The paper “Training Neural Models of Nonlinear Multi-Port Elements Within Wave Digital Structures Through Discrete-Time Simulation” was awarded Second Best Paper at the International Conference on Digital Audio Effects (DAFx), held in Ancona from September 2 to 5, 2025.
The study, conducted by Oliviero Massi, Alessandro Ilic Mezza, Riccardo Giampiccolo, and Alberto Bernardini from the Department of Electronics, Information and Bioengineering – Politecnico di Milano, explores new ways to use neural networks to faithfully reproduce analogue audio instruments and equipment. In practice, the researchers developed a system where the “nonlinear” parts of a circuit—those that produce complex and characteristic sounds—are simulated by a small neural network in the Wave Digital domain. This network can be trained using direct measurements from the nonlinear component or circuit input-output data, making more efficient and accurate the emulation of the electronic circuit.
The main applications are in digital audio processing, particularly musical instrument effects, such as those used for guitar, allowing faithful emulation of analogue equipment in a digital format.