A Fully Convolutional Neural Network for Speech Enhancement

TL;DR

This paper introduces a fully convolutional neural network, R-CED, that effectively enhances speech in noisy environments, especially for hearing aids, by mapping noisy to clean speech spectra with fewer parameters and better performance than recurrent networks.

Contribution

The paper presents R-CED, a convolutional neural network that outperforms recurrent networks in speech enhancement tasks with fewer parameters, suitable for embedded hearing aid devices.

Findings

R-CED is 12 times smaller than recurrent networks.

R-CED achieves better speech enhancement performance.

The model is suitable for embedded systems like hearing aids.

Abstract

In hearing aids, the presence of babble noise degrades hearing intelligibility of human speech greatly. However, removing the babble without creating artifacts in human speech is a challenging task in a low SNR environment. Here, we sought to solve the problem by finding a `mapping' between noisy speech spectra and clean speech spectra via supervised learning. Specifically, we propose using fully Convolutional Neural Networks, which consist of lesser number of parameters than fully connected networks. The proposed network, Redundant Convolutional Encoder Decoder (R-CED), demonstrates that a convolutional network can be 12 times smaller than a recurrent network and yet achieves better performance, which shows its applicability for an embedded system: the hearing aids.