Investigation of Densely Connected Convolutional Networks with Domain Adversarial Learning for Noise Robust Speech Recognition

TL;DR

This paper explores the use of densely connected convolutional networks combined with domain adversarial training to enhance noise robustness in speech recognition, demonstrating significant improvements over existing models.

Contribution

It introduces the integration of DenseNets with domain adversarial learning for noise-robust speech recognition, a novel approach in this field.

Findings

DenseNets outperform other neural networks in noise robustness.

Domain adversarial training further enhances robustness against unknown noise.

The combined approach improves speech recognition accuracy under noisy conditions.

Abstract

We investigate densely connected convolutional networks (DenseNets) and their extension with domain adversarial training for noise robust speech recognition. DenseNets are very deep, compact convolutional neural networks which have demonstrated incredible improvements over the state-of-the-art results in computer vision. Our experimental results reveal that DenseNets are more robust against noise than other neural network based models such as deep feed forward neural networks and convolutional neural networks. Moreover, domain adversarial learning can further improve the robustness of DenseNets against both, known and unknown noise conditions.