kokoro-onnx

To use version 1.0 of the models, you must manually download the following files and place them in your working directory:

• kokoro-v1.0.onnx
• voices-v1.0.bin

Install the kokoro-onnx package using pip to get started with ONNX-based Text-to-Speech.

pip install -U kokoro-onnx

To build and publish a new version of the package, clear the existing dist directory, build the package using uv build, and then publish using uv publish with a PyPI token.

rm -rf dist
uv build
UV_PUBLISH_TOKEN="pypi token here" uv publish

For an isolated Python environment, use uv to manage the project and dependencies. Follow these steps:

1. Install uv:

   pip install uv

2. Initialize a new project with Python 3.12:
   ```console
uv init -p 3.12

3. Add the required dependencies:

   uv add kokoro-onnx soundfile

4. Download the required model files (`kokoro-v1.0.onnx` and `voices-v1.0.bin`) and place them in your project directory.
5. Run your script using `uv run`.

```console
uv init -p 3.12
uv add kokoro-onnx soundfile
uv run hello.py

Use ruff via uv run to format the codebase and check for linting errors.

uv run ruff format
uv run ruff check

You can enable debug-level logging by setting the LOG_LEVEL environment variable to DEBUG when running the main script.

LOG_LEVEL=DEBUG python main.py

The project supports multiple languages and multiple voices. For a complete list of the latest available voices and supported languages, refer to the Kokoro-82M/VOICES.md documentation.

Note: It is recommended to use the misaki g2p package for version 1.0 workflows.

If you encounter a FileNotFoundError during initialization, ensure your paths are correct. If files are missing, you can download them using the following resources:

• Voices file: Download voices-v1.0.bin using: wget https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files-v1.0/voices-v1.0.bin
• Model file: Download from the Kokoro-ONNX releases page.

The Tokenizer class processes text input for the TTS engine. You can initialize it with an optional EspeakConfig object to specify paths for the espeak-ng library and data, or a custom vocab dictionary.

If no configuration is provided, it attempts to locate the espeak-ng library automatically. You can also control the library path by setting the PHONEMIZER_ESPEAK_LIBRARY environment variable.

For low-latency applications, use create_stream to asynchronously yield audio chunks as they are processed. This method runs the heavy computation in a background thread to avoid blocking the event loop.

Parameters: Same as create().

Returns:

• An AsyncGenerator yielding (audio_numpy_array, sample_rate) tuples.