MQAScore measures text to music match in detailed musical aspects
MQAScore: Fine-Grained Text-Music Alignment Evaluation via Large Audio Language Model
SoundArtificial Intelligence
Summary
Matching a text description to music is usually judged by a single overall score, which hides important details about what matches or doesn't. The authors created MQAScore, which checks the match on specific musical ideas like genre, instruments, mood, vocals, and usage by asking yes/no questions to a powerful audio language model. They also developed a way to check how sensitive this method is to real differences in music. Their tests show that this detailed approach gives clearer and more reliable results than just one general score.
What this means in practice
- •For music streaming services: Improve recommendation systems by finely assessing how well songs match text queries across multiple musical attributes.$Commercial implications: Enables advanced text-based song search and matching features that can be marketed to streaming platform users.
- •For video game developers: Accurately select background music matching specific in-game themes or moods by verifying fine-grained musical details.
Authors
Wei-Jaw Lee, Chun-Ping Wang, Xuanjun Chen, Yi-Hsuan Yang
Abstract
Text-to-music alignment is commonly evaluated by the global similarity between text and audio embeddings, which collapses all described attributes into a single score and obscures which ones are missing. We instead introduce MQAScore, which verifies each atomic musical concept along five clip-level dimensions (genre, instrument, mood/theme, vocal, and usage) by the probability that a large audio language model (LALM) answers affirmatively to a binary question. To assess evaluator reliability, we construct verification labels from presence/absence annotations and LLM-guided concept substitutions, and propose the Sensitivity-Invariance Ratio, which measures how selectively an evaluator responds to meaning-changing substitutions. Experiments show that concept-level scoring improves both question-answering and similarity-based evaluators, and MQAScore can be a more reliable and interpretable fine-grained complement to global similarity for evaluating text-to-music alignment.