README.md

metadata

language:
  - ckb
license: other
task_categories:
  - automatic-speech-recognition
pretty_name: Central Kurdish Speech Dataset
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*.parquet
dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: audio_filename
      dtype: string
    - name: transcription
      dtype: string
    - name: category
      dtype: string
    - name: speaker_id
      dtype: string
    - name: gender
      dtype: string
    - name: cer
      dtype: float32
    - name: downvotes
      dtype: int32
    - name: upvotes
      dtype: int32
    - name: validator
      dtype: string
    - name: source
      dtype: string
  splits:
    - name: train
      num_examples: 260571

Central Kurdish Speech Dataset

This dataset contains Central Kurdish speech recordings with audio embedded inside Parquet files.

Loading

from datasets import load_dataset

dataset = load_dataset("aranemini/bawajai", split="train")
print(dataset)
print(dataset[0])

Fields

• audio: embedded audio
• audio_filename: original audio filename
• transcription: Kurdish transcription
• category: validation category
• speaker_id: anonymous speaker identifier
• gender: speaker gender or -
• cer: character error rate
• downvotes: validation downvotes
• upvotes: validation upvotes
• validator: validator identifier
• source: original source

The validated_at field is intentionally excluded.

Statistics

• Recordings: 260,571
• Speakers: 496
• Parquet files: 261
• Maximum examples per Parquet: 1,000

Categories

• manually-validated: 136,285
• automatically-validated: 108,459
• automatically-qualified: 15,827

Gender

• female: 132,403
• male: 87,105
• -: 41,063