Repolode
Submit
k2-fsa

k2-fsa/OmniVoice

Open-source massively multilingual zero-shot text-to-speech model supporting 600+ languages, with voice cloning and voice design.

Open-source alternative to:ElevenLabs

View on GitHub
Stars
14,781
Forks
2,206
Open issues
67
Last commit
8 days ago

What is it?

OmniVoice is a text-to-speech model that generates speech in over 600 languages without per-language training. It is built on a diffusion language model-style architecture and supports zero-shot voice cloning from a short reference clip, as well as voice design, where a voice is described with attributes such as gender, age, pitch, accent, or Chinese dialect.

Beyond basic synthesis, it supports inline non-verbal tags such as [laughter] and [sigh], pinyin tone markers for correcting Chinese pronunciation, and CMU dictionary phonemes for English. Output is 24 kHz audio. It runs on NVIDIA, Apple Silicon (MPS), and Intel Arc (XPU) hardware, and can be used through a Python API, a Gradio web demo, or command-line tools, including a multi-GPU batch inference tool.

For faster serving on NVIDIA GPUs, an optional FlashInfer integration is available.

Key features

  • 01Zero-shot voice cloning from a short reference audio clip.
  • 02Voice design by specifying gender, age, pitch, style, accent, or Chinese dialect.
  • 03Non-verbal tags like [laughter] and [sigh] inserted into text.
  • 04Pinyin and CMU phoneme input for pronunciation correction.
  • 05Runs on NVIDIA, Apple Silicon, and Intel Arc GPUs.
  • 06Python API, Gradio web demo, and CLI tools including multi-GPU batch inference.

Use cases

  • Generating multilingual voiceovers or narration across many languages.
  • Cloning a specific speaker's voice for personalized speech output, with their consent.
  • Batch-producing large volumes of speech audio across multiple GPUs.
  • Prototyping TTS features in a local Gradio demo before integrating an API.

Quick start

  1. 1

    Install PyTorch for your hardware (NVIDIA GPU example)

    pip install torch==2.8.0+cu128 torchaudio==2.8.0+cu128 --extra-index-url https://download.pytorch.org/whl/cu128
  2. 2

    Install OmniVoice from PyPI

    pip install omnivoice
  3. 3

    Launch the local web demo

    omnivoice-demo --ip 0.0.0.0 --port 8001

Things to know

  • Voice design is trained only on Chinese and English data, so other languages may be unstable.
  • Running it well generally needs a GPU (CUDA, Apple Silicon MPS, or Intel Arc XPU), and the pre-trained model must be downloaded from Hugging Face.
  • FlashInfer acceleration is limited to NVIDIA GPUs.

Topics

#text-to-speech#voice-cloning#speech-synthesis#multilingual#diffusion#pytorch

FAQ

What is OmniVoice?
Open-source massively multilingual zero-shot text-to-speech model supporting 600+ languages, with voice cloning and voice design. OmniVoice is a text-to-speech model that generates speech in over 600 languages without per-language training. It is built on a diffusion language model-style architecture and supports zero-shot voice cloning from a short reference clip, as well as voice design, where a voice is described with attributes such as gender, age, pitch, accent, or Chinese dialect.
Is OmniVoice free to use? What license does it have?
OmniVoice is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is OmniVoice still maintained?
Actively maintained. The last commit to OmniVoice was on Oct 12, 2026.
What can I use OmniVoice for?
Generating multilingual voiceovers or narration across many languages. Cloning a specific speaker's voice for personalized speech output, with their consent. Batch-producing large volumes of speech audio across multiple GPUs. Prototyping TTS features in a local Gradio demo before integrating an API.
How do I install OmniVoice?
Install PyTorch for your hardware (NVIDIA GPU example): pip install torch==2.8.0+cu128 torchaudio==2.8.0+cu128 --extra-index-url https://download.pytorch.org/whl/cu128 Install OmniVoice from PyPI: pip install omnivoice Launch the local web demo: omnivoice-demo --ip 0.0.0.0 --port 8001
Where can I get OmniVoice?
The source code and installation instructions are on GitHub: https://github.com/k2-fsa/OmniVoice

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

Related projects