English

MetricGAN: Generative Adversarial Networks based Black-box Metric Scores Optimization for Speech Enhancement

Sound 2019-05-14 v1 Machine Learning Audio and Speech Processing

Abstract

Adversarial loss in a conditional generative adversarial network (GAN) is not designed to directly optimize evaluation metrics of a target task, and thus, may not always guide the generator in a GAN to generate data with improved metric scores. To overcome this issue, we propose a novel MetricGAN approach with an aim to optimize the generator with respect to one or multiple evaluation metrics. Moreover, based on MetricGAN, the metric scores of the generated data can also be arbitrarily specified by users. We tested the proposed MetricGAN on a speech enhancement task, which is particularly suitable to verify the proposed approach because there are multiple metrics measuring different aspects of speech signals. Moreover, these metrics are generally complex and could not be fully optimized by Lp or conventional adversarial losses.

Keywords

Cite

@article{arxiv.1905.04874,
  title  = {MetricGAN: Generative Adversarial Networks based Black-box Metric Scores Optimization for Speech Enhancement},
  author = {Szu-Wei Fu and Chien-Feng Liao and Yu Tsao and Shou-De Lin},
  journal= {arXiv preprint arXiv:1905.04874},
  year   = {2019}
}

Comments

Accepted by Thirty-sixth International Conference on Machine Learning (ICML) 2019