Real-Time Modeling of Audio Distortion Circuits with Deep Learning

Eero-Pekka Damskägg, Lauri Juvela, and Vesa Välimäki

Demo page for the paper submitted to SMC 2019.

This page includes sound samples of the proposed WaveNet-based black-box model. Three guitar distortion pedals were modeled: the Ibanez Tube Screamer, the Boss DS-1 and the Electro-Harmonix Big Muff Pi. Sound samples are included of the three selected WaveNet models. The model hyperparameters are detailed in the paper. WaveNet1 is the computationally cheapest and the least accurate model. WaveNet3 is the computationally most expensive and the most accurate model. WaveNet2 is an intermediate model.

A real-time implementation of the proposed deep neural network was developed in C++, using the JUCE framework. All three models were shown capable of running in real time on an Apple iMac with a 2.8 GHz Intel Core i5 processor. The source code with the pre-trained models can be found here.