VoxCPM-0.5B is a speech model released by OpenBMB.
Technical highlights
：Unlike mainstream approaches that convert speech into discrete tokens, VoxCPM uses an end-to-end diffuse autoregressive architecture to generate continuous speech representations directly from text. Based on the MiniCPM-4 backbone, it implements implicit semantic-acoustic decoupling through hierarchical language modeling and FSQ constraints, greatly enhancing expressiveness and generation stability.