Parallel WaveNet: Fast High-Fidelity Speech Synthesis

TL;DR

This paper introduces a parallel neural network for speech synthesis that achieves high fidelity and over 20 times faster-than-real-time generation, enabling practical deployment in real-time systems like Google Assistant.

Contribution

It presents Probability Density Distillation, a novel training method to convert WaveNet into a fast, parallel model without quality loss.

Findings

Achieves over 20x real-time speech synthesis speed

Maintains high speech quality comparable to WaveNet

Deployed in Google Assistant for multiple voices

Abstract

The recently-developed WaveNet architecture is the current state of the art in realistic speech synthesis, consistently rated as more natural sounding for many different languages than any previous system. However, because WaveNet relies on sequential generation of one audio sample at a time, it is poorly suited to today's massively parallel computers, and therefore hard to deploy in a real-time production setting. This paper introduces Probability Density Distillation, a new method for training a parallel feed-forward network from a trained WaveNet with no significant difference in quality. The resulting system is capable of generating high-fidelity speech samples at more than 20 times faster than real-time, and is deployed online by Google Assistant, including serving multiple English and Japanese voices.