GAN-Based Speech Enhancement for Low SNR Using Latent Feature Conditioning

TL;DR

This work proposes DisCoGAN, which is a time-frequency-domain generative adversarial network (GAN) conditioned by the latent features of a discriminative model pre-trained for speech enhancement in low SNR scenarios that achieves superior performance compared to state-of-the-art discriminative methods and also surpasses end-to-end (E2E) trained GAN models.

Abstract

Enhancing speech quality under adverse SNR conditions remains a significant challenge for discriminative deep neural network (DNN)-based approaches. In this work, we propose DisCoGAN, which is a time-frequency-domain generative adversarial network (GAN) conditioned by the latent features of a discriminative model pre-trained for speech enhancement in low SNR scenarios. Our proposed method achieves superior performance compared to state-of-the-arts discriminative methods and also surpasses end-to-end (E2E) trained GAN models. We also investigate the impact of various configurations for conditioning the proposed GAN model with the discriminative model and assess their influence on enhancing speech quality

Figures (2)

• Figure 1: Overview of the proposed SEANet-based discriminatively conditioned GAN (DisCoGAN)
• Figure 2: PESQ and SNR improvement for the generative models (solid line) and discriminative models (dotted line) for low SNR evaluation datasets (Please note that models are ordered in descending parameter count for both generative and discriminative models)