README.md

metadata

license: cc-by-nc-4.0
language:
  - he
tags:
  - hebrew
  - asr
  - automatic-speech-recognition
  - text-to-speech
task_categories:
  - text-to-speech

🎙️ Osim Historia (Ran Levi) - 40 Hours Hebrew Audio Dataset IPA Transcribed

Dataset Description

This dataset contains approximately 40 hours of high-quality, processed Hebrew speech audio sourced from the popular Israeli podcast "Osim Historia" (עושים היסטוריה), hosted by Ran Levi. The audio features clear, narrative-style speaking with professional recording quality, making it an exceptional resource for training and fine-tuning Hebrew speech models.

• Language: Hebrew (he)
• Total Duration: ~40 Hours
• Primary Speaker: Ran Levi (Male)
• Audio Format: WAV, 44.1kHz (Mono)

Uses

This dataset is uniquely suited for advancing Hebrew audio AI. Primary use cases include: Text-to-Speech (TTS): Excellent for fine-tuning modern acoustic models and vocoders to achieve highly natural, prosody-rich Hebrew synthesis with IPA as input text.

Dataset Structure

The dataset is structured for immediate compatibility with the datasets library.