What VoxCPM is
VoxCPM is a free, open-source text-to-speech system. Its VoxCPM2 model generates expressive speech in 30 languages and lists nine Chinese dialects. Users can describe a voice in natural language without supplying reference audio, or use the voice-cloning capability. The model produces 48 kHz audio and accepts reference audio at 16 kHz; WAV is a listed export format. A streaming generation API is available. The project documents Python and command-line use, a local web demo, production serving, and on-device inference. Its vLLM-Omni option provides an OpenAI-compatible speech endpoint, and Nano-vLLM supports concurrent requests. Full and LoRA fine-tuning are supported, with adaptation possible from 5–10 minutes of audio. The code and model weights use the Apache-2.0 license and are free for commercial use. The documented Python installation requires Python 3.10 or later but below 3.13, PyTorch 2.5.0 or later, and CUDA 12.0 or later. Voice design and controllable cloning can vary between runs, and unsupported languages may need testing or fine-tuning.
Who it suits
VoxCPM suits developers and teams building speech generation into applications or running it locally. Its commercial-use license and fine-tuning options may also suit projects that need custom voices, provided they can meet the documented installation requirements.
Good to know, and what to watch
- Supports 30 languages and nine listed Chinese dialects
- Voice design works from a text description
- Streaming generation API is available
- Full and LoRA fine-tuning are supported
- Apache-2.0 licensed for commercial use
- Voice design and cloning results can vary between runs
- Python setup requires specified Python, PyTorch, and CUDA versions
- Unsupported languages may need testing or fine-tuning
Plans
Every plan and priceFree & Open Source
Open source
Getting started
- Visit the VoxCPM website or repository
- Install Python 3.10 or later but below 3.13
- Install PyTorch 2.5.0 or later and CUDA 12.0 or later
- Run the documented Python or CLI setup
- Use the local web demo or production serving options
- Connect through vLLM-Omni or Nano-vLLM as needed
Limits to check
The documented Python installation depends on Python 3.10–3.12, PyTorch 2.5.0 or later, and CUDA 12.0 or later. Results for Voice Design and controllable cloning can vary, and unsupported languages may need testing or fine-tuning.
The verdict
VoxCPM combines multilingual speech generation, voice design, cloning, and several deployment paths in a free, commercially usable project. Results can vary, and its documented Python setup has specific version requirements.
Questions people ask
Is VoxCPM free to use?
Yes. The Free & Open Source plan costs nothing, with no paid tiers or usage limits stated.
Can VoxCPM be used commercially?
Yes. The Apache-2.0 license covers the model weights and code, and commercial use is permitted.
What languages does VoxCPM2 support?
It supports 30 languages and lists nine Chinese dialects.
Does VoxCPM offer an API?
Yes. It provides a streaming generation API, and vLLM-Omni exposes an OpenAI-compatible /v1/audio/speech endpoint.
What audio formats and rates are available?
Generated audio is 48kHz, reference audio is accepted at 16kHz, and WAV is a listed export format.
What deployment options are documented?
The project documents Python, CLI, a local web demo, production serving, on-device inference, Apple Silicon MPS, and C++ CPU inference.
What the maker publishes
- API access
- Yes source
- API integration
- Its vLLM-Omni serving option exposes an OpenAI-compatible /v1/audio/speech endpoint. source
- Audio output
- The model outputs 48kHz audio and accepts 16kHz reference audio. source
- Commercial use
- Yes source
- Commercial use
- The project states its code and weights use the Apache-2.0 license and are free for commercial use. source
- Deployment
- The repository documents Python and CLI use, a local web demo, production serving, and on-device inference. source
- Export formats
- WAV source
- Fine-tuning
- It supports full and LoRA fine-tuning, with adaptation possible from 5–10 minutes of audio. source
- Headquarters
- Beijing, China source
- Integrations
- Documented serving options include vLLM-Omni with an OpenAI-compatible speech endpoint and Nano-vLLM for concurrent requests. source
- Languages
- VoxCPM2 supports 30 languages and lists nine Chinese dialects. source
- License and cost
- The model weights and code use the Apache-2.0 license, and the repository says they are free for commercial use. source
- Limitations
- Voice Design and controllable cloning results can vary between runs, and the project says unsupported languages may require testing or fine-tuning. source
- Maker
- The site identifies Modelbest Technology Co., Ltd. and offers custom solutions through [email protected]. source
- Platforms
- web,macos,linux,api,self_hosted source
- Platform support
- The README documents Apple Silicon MPS support and C++ inference on CPU, Metal, CUDA, or Vulkan. source
- Product
- VoxCPM2 is a tokenizer-free text-to-speech system that generates expressive multilingual speech. source
- Requirements
- The documented Python installation requires Python 3.10 or later but below 3.13, PyTorch 2.5.0 or later, and CUDA 12.0 or later. source
- Responsible use
- The project forbids using VoxCPM for impersonation, fraud, or disinformation and recommends labeling AI-generated content. source
- Streaming
- It provides a streaming generation API, and the repository reports real-time factor as low as about 0.3 on an NVIDIA RTX 4090. source
- Support
- The site offers custom business solutions and lists [email protected]; the repository links Discord and Feishu community channels. source
- Support
- The project links to Feishu, Discord, and its documentation wiki for community discussion and support. source
- Voice cloning
- Yes source
- Voice cloning
- It supports controllable voice cloning from a short reference clip and continuation cloning with reference audio and its transcript. source
Checked against the maker's pages on 8 Oct 2026.