▲ 466 today40.4k starsPythonHN ×1
Tokenizer-Free TTS Model Speaks 30 Languages Without a Tokenizer
What is VoxCPM?
VoxCPM2 generates speech directly from text with a diffusion autoregressive model, skipping the usual tokenization step. It can design new voices from a plain-language description, clone a voice from a short clip, and output studio-grade 48 kHz audio in 30 languages.
Who it is for
Audio engineers, game developers, and creators who need realistic, multilingual speech can prototype custom voices or clone existing ones without gathering massive datasets.