Neural Synthesis of Footsteps Sound Effects with Generative Adversarial Networks

TL;DR

This paper explores neural network-based synthesis of footstep sounds using GANs, achieving realism comparable to real recordings and outperforming traditional methods.

Contribution

It introduces the first neural synthesis approach for footstep sounds with two GAN architectures and compares them to existing traditional synthesis techniques.

Findings

GAN architectures achieved realism scores close to real recordings

Neural synthesis outperformed six traditional sound synthesis methods

Encouraging results suggest viability of neural methods for sound effect generation

Abstract

Footsteps are among the most ubiquitous sound effects in multimedia applications. There is substantial research into understanding the acoustic features and developing synthesis models for footstep sound effects. In this paper, we present a first attempt at adopting neural synthesis for this task. We implemented two GAN-based architectures and compared the results with real recordings as well as six traditional sound synthesis methods. Our architectures reached realism scores as high as recorded samples, showing encouraging results for the task at hand.