ivrit.ai

ivrit.ai is an effort to provide high-quality Hebrew datasets under a permissive license. It is our hope that such datasets will be used to enable first-class support for Hebrew in AI models.

More about us can be found at ivrit.ai.

We are proud to present our latest achievements:

1. A state-of-the-art Hebrew speech-to-text model: https://huggingface.co/ivrit-ai/whisper-large-v3-ct2

2. A turbo-based Hebrew speech-to-text model: https://huggingface.co/ivrit-ai/whisper-large-v3-turbo-ct2

3. Our newest comprehensive Hebrew language dataset: https://huggingface.co/datasets/ivrit-ai/crowd-transcribe-v5

Paper: https://www.isca-archive.org/interspeech_2025/marmor25_interspeech.pdf

If you use our datasets or models, the following quote is preferable:

@inproceedings{marmor2025building,
  title={Building an Accurate Open-Source Hebrew ASR System through Crowdsourcing},
  author={Marmor, Yanir and Lifshitz, Yair and Snapir, Yoad and Misgav, Kinneret},
  booktitle={Proc. Interspeech 2025},
  pages={723--727},
  year={2025}
}