Hysen LabsVoxCPM
VoxCPM2: Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning
VoxCPM: text to speech that skips tokenization
A tokenizer free TTS system that generates continuous speech representations through an end to end diffusion autoregressive architecture, now in a 2 billion parameter release.
The approach
VoxCPM is a tokenizer free text to speech system. It directly generates continuous speech representations through an end to end diffusion autoregressive architecture, bypassing discrete tokenization for natural and expressive synthesis. Python is the primary language, the license is Apache-2.0, and the homepage is voxcpm.com.
VoxCPM2 by the numbers
VoxCPM2 is the latest major release: a 2 billion parameter model trained on over 2 million hours of multilingual speech data. It supports 30 languages, voice design, controllable voice cloning, and 48kHz studio quality audio output, and it is built on a MiniCPM-4 backbone.
Voice design and cloning
The highlights break into three tiers. Voice design creates a new voice from a natural language description alone, with no reference audio required. Controllable cloning works from a short reference clip with optional style guidance. Ultimate cloning preserves every vocal detail, using reference audio and a transcript.
Multilingual by default
The 30 language multilingual synthesis needs no language tag at all, which removes a whole class of setup friction. The same architecture that skips tokenization also skips the usual language selection step.
Community channels
Discussion and support run through Feishu and Discord, with a MiniCPM wiki linked as well. The project sits inside the OpenBMB organization, which the README makes clear through its links and attributions.
Editorial conclusion
The release notes do the talking: no language tags, voice design from a description, cloning from a short clip. The architecture section explains how, and the community section shows where to argue about it.
Community notes