APEX Framework Predicts Popularity of AI-Generated Music
Researchers have introduced APEX, the first large-scale multi-task learning framework for predicting the popularity of AI-generated music. Trained on over 211,000 songs (10,000 hours of audio) from platforms Suno and Udio, APEX jointly predicts engagement-based popularity signals—streams and likes scores—alongside five perceptual aesthetic quality dimensions. The framework uses frozen audio embeddings from MERT, a self-supervised music understanding model. The study highlights that aesthetic quality and popularity capture complementary aspects of music, proving valuable in an out-of-distribution setting. This research addresses the challenge of evaluating AI-generated music, which lacks traditional markers of artist reputation or label backing.
Key facts
- APEX is the first large-scale multi-task learning framework for AI-generated music popularity prediction.
- Trained on over 211k songs (10k hours of audio) from Suno and Udio.
- Jointly predicts streams, likes scores, and five aesthetic quality dimensions.
- Uses frozen audio embeddings from MERT, a self-supervised music understanding model.
- Aesthetic quality and popularity capture complementary aspects of music.
- The framework is effective in out-of-distribution settings.
- Addresses the new landscape of AI-generated music without traditional artist or label markers.
- Published on arXiv under ID 2605.03395.
Entities
Institutions
- arXiv