WaveFlow: A Compact Flow-based Model for Raw Audio
Wei Ping, Kainan Peng, Kexin Zhao, Zhao Song
Abstract
In this work, we propose WaveFlow, a small-footprint generative flow for raw audio, which is directly trained with maximum likelihood. It handles the long-range structure of 1-D waveform with a dilated 2-D convolutional architecture, while modeling the local variations using expressive autoregressive functions. WaveFlow provides a unified view of likelihood-based models for 1-D data, including WaveNet and WaveGlow as special cases. It generates high-fidelity speech as WaveNet, while synthesizing several orders of magnitude faster as it only requires a few sequential steps to generate very long waveforms with hundreds of thousands of time-steps. Furthermore, it can significantly reduce the likelihood gap that has existed between autoregressive models and flow-based models for efficient synthesis. Finally, our small-footprint WaveFlow has only 5.91M parameters, which is 15{\texttimes} smaller than WaveGlow. It can generate 22.05 kHz high-fidelity audio 42.6{\texttimes} faster than real-time (at a rate of 939.3 kHz) on a V100 GPU without engineered inference kernels.
BibTeX
@InProceedings{pmlr-v119-ping20a,
title = {{W}ave{F}low: A Compact Flow-based Model for Raw Audio},
author = {Ping, Wei and Peng, Kainan and Zhao, Kexin and Song, Zhao},
booktitle = {Proceedings of the 37th International Conference on Machine Learning},
pages = {7706--7716},
year = {2020},
editor = {III, Hal Daumé and Singh, Aarti},
volume = {119},
series = {Proceedings of Machine Learning Research},
month = {13--18 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v119/ping20a/ping20a.pdf},
url = {https://proceedings.mlr.press/v119/ping20a.html},
abstract = {In this work, we propose WaveFlow, a small-footprint generative flow for raw audio, which is directly trained with maximum likelihood. It handles the long-range structure of 1-D waveform with a dilated 2-D convolutional architecture, while modeling the local variations using expressive autoregressive functions. WaveFlow provides a unified view of likelihood-based models for 1-D data, including WaveNet and WaveGlow as special cases. It generates high-fidelity speech as WaveNet, while synthesizing several orders of magnitude faster as it only requires a few sequential steps to generate very long waveforms with hundreds of thousands of time-steps. Furthermore, it can significantly reduce the likelihood gap that has existed between autoregressive models and flow-based models for efficient synthesis. Finally, our small-footprint WaveFlow has only 5.91M parameters, which is 15{\texttimes} smaller than WaveGlow. It can generate 22.05 kHz high-fidelity audio 42.6{\texttimes} faster than real-time (at a rate of 939.3 kHz) on a V100 GPU without engineered inference kernels.}
}