GAN Vocoder: Multi-Resolution Discriminator Is All You Need

TL;DR

This paper demonstrates that a multi-resolution discriminator is the key factor behind the success of recent GAN-based vocoders, rather than specific architectural or training details, by experimentally validating its effectiveness across multiple generators.

Contribution

The study shows that a shared multi-resolution discriminating framework alone can achieve high-quality speech synthesis across different generators, simplifying the design of GAN vocoders.

Findings

Multi-resolution discriminator is the main factor in GAN vocoder success.

Different generator architectures perform similarly with the shared discriminator.

The approach outperforms or matches existing methods in perceptual metrics.

Abstract

Several of the latest GAN-based vocoders show remarkable achievements, outperforming autoregressive and flow-based competitors in both qualitative and quantitative measures while synthesizing orders of magnitude faster. In this work, we hypothesize that the common factor underlying their success is the multi-resolution discriminating framework, not the minute details in architecture, loss function, or training strategy. We experimentally test the hypothesis by evaluating six different generators paired with one shared multi-resolution discriminating framework. For all evaluative measures with respect to text-to-speech syntheses and for all perceptual metrics, their performances are not distinguishable from one another, which supports our hypothesis.