crowd-recital-whisper-training

# Dataset Card for ivrit.ai - Crowd Recital <!-- Provide a quick summary of the dataset. --> ## Dataset Details ### Dataset Description <!-- Provide a longer summary of what this dataset is. --> # License The dataset is released under the ivrit.ai License, which enables broad research and commercial use. - Full license: https://www.ivrit.ai/en/the-license/ - FAQs: https://www.ivrit.ai/en/license-faqs/ - ## Dataset Structure ### Data Fields Each example in the dataset contains: - `audio`: An audio column containing: - `bytes`: The audio data encoded in MP3 format - `path`: A string identifier derived from the source entry ID - Sampling rate: Fixed at 16000 Hz - `transcript`: A string containing the text with potentially Whisper-style timestamp tokens (e.g., `<|0.00|>text<|2.40|>`) if "has_timestamps" is true - `metadata`: A dictionary containing: - `seek`: Float indicating the start time of this slice in the original source audio - `source`: String identifier for the source of the audio (Name of podcast, production system, etc.) - `entry_id`: Unique identifier for the source entry - `has_prev`: Boolean indicating if this slice has transcript from the previous slice within the audio source - `has_timestamps`: Boolean indicating if the transcript contains timestamp tokens - `prev_transcript`: String containing the transcript of the previous slice (empty if `has_prev` is false)