VoxCPM
OpenBMB/VoxCPMReviewed by AI
Checked by RepoPicks Radar on
What is VoxCPM?
VoxCPM2 turns written text into speech on your own machine. It can invent a voice from a written description, or copy a voice from a short recording. The README lists 30 languages and 48kHz audio output. It is for people comfortable with a terminal: you install it with "pip install voxcpm", and the README asks for Python 3.10 or newer, PyTorch 2.5.0 or newer and CUDA 12.0 or newer. The model table lists about 8 GB of VRAM for VoxCPM2. There is also a browser page you start yourself with "python app.py". Code and model weights are under Apache-2.0.
VoxCPM is free and open source (Apache-2.0) and can replace ElevenLabs, Murf, and Speechify.
How easy is it to install VoxCPM?
D · terminal, several tools or a specific graphics card · Windows
- Windows D
- macOS D
- Linux D
The best grade is D and it is the same on Windows, macOS and Linux: the 14 GitHub releases carry no ready-to-run file, so you install Python yourself, run "pip install voxcpm" and then type commands, and the README's model table lists about 8 GB of VRAM for VoxCPM2. macOS is not easier: the README only says that "python app.py --device auto" uses MPS on Apple Silicon, so the same install steps apply.
Coming soon: install in 1 click with the RepoPicks app.
VoxCPM on GitHub
- Stars
- 42,206on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +1,201stars in 7 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Python, Graphics card
- To build from source
- Python, Graphics card
- Disk space
- 950 MB–4.3 GB · estimate
- Accounts or keys
- None
- Cost
- Free · License is Apache-2.0 and the README says the weights and code are "free for commercial use". No price or paid plan is mentioned.
- License
- Apache-2.0 (Permissive)
- Can replace
- ElevenLabs, Murf, Speechify
- Last change on GitHub
- Nov 9, 2026
VoxCPM risk signals (4)
- Attentionvoice cloning can imitate real people
The README has a "Risks and Limitations" section that says: "Potential for Misuse: VoxCPM's voice cloning can generate highly realistic synthetic speech. It is strictly forbidden to use VoxCPM for impersonation, fraud, or disinformation. We strongly recommend clearly marking any AI-generated content." The same README shows how to copy a voice from a short clip, for example the command "voxcpm clone --text \"This is a voice cloning demo.\" --reference-audio path/to/voice.wav --output out.wav". Many countries have laws about using a person's voice without that person's permission.
AI review
- Infomodel weights downloaded on first use
The README's first example calls VoxCPM.from_pretrained("openbmb/VoxCPM2"), which pulls the 2B-parameter weights from Hugging Face (or from ModelScope with snapshot_download("OpenBMB/VoxCPM2")) the first time you run it, so the download happens outside the repository.
AI review
- Infopackage marked alpha by the authors
pyproject.toml lists the classifier "Development Status :: 3 - Alpha" for the published "voxcpm" package.
AI review
- Infoaccuracy varies a lot by language
In the README's Minimax-MLS-test WER table, VoxCPM2 scores 2.289 for English and 0.913 for Dutch but 38.584 for Cantonese, 24.132 for Czech, 21.577 for Romanian and 19.699 for Hindi, so quality is uneven across the 30 listed languages. The README's own notes add that "Voice Design and Controllable Voice Cloning results can vary between runs".
AI review
Evidence is shown in its original language.
What Radar could not check about VoxCPM
- Whether the install really works on Windows, macOS and Linux, since no release carries a ready-to-run file.
- Whether it runs usefully without an NVIDIA card: the README asks for CUDA 12.0 or newer, while "python app.py --device" also accepts cpu and mps.
- The real disk space used after the VoxCPM2 weights are downloaded (the checks estimate 950-4450 MB by heuristic).
- Whether the speed figures in the README (RTF ~0.30 on an RTX 4090, ~0.13 with Nano-vLLM) and the benchmark tables hold on your own machine.
- Whether about 8 GB of VRAM is enough in practice for long texts or several requests at once.