with a random factor to generate BC examples for network training.…”

General audio tagging with ensembling convolutional neural networks and statistical features

“…MFCC follows similar generation procedure, except for the size. To prevent over-fitting, we apply mixup-data augmentation [22] with an ratio of 0.2 [23].…”