OmniVoice: Free speech model clones voices in over 600 languages
OmniVoice is a free text-to-speech model that can recreate a voice from just a few seconds of audio and supports over 600 languages. Alternatively, you can design a voice from scratch by choosing gender, age and pitch. You can run it locally in your browser.
What it's about
OmniVoice is a free, open-source text-to-speech (TTS) model, an AI model that turns written text into spoken language. The project is on GitHub at k2-fsa/OmniVoice. The Japanese site GIGAZINE presented it and tried it out locally on Windows 11. According to the source, it supports over 600 languages and can recreate voices from short audio samples or generate new ones from traits such as gender, age and pitch.
What OmniVoice can do
The model handles what's known as "zero-shot speech cloning". That means it reproduces a voice matching the speech sample you give it, without needing to be retrained for it. A short recording is enough as a template.
The official demo page ("Massive Multilingual Zero-shot TTS") shows examples where you can directly compare the original template ("prompt"), the text being read and the generated speech. The dropdown menu offers examples in 102 languages, including Japanese. When you switch languages, you can hear how the voice characteristics of the template are preserved even when a different text is spoken.
The numbers need careful reading: over 600 languages are supported, but the demo page showcases 102.
Designing voices and fine control
If you don't have a reference voice, use "Voice Design" mode. There you pick, for example, gender ("Female"), age ("Young Adult") and pitch ("High Pitch") and get a new voice, with no reference audio at all.
There is also finer control ("Fine-Grained Control"). It lets you add sounds beyond words, such as laughter, by writing "[laughter]" into the text. The source only hints at other features without describing them in detail.
Running it locally
GIGAZINE shows the process on a Windows 11 PC with an NVIDIA graphics card, with Python and Git for Windows (Git Bash) installed:
- Create and activate a virtual environment with Python 3.11.
- Install PyTorch 2.8.0 and torchaudio 2.8.0 in the CUDA 12.8 build.
- Install the "omnivoice" package with pip.
- Start the web interface with
omnivoice-demo --ip 127.0.0.1 --port 8001.
Then open http://127.0.0.1:8001 in your browser. Enter your text under "Text to Synthesize". Under "Reference Audio", record 3 to 10 seconds with a microphone or upload a file. Then select the language (Japanese in the test), click "Generate" and listen to the result under "Output Audio".
What you need to try it
The source only describes the test system: Windows 11, NVIDIA GPU, Python 3.11, Git Bash. The article gives no exact VRAM figure, model size or memory requirements. Whether it also runs on a Mac, on CPU alone or in Google Colab isn't stated in the source. Expect to need several gigabytes of space for PyTorch and the model files.
Our take
A free solution that covers this many languages and clones voices from seconds of audio is interesting for video dubbing, learning materials and accessibility, especially for languages that commercial services barely cover. How good the quality really is in rare languages can't be judged from the article, as there are no benchmarks. License and model size aren't mentioned either. Misuse potential is a concern: voice clones can be used for deception and fraud. Only use voices you have permission to use.
Sources
- Google News (US) gigazine.net
Try it yourself: OmniVoice
Explained step by step, no prior knowledge needed – on your own computer or free in your browser.
What can I do with it?
OmniVoice reads text aloud in over 600 languages. You can either recreate a voice from a recording of 3 to 10 seconds or design a new voice by choosing gender, age and pitch. You can also insert laughter with [laughter].
Who is it worth it for?
Interesting for video makers, learning-material creators and tinkerers who want free, local multilingual voiceovers. Less suitable if you don't have an NVIDIA GPU, since only that setup is described.
What your computer needs
The source mentions no variants or quantizations, only the "omnivoice" package. Use the standard package with an NVIDIA graphics card.
- Tested on Windows 11 with an NVIDIA GPU and CUDA 12.8.
- Requires Python 3.11, PyTorch 2.8.0 and torchaudio 2.8.0.
- The source doesn't state VRAM requirements; a current graphics chip with several GB makes sense.
- Whether CPU, Mac or Colab work isn't stated in the source.
How do I find out what my computer has?
- Windows: Press Ctrl + Shift + Esc → Task Manager → Performance tab → click GPU on the left. At the bottom you'll see Dedicated GPU memory (that's your VRAM). You'll find system memory under Memory.
- Mac: Apple menu at the top left → About This Mac. You'll see your chip (e.g. M2) and memory (e.g. 16 GB). On Apple chips, graphics and processor share this memory.
- Linux: Type
nvidia-smiin the terminal (shows your graphics card and its memory) andfree -hfor system memory.
Guide
This model runs with its developers' own program (k2-fsa/OmniVoice on GitHub). The commands below come from the Model Card and the developers' instructions.
In the browser (no installation)
The developers have an online demo on Hugging Face. You can try the model right away there without installing anything. Sometimes you have to wait a moment for the demo to wake up.
Windows 10 / 11
Windows 10 / 11: not officially tested. The source doesn't explicitly mention Windows. It should work with an NVIDIA graphics card; the Linux steps are simply carried over here. Alternative with no installation: the online demo on Hugging Face (Space k2-fsa/OmniVoice) or Google Colab.
- Check the prerequisites. You need Python (3.10 or newer), Git and an NVIDIA graphics card with an up-to-date driver. The source doesn't name an exact Python version; 3.10 or newer is a safe choice. Check in PowerShell with the commands below.
python --version git --version nvidia-smi - Get the program. Download the program code from GitHub into a new folder.
git clone https://github.com/k2-fsa/OmniVoice.git cd OmniVoice - Create an environment. A virtual environment is its own self-contained Python folder. That keeps everything from getting mixed up. If activation is blocked, allow scripts once with the second command.
python -m venv .venv .venv\Scripts\Activate.ps1 - Install PyTorch. PyTorch is the computing library for AI. This command is the NVIDIA variant with CUDA 12.8. If you have a different CUDA version, check the PyTorch website.
pip install torch==2.8.0+cu128 torchaudio==2.8.0+cu128 --extra-index-url https://download.pytorch.org/whl/cu128 - Install OmniVoice. Now comes the actual program.
pip install omnivoice - Generate your first speech. On first launch, the program downloads the model automatically (about 3.3 GB). That takes a moment. Here a voice is generated from a description, so you don't need your own audio file.
omnivoice-infer --model k2-fsa/OmniVoice --text "This is a test for text to speech." --instruct "male, British accent" --output hello.wav - Result. The folder contains the file hello.wav with the spoken sentence "This is a test for text to speech."
macOS
macOS: supported. The source explicitly mentions Apple Silicon (M1 and newer), using device_map="mps" in the Python code. Whether the command line picks the Mac's GPU by itself isn't stated in the source. If it doesn't run, use the online demo on Hugging Face.
- Check the prerequisites. You need a Mac with an Apple chip (M1 or newer), Python 3 (3.10 or newer is a safe choice) and Git. If Git is missing, macOS offers the Xcode tools the first time you call it. Install them when prompted.
python3 --version git --version - Get the program. Download the program code from GitHub into a new folder.
git clone https://github.com/k2-fsa/OmniVoice.git cd OmniVoice - Create an environment. A virtual environment is its own self-contained Python folder. That keeps everything from getting mixed up.
python3 -m venv .venv source .venv/bin/activate - Install PyTorch. PyTorch is the computing library for AI. This is the variant for Apple Silicon.
pip install torch==2.8.0 torchaudio==2.8.0 - Install OmniVoice. Now comes the actual program.
pip install omnivoice - Generate your first speech. On first launch, the program downloads the model automatically (about 3.3 GB). Here a voice is generated from a description, so you don't need your own audio file.
omnivoice-infer --model k2-fsa/OmniVoice --text "This is a test for text to speech." --instruct "male, British accent" --output hello.wav - Result. The folder contains the file hello.wav with the spoken sentence "This is a test for text to speech."
Linux
Linux: supported. The source describes installation with NVIDIA cards (CUDA). Intel Arc cards run via XPU, tested on an Arc A310 with 4 GB. The model is released for non-commercial use only (CC-BY-NC).
- Check the prerequisites. You need Python (3.10 or newer is a safe choice), Git and an NVIDIA graphics card with the driver installed. Check with the commands below.
python3 --version git --version nvidia-smi - Get the program. Download the program code from GitHub into a new folder.
git clone https://github.com/k2-fsa/OmniVoice.git cd OmniVoice - Create an environment. A virtual environment is its own self-contained Python folder. That keeps everything from getting mixed up.
python3 -m venv .venv source .venv/bin/activate - Install PyTorch. PyTorch is the computing library for AI. This command is the NVIDIA variant with CUDA 12.8. If you have a different CUDA version, check the PyTorch website.
pip install torch==2.8.0+cu128 torchaudio==2.8.0+cu128 --extra-index-url https://download.pytorch.org/whl/cu128 - Install OmniVoice. Now comes the actual program.
pip install omnivoice - Generate your first speech. On first launch, the program downloads the model automatically (about 3.3 GB). Here a voice is generated from a description, so you don't need your own audio file.
omnivoice-infer --model k2-fsa/OmniVoice --text "This is a test for text to speech." --instruct "male, British accent" --output hello.wav - Result. The folder contains the file hello.wav with the spoken sentence "This is a test for text to speech."
Try it
Welcome to our learning video! Today we'll explain how photosynthesis works.That was really funny [laughter] – tell me more!Good morning, here are today's news headlines in three sentences.If something goes wrong
You need exactly the right PyTorch version (2.8.0 with cu128) installed, otherwise GPU use may fail.
On Windows, Git Bash was used; the commands for activating the environment (source .venv/Scripts/activate) are written for it.
The web interface is only reachable at http://127.0.0.1:8001 after you start omnivoice-demo.
For voice cloning, choose the language in the menu that matches the text.
“out of memory,” a crash, or your computer freezes
The model is too big for your VRAM (video memory). Pick a smaller variant from the table above and close other programs, especially games and browsers with lots of tabs.
It runs, but extremely slowly
Your CPU is probably doing the work instead of your graphics card. In Ollama, the PROCESSOR column of ollama ps shows how much runs on the GPU. If it says "100% CPU", update your graphics driver or pick a smaller variant that fits entirely into VRAM (video memory).
“ollama” or “python” is not found
Close the terminal window and open it again so it recognizes the newly installed program. For Python on Windows: reinstall it and check Add python.exe to PATH.
Windows: "Running scripts is disabled on this system"
Run the script with exactly the command above: powershell -ExecutionPolicy Bypass -File try_windows.ps1. This allows execution for this one time only.
Error message when loading the model (version, "trust_remote_code", unknown architecture)
Brand-new models often need the latest software version. With Ollama: update Ollama. With Python: pip install -U transformers in the activated environment. The Model Card usually lists the exact version.
Tokenstrom generates the scripts and notebook automatically. Please take a quick look before running them – they're plain, readable text. Mind the model's license: custom license (see Model Card).
Ask the AI assistant
Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.