Diptych: Scoped, AI-Interpreted Comparison for Reference Listening in Music Production

Diptych. Scoped, AI-Interpreted Comparison for Reference Listening in Music Production. By Chongjun Zhong and four others, from Zhejiang University and the Singapore University of Technology and Design. Published on the thirtieth of September. No code repository is linked to it.

Reference listening is a common strategy in music production: you judge your own work by comparing it against a reference. Current comparison tools, though, often obscure a key human judgment, and it is the one that comes first. Before anything is compared, someone has to decide what should be compared.

Diptych is an AI-assisted system that hands that decision back to the user. Users define the comparison scope themselves, either across whole tracks or across segments they select independently. Within whatever scope they choose, they can inspect structured audio features alongside AI interpretations that are specific to that scope.

Two kinds of material sit side by side. One is evidence, in the form of structured audio features that can be inspected directly. The other is interpretation, produced by AI for the particular scope the user has set. Because the segments are chosen independently, the scope is never fixed by the tool: it is whatever the musician decided belongs side by side.

Diptych was evaluated in a within-participants study with twelve musicians, complemented by source-blinded ratings from four expert listeners. Participants used the system to surface additional differences, and nine of ten of those differences received at least partial support from the experts. Participants also reported good usability, and greater clarity about possible next steps.

The authors draw a design lesson rather than a ranking. AI support for creative comparison, they suggest, should prioritize user-defined scope, inspectable evidence and actionable guidance. It should also avoid authoritative judgments that go beyond what the evidence can support. The musician decides what to compare, and the system shows its work.