MIRA improves user intent matching in text to music generation
MIRA: A Musical Intent Refinement Agent for Aligning Text-to-Music Generation with User Intent
SoundArtificial Intelligence
Summary
Text-to-music tools can create believable songs from written prompts, but often miss exactly what the user wants. The authors created a way to check music against detailed lists of user needs, like instruments or mood changes, to see how well the music fits. They also made MIRA, an assistant that refines the prompts and tries multiple generations to better match the user's intent. Tests show MIRA helps open-source music generators compete with commercial ones in meeting user requests.
What this means in practice
- •For music app developers: Improve text-to-music features by refining user prompts to produce songs that better meet specific musical requests and preferences.
- •For advertising agencies: Generate customized background music for ads that closely aligns with specified moods, instruments, and rhythms by iteratively refining text descriptions.$Commercial implications: Enables sales of targeted music generation services that deliver closer matches to client briefings, improving creative outcomes.
Authors
Zekai Liu, Zhilin Wang, Xuzheng He, Yu Cheng, Yang Yang
Abstract
Text-to-music systems produce increasingly convincing audio, yet evaluation reveals little about whether the result matches user intent. A global text-audio relevance score can overlook the implicit intent in underspecified prompts and mask failures in specific requirements, such as instrumentation, structure, rhythm, or mood progression. To bridge this gap, we formulate text-to-music intent alignment as satisfying a per-request rubric of independently verifiable items covering both a request's explicit requirements and its implied musical intent. Scoring items individually makes evaluation diagnostic by intent source and musical dimension, rather than a single opaque score. We instantiate this as MuRA-Bench, a benchmark of real-world platform requests curated by music experts. We further propose MIRA (Musical Intent Refinement Agent), a test-time agent that first grounds a request's intent into rubrics, then searches over prompt revisions for a black-box generator under a bounded budget, iteratively generating music, verifying it against the rubrics, and using this feedback to guide a trajectory-aware tree search. Experiments across open-source and commercial backends show that MIRA improves intent alignment, enabling an open-source generator to achieve performance comparable to representative commercial systems (e.g. Suno and Mureka). Project page: https://mirareview.github.io/.