Voice, dictation & realtime
OpenBMB/VoxCPM
Tokenizer-free multilingual TTS you can run locally
- 36k
- Apache-2.0
- Python
What it is and why it matters
VoxCPM is a text-to-speech model aimed at multilingual generation, voice design and cloning. The podcast discusses a 2 billion parameter model needing about 8 GB of RAM, small enough to run on a regular laptop. It is framed as a self-hosted, free alternative to paid services like ElevenLabs, in exchange for managing the infrastructure yourself.
Key points
- Small model (2B parameters, about 8 GB of RAM) suitable for local use
- Decent voice cloning, though ElevenLabs still wins on naturalness and polish
- Free and self-hosted: you give up managed infrastructure and support
Where we saw it
Same category