ArthurYeghinyan/Mixed_dataset_speech

Mixed Armenian Whisper Speech Dataset

This dataset is a mixed Whisper-ready Armenian speech dataset built from:

• Arthuryann/Youtube_azatutyun_speech
• Arthuryann/Youtube_panarmeniantv_speech
• Arthuryann/Youtube_armcomedy_speech

Splits are created by source + video_id, so segments from the same source video do not leak across train/validation/test.

Stats

{
  "splits": {
    "train": {
      "rows": 106961,
      "hours": 668.4926,
      "shards": 111
    },
    "validation": {
      "rows": 2433,
      "hours": 14.6586,
      "shards": 3
    },
    "test": {
      "rows": 1915,
      "hours": 11.8534,
      "shards": 3
    }
  },
  "source_repos": [
    "Arthuryann/Youtube_azatutyun_speech",
    "Arthuryann/Youtube_panarmeniantv_speech",
    "Arthuryann/Youtube_armcomedy_speech"
  ],
  "source_split": "whisper_clean",
  "total_rows": 111309,
  "total_hours": 695.0046,
  "split_policy": {
    "key": "source::video_id",
    "validation_ratio_by_video": 0.02,
    "test_ratio_by_video": 0.02,
    "seed": 42
  }
}