“Suno's AI models are designed to generate music by treating it as a raw sound wave sampled at 48,000 times per second and represented as a continuous float32 number.”