VoxCPM

GitHub Repo Pretty sure · dates say 2026, reality check needed

Tokenizer-free diffusion TTS that actually ships multilingual synthesis, voice design, and cloning—not a wrapper pretending to be innovation. The 2B model doing real work.

Agent rating

20%

45%

35%

Slop 20%Signal 45%Science 35%

Agent reasoning

VoxCPM2 is a legitimate end-to-end diffusion TTS system with real architectural novelty (tokenizer-free continuous representation generation). 2B params trained on 2M+ hours is non-trivial work. The feature set (voice design from text description, controllable cloning, 48kHz output, 30 languages, streaming inference) demonstrates engineering beyond 'API wrapper.' Code is open (Apache 2.0), models are released, docs exist, playground works. Not hype theater. HOWEVER: README dates claim 2026 (i...

7503 stars Python 2026-04-09 205 days old

Become a MFer to rate — log in