FA-GAN: Artifacts-free and Phase-aware High-fidelity GAN-based Vocoder

TL;DR

FA-GAN is a novel GAN-based vocoder that significantly reduces spectral artifacts and improves speech fidelity by introducing anti-aliased upsampling and phase-aware loss functions, leading to higher quality synthesized speech especially for unseen speakers.

Contribution

The paper proposes FA-GAN, a new vocoder with anti-aliased modules and phase-aware loss, addressing spectral artifacts and enhancing high-fidelity speech synthesis.

Findings

FA-GAN outperforms existing methods in speech quality.

It effectively reduces spectral artifacts.

It generalizes well to unseen speakers.

Abstract

Generative adversarial network (GAN) based vocoders have achieved significant attention in speech synthesis with high quality and fast inference speed. However, there still exist many noticeable spectral artifacts, resulting in the quality decline of synthesized speech. In this work, we adopt a novel GAN-based vocoder designed for few artifacts and high fidelity, called FA-GAN. To suppress the aliasing artifacts caused by non-ideal upsampling layers in high-frequency components, we introduce the anti-aliased twin deconvolution module in the generator. To alleviate blurring artifacts and enrich the reconstruction of spectral details, we propose a novel fine-grained multi-resolution real and imaginary loss to assist in the modeling of phase information. Experimental results reveal that FA-GAN outperforms the compared approaches in promoting audio quality and alleviating spectral artifacts, and exhibits superior performance when applied to unseen speaker scenarios.