Time-Varying Audio Effect Modeling by End-to-End Adversarial Training

TL;DR

This paper presents a novel GAN-based approach for modeling time-varying audio effects solely from input-output recordings, eliminating the need for control signal extraction and enabling black-box modeling of dynamic audio effects.

Contribution

It introduces a two-stage training strategy with adversarial learning and state prediction for effective time-varying effect modeling without control signals.

Findings

Successfully models a vintage hardware phaser's time-varying behavior

Develops a new chirp-train based metric for modulation accuracy

Demonstrates improved black-box modeling of dynamic audio effects

Abstract

Deep learning has become a standard approach for the modeling of audio effects, yet strictly black-box modeling remains problematic for time-varying systems. Unlike time-invariant effects, training models on devices with internal modulation typically requires the recording or extraction of control signals to ensure the time-alignment required by standard loss functions. This paper introduces a Generative Adversarial Network (GAN) framework to model such effects using only input-output audio recordings, removing the need for modulation signal extraction. We propose a convolutional-recurrent architecture trained via a two-stage strategy: an initial adversarial phase allows the model to learn the distribution of the modulation behavior without strict phase constraints, followed by a supervised fine-tuning phase where a State Prediction Network (SPN) estimates the initial internal states required to synchronize the model with the target. Additionally, a new objective metric based on chirp-train signals is developed to quantify modulation accuracy. Experiments modeling a vintage hardware phaser demonstrate the method's ability to capture time-varying dynamics in a fully black-box context.