israwave

Mission to create a Hebrew TTS model as powerful and user-friendly as WaveNet

Features

• Generate sentence in less than 1ms on CPU
• Powerful text processor by espeak-ng
• Support for SSML (soon)

Play with it!

You can play with it on HuggingFace Space

Samples

https://github.com/user-attachments/assets/3212a800-406f-4d79-8aa1-d814eed815d6

Setup

pip install -U israwave

• You also need to place the files israwave.onnx, nakdimon.onnx, and extracted espeak-ng-data in the project folder.
• We recommend use uv for Python.
• Python 3.12 is recommended. (uv init -p 3.12)
• Please see examples.

Examples

See examples

Dataset

The model trained on saspeech gold standard.

Thanks

Thanks to Kan11 and Shaul for providing the dataset.

Thanks to elazarg for sharing the Nakdimon diacritics model, which was instrumental in our project.

For mush42 for their excellent TTS training recipe.