Masked Conditional Neural Networks for Automatic Sound Events Recognition

TL;DR

This paper introduces the Masked Conditional Neural Network (MCLNN), a novel neural architecture that improves environmental sound recognition by focusing on frequency bands and feature combinations, achieving competitive results with fewer parameters.

Contribution

The work presents the MCLNN, which enforces systematic sparsity and frequency shift invariance, enabling more effective sound recognition compared to traditional CNNs.

Findings

MCLNN achieved competitive accuracy on ESC-10 and ESC-50 datasets.

The model used only 12% of the parameters of state-of-the-art CNNs.

No data augmentation was needed for strong performance.

Abstract

Deep neural network architectures designed for application domains other than sound, especially image recognition, may not optimally harness the time-frequency representation when adapted to the sound recognition problem. In this work, we explore the ConditionaL Neural Network (CLNN) and the Masked ConditionaL Neural Network (MCLNN) for multi-dimensional temporal signal recognition. The CLNN considers the inter-frame relationship, and the MCLNN enforces a systematic sparseness over the network's links to enable learning in frequency bands rather than bins allowing the network to be frequency shift invariant mimicking a filterbank. The mask also allows considering several combinations of features concurrently, which is usually handcrafted through exhaustive manual search. We applied the MCLNN to the environmental sound recognition problem using the ESC-10 and ESC-50 datasets. MCLNN achieved competitive performance, using 12% of the parameters and without augmentation, compared to state-of-the-art Convolutional Neural Networks.