DiffRhythm

DiffRhythm is the first latent diffusion-based song generation model capable of synthesizing complete songs with both vocals and accompaniment for durations up to 4 minutes and 45 seconds in just ten seconds. Unlike existing models that generate either vocal or accompaniment tracks separately or rely on complex multi-stage architectures, DiffRhythm offers an end-to-end solution with high musicality and intelligibility while maintaining fast inference speeds.