MetricGAN+: An Improved Version of MetricGAN for Speech Enhancement

TL;DR

This study proposes a MetricGAN+ in which three training techniques incorporating domain-knowledge of speech processing are proposed and can increase PESQ score by 0.3 compared to the previous metricGAN and achieve state-of-the-art results.

Abstract

The discrepancy between the cost function used for training a speech enhancement model and human auditory perception usually makes the quality of enhanced speech unsatisfactory. Objective evaluation metrics which consider human perception can hence serve as a bridge to reduce the gap. Our previously proposed MetricGAN was designed to optimize objective metrics by connecting the metric with a discriminator. Because only the scores of the target evaluation functions are needed during training, the metrics can even be non-differentiable. In this study, we propose a MetricGAN+ in which three training techniques incorporating domain-knowledge of speech processing are proposed. With these techniques, experimental results on the VoiceBank-DEMAND dataset show that MetricGAN+ can increase PESQ score by 0.3 compared to the previous MetricGAN and achieve state-of-the-art results (PESQ score = 3.15).

Figures (4)

• Figure 1: Training flow of MetricGAN.
• Figure 2: Sigmoid function with different$\alpha$.
• Figure 3: Learned values of$\alpha$ in learnable sigmoid function.
• Figure 4: Learning curves of different settings (structure of$G$ is fixed).