WaveGlow: A Flow-based Generative Network for Speech Synthesis

TL;DR

WaveGlow introduces a non-autoregressive, flow-based vocoder that learns via maximum likelihood and uses invertible 1x1 convolutions plus affine coupling conditioned on mel-spectrograms. The model achieves high-quality speech with real-time or faster-than-real-time inference on GPUs, significantly simplifying training by using a single network and loss. Empirical results show MOS comparable to WaveNet while providing substantial speed advantages, highlighting practical deployment benefits. This approach offers a simpler, scalable alternative to prior multi-network flow methods for fast, high-fidelity speech synthesis.

Abstract

In this paper we propose WaveGlow: a flow-based network capable of generating high quality speech from mel-spectrograms. WaveGlow combines insights from Glow and WaveNet in order to provide fast, efficient and high-quality audio synthesis, without the need for auto-regression. WaveGlow is implemented using only a single network, trained using only a single cost function: maximizing the likelihood of the training data, which makes the training procedure simple and stable. Our PyTorch implementation produces audio samples at a rate of more than 500 kHz on an NVIDIA V100 GPU. Mean Opinion Scores show that it delivers audio quality as good as the best publicly available WaveNet implementation. All code will be made publicly available online.

Figures (1)

• Figure 1: WaveGlow network