VoxCPM is OpenBMB's open text-to-speech model; version 2 speaks 30 languages and can invent a voice from a written description or copy one from a short clip.
In plain words
VoxCPM makes speech directly as continuous sound data instead of first splitting it into separate tokens, which its makers say sounds more natural and expressive. VoxCPM2 has 2 billion parameters, learned from over 2 million hours of speech in many languages and outputs 48 kHz audio. You can describe a new voice by age, tone or pace, or clone a voice from a short sample and steer its emotion. Weights and code use the Apache 2.0 licence, which allows commercial use.
A quick example
Imagine a teacher making audio lessons. They type a description such as a calm, older voice at a slow pace to design a narrator, then paste in each lesson's text to hear it read aloud in that voice.
Where you'll meet it
- Narration, audiobooks and lessons.
- Voices for voice agents that must answer quickly.
- Designing or cloning voices for games and videos.
Related
Sources
- DocsOpenBMB/VoxCPM, OpenBMB (GitHub) · read 23 Oct 2026
- Docsopenbmb/VoxCPM2, OpenBMB (Hugging Face) · read 23 Oct 2026
Checked by WhatIsAI against the sources above · 23 Oct 2026. A full five-minute explainer with a diagram may follow. Spotted an error? Report it.