Fast Wavenet Generation Algorithm

TL;DR

This paper introduces Fast Wavenet, an efficient algorithm that reduces the computational complexity of Wavenet generation from exponential to linear by caching previous calculations, enabling faster autoregressive synthesis.

Contribution

The paper proposes a caching-based method that significantly accelerates Wavenet generation, applicable to any autoregressive model with dilated convolutions.

Findings

Achieves linear time complexity in generation

Demonstrates significant speed improvements over naive implementation

Applicable to other autoregressive models with dilated convolutions

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.