Neural network estimates plate reverb settings from sound impulse responses
Simulation-Based Inference for Plate Reverb System Identification
Machine Learning
Summary
Plate reverb is a type of audio effect created by vibrating a large metal plate. To recreate or adjust this effect digitally, you need to know specific physical settings of the plate. The authors built a system that trains a neural network to guess these settings just by listening to a short sound called an impulse response. Their method quickly refines its guesses for new sounds by simulating similar sounds and learning from them, without rerunning complex physical simulations each time.
What this means in practice
- •For audio software developers: Create tools that automatically extract plate reverb settings from recorded impulse responses for sound design and mixing.
- •For acoustic engineers: Estimate physical parameters of plate reverberation systems from measured impulse responses to aid in system calibration and analysis.
Authors
Dylan Sechet, Marc Evrard, Matthieu Kowalski
Abstract
We address Task A of the 1st DAFx Parameter Estimation Challenge, which aims to retrieve the physical parameters of a plate model from an impulse response. To do so, we use the Simulation-Based Inference (SBI) framework, in which we train a neural network to estimate a density over plate parameters given an impulse response, using a dataset generated by the simulator. Inference for a new impulse response then requires only a forward pass through the network, without involving the simulator. For each test observation, we fine-tune a specific network: additional simulation rounds are performed by sampling parameters from the current estimated distribution, simulating the corresponding impulse responses, and fine-tuning to produce the specialized network.