Wei Tan

This paper presents SongBloom, the first unified autoregressive diffusion model for long-form song generation, achieving state-of-the-art performance compared to both commercial and non-commercial methods.