Repolode
Submit
OpenBMB

OpenBMB/VoxCPM

VoxCPM2 is an open-source, tokenizer-free text-to-speech model supporting 30 languages, voice design from text descriptions, and voice cloning from short reference clips.

Open-source alternative to:ElevenLabs

View on GitHubWebsite
Stars
39,635
Forks
4,500
Open issues
135
Last commit
13 days ago

What is it?

VoxCPM2 is a text-to-speech system from OpenBMB that generates continuous speech representations with a diffusion autoregressive architecture instead of discrete tokens. It is a 2B-parameter model built on a MiniCPM-4 backbone and outputs 48kHz audio.

The model covers several workflows: plain multilingual synthesis across 30 languages and several Chinese dialects, voice design from a natural-language description with no reference audio, and cloning from a short reference clip with optional style control. A higher-fidelity mode takes both reference audio and its transcript so the output continues from the reference.

It is usable through a Python API, a CLI, a local web demo, and streaming generation. Serving options include Nano-vLLM, vLLM-Omni with an OpenAI-compatible endpoint, and llama.cpp-omni for on-device use. Fine-tuning with full SFT or LoRA is supported, and the weights and code are released under Apache-2.0.

Key features

  • 01Speech synthesis in 30 languages, including several Chinese dialects, without a language tag.
  • 02Voice design from a text description such as gender, age, and tone.
  • 03Voice cloning from a short reference clip, with optional style and emotion control.
  • 0448kHz audio output from 16kHz reference input.
  • 05Streaming generation through a Python API, CLI, and web demo.
  • 06Full fine-tuning and LoRA fine-tuning support.

Use cases

  • Generating multilingual voiceovers or narration from scripts.
  • Creating custom character voices for games, video, or virtual assistants.
  • Dubbing or localizing video content with a cloned speaker voice.
  • Running offline speech synthesis on-device via llama.cpp-omni.

Quick start

  1. 1

    Install the package with pip

    pip install voxcpm
  2. 2

    Load the model and generate speech from Python

  3. 3

    Or launch the local web demo

    python app.py --port 8808

Things to know

  • Requires Python 3.10 to below 3.13, PyTorch 2.5 or newer, and CUDA 12.0 or newer for the documented setup.
  • Full-quality inference is GPU-oriented; the README reports real-time factors measured on an NVIDIA RTX 4090.
  • The model is 2B parameters and needs a sizable download from Hugging Face or ModelScope.

Topics

#text-to-speech#voice-cloning#speech-synthesis#multilingual#pytorch#diffusion#audio#deeplearning#minicpm#python

FAQ

What is VoxCPM?
VoxCPM2 is an open-source, tokenizer-free text-to-speech model supporting 30 languages, voice design from text descriptions, and voice cloning from short reference clips. VoxCPM2 is a text-to-speech system from OpenBMB that generates continuous speech representations with a diffusion autoregressive architecture instead of discrete tokens. It is a 2B-parameter model built on a MiniCPM-4 backbone and outputs 48kHz audio.
Is VoxCPM free to use? What license does it have?
VoxCPM is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is VoxCPM still maintained?
Actively maintained. The last commit to VoxCPM was on Oct 8, 2026.
What can I use VoxCPM for?
Generating multilingual voiceovers or narration from scripts. Creating custom character voices for games, video, or virtual assistants. Dubbing or localizing video content with a cloned speaker voice. Running offline speech synthesis on-device via llama.cpp-omni.
How do I install VoxCPM?
Install the package with pip: pip install voxcpm Load the model and generate speech from Python Or launch the local web demo: python app.py --port 8808
Where can I get VoxCPM?
The source code and installation instructions are on GitHub: https://github.com/OpenBMB/VoxCPM

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects