ivrit-ai/crowd-whatsapp-yi-whisper-training

Dataset Card for ivrit.ai - Crowd Whatsapp - Yiddish

See more details on the source dataset card.

Dataset Details

Dataset Description

This is a derived dataset for structured for whisper training:

• Excludes low quality segments (judged by probabilities of the text-audio auto alignment process)

• Encodes timestamps along segments of text + previous text

• Audio encoded to 16K sample-rate, mono

• Total audio duration - ~19h

• License: other

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)