What is it about?

Interactive music systems often need to recognize a performer's audio gesture, such as a short musical phrase or characteristic sound, and respond with sound, visuals, or another digital action. AGR is an open-source toolkit built in Max that lets musicians record examples during rehearsal, train a small recognition model, test the results, and revise what happens next without leaving the performance environment. The paper explains this rehearsal-to-performance workflow, evaluates it with audio gestures recorded from clarinet, flute, trumpet, and trombone, and discusses uses ranging from solo pieces to a production involving 12 ensemble models.

Featured Image

Why is it important?

AGR shifts attention from building a universal audio-gesture recognizer to making recognition editable by artists. When the system confuses two gestures, performers and composers can add examples, simplify categories, adjust listening boundaries, or change the electronic response as part of rehearsal. This helps machine listening function as practical compositional material rather than a fixed black box. AGR does not introduce a new machine-learning algorithm; its contribution is connecting capture, training, testing, and performance routing in one Max workflow.

Perspectives

I developed AGR through rehearsals and performances where the technical system had to remain open to musical change. The most important lesson was that recognition errors are not only engineering problems: they can prompt performers and composers to redefine a gesture, revise the electronic response, or make instability part of the piece. I hope the paper helps artists treat machine listening as something they can negotiate and reshape in rehearsal.

Hongshuo Fan
Texas A&M University College Station

Read the Original

This page is a summary of: AGR: A Rehearsal-to-Performance Workflow for Programming Audio Gestures in Max, August 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3830435.3830948.
You can read the full text:

Read

Contributors

The following have contributed to this page