ivrit.ai — open Hebrew speech & language AI

ivrit.ai is a non-profit building open, permissively licensed Hebrew (and Yiddish) datasets and models. Models: apache-2.0. Datasets: the ivrit.ai License (v2) — which allows broad research and commercial use.

🚀 Start here — which model should I use?

# pip install faster-whisper
from faster_whisper import WhisperModel
model = WhisperModel("ivrit-ai/whisper-large-v3-turbo-ct2")
segments, _ = model.transcribe("audio.wav", language="he")
print(" ".join(s.text for s in segments))

No code? Try it in .

📊 Leaderboards & arenas

• Hebrew Transcription Leaderboard — compare Hebrew ASR models
• TTS Arena Hebrew — vote on Hebrew text-to-speech
• Hebrew Handwriting OCR Leaderboard

🗂️ Datasets

Older versions stay online but are marked Deprecated on their cards.

🤝 Contribute

• Record or transcribe Hebrew at ivrit.ai
• Code & issues: GitHub
• Questions: open a Discussion on any repo or email to info at ivrit.ai

📝 Cite

Marmor, Lifshitz, Snapir, Misgav. Building an Accurate Open-Source Hebrew ASR System through Crowdsourcing. Interspeech 2025. doi:10.21437/Interspeech.2025-460

ivrit.ai: A Comprehensive Dataset of Hebrew Speech for AI Research and Development — arXiv:2307.08720