VoxCPM2: open 2B-parameter TTS model with voice cloning and design

OpenBMB released VoxCPM2, an open-source text-to-speech model with 2 billion parameters under Apache-2.0. It clones voices from a few seconds of audio, invents new voices from text descriptions and outputs 48 kHz speech in 30 languages, running locally on a GPU.

• 2B parameters, Apache-2.0 license, released April 2026
• Zero-shot voice cloning and text-prompted voice design
• 30 languages plus 9 Chinese dialects, 48 kHz output
• OpenAI-compatible /v1/audio/speech endpoint via vLLM-Omni