OpenBMB open-source speech synthesis model VoxCPM2

OpenBMB's open-source 2B-parameter speech synthesis model, VoxCPM2, employs a segmenter-free diffusion autoregressive architecture, supports 30 languages and Chinese dialects, and outputs 48kHz studio-quality audio. The model pioneers the Voice Design feature, allowing users to create voices from scratch using text descriptions; it also supports controllable voice cloning and an ultimate cloning mode. The model has been trained on 2.36 million hours of data, achieving a real-time rate as low as 0.13, and is commercially viable under the Apache-2.0 license.