This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
WildFX: A DAW-Powered Pipeline for In-the-Wild Audio FX Graph Modeling
Created by
Haebom
Author
Qihui Yang, Taylor Berg-Kirkpatrick, Julian McAuley, Zachary Novack
Outline
This paper addresses the issue that despite the rapid progress in end-to-end AI music generation, modeling professional digital signal processing (DSP) workflows with AI remains challenging. In particular, while there is growing interest in neural network black-box modeling of audio effect graphs such as reverb, compression, and equalization, AI-based approaches struggle to replicate the subtle signal flows and parameter interactions used in professional workflows. Existing differentiable plugin approaches often deviate from real-world tools and underperform simplified neural network controllers under the same computational constraints. In this paper, we present WildFX, a Docker-contained pipeline for generating multitrack audio mixing datasets with rich effect graphs based on a professional digital audio workstation (DAW) backend. WildFX seamlessly integrates cross-platform commercial or other plugins in VST/VST3/LV2/CLAP formats to support structural complexities such as sidechains and crossovers, and achieves efficient parallel processing. A minimal metadata interface simplifies project/plugin configuration. We validate the pipeline through experiments with mixing graphs, blind estimation of plugin/gain parameters, and the ability to bridge AI research with real-world DSP requirements. The code is available at https://github.com/IsaacYQH/WildFX .
