Machine Heart

May 19, 2026 · Artificial Intelligence

When Does a Song’s Climax Start? GaMMA Lets Multimodal Models Grasp Music Timelines

GaMMA is a multimodal large model that jointly learns global music semantics and fine‑grained temporal dynamics via a dual‑encoder fusion network and a three‑stage progressive training pipeline, and its accompanying MusicBench benchmark shows state‑of‑the‑art performance on both global and temporal music understanding tasks, surpassing Gemini‑3.0 Pro.