Fast Wavenet Generation Algorithm

Abstract

This paper presents an efficient implementation of the Wavenet generation process called Fast Wavenet. Compared to a naive implementation that has complexity O(2^L) (L denotes the number of layers in the network), our proposed approach removes redundant convolution operations by caching previous calculations, thereby reducing the complexity to O(L) time. Timing experiments show significant advantages of our fast implementation over a naive one. While this method is presented for Wavenet, the same scheme can be applied anytime one wants to perform autoregressive generation or online prediction using a model with dilated convolution layers. The code for our method is publicly available.

Technical Report

Cited by in corpus (11)

• MidiNet: A Convolutional Generative Adversarial Network for Symbolic-domain Music Generation
• Solving Statistical Mechanics Using Variational Autoregressive Networks
• WaveFlow: A Compact Flow-based Model for Raw Audio
• ClearBuds: Wireless Binaural Earbuds for Learning-Based Speech Enhancement
• The Effect of Explicit Structure Encoding of Deep Neural Networks for Symbolic Music Generation
• Semantic Hearing: Programming Acoustic Scenes with Binaural Hearables
• Hierarchical Autoregressive Image Models with Auxiliary Decoders
• FastWave: Accelerating Autoregressive Convolutional Neural Networks on FPGA
• Neural source-filter waveform models for statistical parametric speech synthesis
• Wasserstein GAN and Waveform Loss-based Acoustic Model Training for Multi-speaker Text-to-Speech Synthesis Systems Using a WaveNet Vocoder
• A Fully Time-domain Neural Model for Subband-based Speech Synthesizer