Parallel WaveNet: Fast High-Fidelity Speech Synthesis
Aaron Oord, Yazhe Li, Igor Babuschkin, Karen Simonyan, Oriol Vinyals, Koray Kavukcuoglu, George Driessche, Edward Lockhart
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, a 1000x speed up relative to the original WaveNet, and capable of serving multiple English and Japanese voices in a production setting.
BibTeX
@InProceedings{pmlr-v80-oord18a,
title = {Parallel {W}ave{N}et: Fast High-Fidelity Speech Synthesis},
author = {van den Oord, Aaron and Li, Yazhe and Babuschkin, Igor and Simonyan, Karen and Vinyals, Oriol and Kavukcuoglu, Koray and van den Driessche, George and Lockhart, Edward and Cobo, Luis and Stimberg, Florian and Casagrande, Norman and Grewe, Dominik and Noury, Seb and Dieleman, Sander and Elsen, Erich and Kalchbrenner, Nal and Zen, Heiga and Graves, Alex and King, Helen and Walters, Tom and Belov, Dan and Hassabis, Demis},
booktitle = {Proceedings of the 35th International Conference on Machine Learning},
pages = {3918--3926},
year = {2018},
editor = {Dy, Jennifer and Krause, Andreas},
volume = {80},
series = {Proceedings of Machine Learning Research},
month = {10--15 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v80/oord18a/oord18a.pdf},
url = {https://proceedings.mlr.press/v80/oord18a.html},
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, a 1000x speed up relative to the original WaveNet, and capable of serving multiple English and Japanese voices in a production setting.}
}