GAN Vocoder: Multi-Resolution Discriminator Is All You Need
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.
中文速览
六款基于生成对抗网络（GAN）的声码器（vocoder）在近年来表现亮眼，但研究者们对各自成功的原因众说纷纭，有人归功于生成器结构，有人强调损失函数设计。这项研究提出了一个不同的假设：真正关键的共同因素是"多分辨率判别框架"（multi-resolution discriminating framework），而非那些细枝末节的架构差异。为了验证这一点，作者将六种不同的生成器统一搭配同一套来自HiFi-GAN的多分辨率判别器进行训练和测试，并通过主观平均意见分（MOS）和客观梅尔倒谱失真（MCD）两类指标综合评估。结果显示，在文本转语音合成场景下，六个模型的主客观指标均无统计显著差异，彼此置信区间高度重叠，有力地支持了"判别框架才是决定性因素"这一核心论断。这一发现对声码器研究具有重要的方向性意义——与其在生成器上做精细打磨，不如将更多精力投入到判别框架的设计与改进。
原文 arXiv:2103.05236;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2103.05236v2