Several studies have used deep learning methods to create digital twins of amps, speakers, and effects pedals. This paper presents a novel method for creating a digital twin of a physical loudspeaker with stereo output. Two neural network architectures are considered: a Recurrent Neural Network (RNN) and a WaveNet-style Convolutional Neural Network (CNN). The models were tested on two datasets containing speech and music, respectively. The method of recording and preprocessing the target audio data addresses the challenge of lacking a direct output line to digitize the effect of nonlinear circuits. Both model architectures successfully create a digital twin of the loudspeaker with no direct output line and stereo audio. The RNN model achieved the best result on the music dataset, while the WaveNet model achieved the best result on the speech dataset.

Authors: Louise, Bryn; Kerimovs, Teodors; Schlecht, Sebastian J.
Affiliations: Aalto University, Espoo, Finland; Aalto University, Espoo, Finland; Aalto University, Espoo, Finland(See document for exact affiliation information.)
AES Convention: 154 (May 2023) Paper Number: 10642
Publication Date: May 13, 2023 Import into BibTeX
Subject: Transducers
Permalink: http://www.aes.org/e-lib/browse.cfm?elib=22049

Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!

This paper costs $33 for non-members and is free for AES members and E-Library subscribers.

Learn more about the AES E-Library

E-Library Location:   /conv/154/10642.pdf

Start a discussion about this Transducers!