MetricGAN+: An Improved Version of MetricGAN for Speech Enhancement

By Szu-Wei Fu, Cheng Yu, Tsun-An Hsieh, Peter William VanHarn Plantinga, Mirco Ravanelli, Xugang Lu, Yu Tsao. Interspeech, 2021

305 citations recorded by the source corpus.

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).

Subfields: Artificial Intelligence, Electrical and Electronic Engineering, Signal Processing

View Paper: https://www.semanticscholar.org/paper/6af96d82d7ba1f4945dc208b6b0ab527bc6d8874

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Electrical and Electronic Engineering

In Signal Processing

Citations in the corpus

Generative Pre-training for Speech with Flow Matching (2023, methodology): “MetricGAN+ (Fu et al., 2021) proposed to optimize non-differentiable metrics such as PESQ via adversarial training against their approximation using discriminators.”
Universal Speech Enhancement with Score-based Diffusion (2022, methodology): “…Demucs (Défossez et al., 2020) 3.07 3.63 0.95 SN-Net (Zheng et al., 2021) 3.12 3.60 3.39 SE-Conformer (Kim & Seo, 2021) 3.12 3.82 0.95 MetricGAN+ (Fu et al., 2021) 3.15 3.64 PERL-AE (Kataria et al., 2021) 3.17 3.83 0.95 HiFi-GAN-2 (Su et al., 2021) 3 Table 4 shows the results for the denoising…”
HiFi++: a Unified Framework for Neural Vocoding, Bandwidth Extension and Speech Enhancement (2022, methodology): “An interesting observation is the performance of the Metri-GAN+ model (Fu et al., 2021).”
Speech Enhancement and Dereverberation With Diffusion-Based Generative Models (2022, unknown): “In fact, listening to the enhanced files, it can be recognized that the energy of the speech signal estimated by MetricGAN+ is mainly concentrated in the low-and mid-frequency area of the spectrogram, while high frequencies are strongly attenuated.”
SpeechBrain: A General-Purpose Speech Toolkit (2021, methodology): “In addition, it supports a variety of more sophisticated multi-model training techniques such as Mimic Loss [40] and MetricGAN+ [39].”