WildFX: A DAW-Powered Pipeline for In-the-Wild Audio FX Graph Modeling

A preprint presenting a tool that renders multitrack effect-chain datasets using a real DAW.

Author
Qihui Yang, Taylor Berg-Kirkpatrick, Julian McAuley and Zachary Novack, UC San Diego
Published
17 July 2025
Kind
arXiv preprint presenting a dataset-generation tool with one experiment
Original
arxiv.org/html/2507.10534
Subject
Audio models and analysis
Themes
Art through code

What it is

The authors are researchers at UC San Diego. The paper presents WildFX, which generates mixing projects with commercial plugins, sidechains and splitters. The aim is to connect neural audio effect research to professional tools.

What it says

What evidence it gives

The paper gives a pipeline design, open code and one experiment. The experiment trains an existing graph-estimation method on two generated datasets. Results appear in one table of four settings, with no comparison to other pipelines. No listening or user test appears.

Interests and limits

The authors are academics presenting their own tool. Only five plugins are used, and one estimation method is tested. The authors say the results trail the earlier work. They still call them competitive and state of the art, with no comparison table. Headless REAPER offers no error reporting, so steps were checked manually. The paper claims to be the first such pipeline, to the authors' knowledge.

Connections

These links are this wiki's own reading, not claims the sources make.