{
  "fetched_at": "2026-10-19T06:29:04Z",
  "source_url": "https://www.kaggle.com/competitions/automatic-speech-recognition-hw4p2-slack/data",
  "pages": [
    {
      "name": "data-description",
      "content": "The dataset is given in a similar format to previous homework in this course.\n\n## Subdirectories\n\n*   **train-clean-50** - the training set\n*   **train-clean-100** - the training set\n*   **dev-clean** - the training set\n*   **test-clean** - the training set\n\n## Files\n\n*   `encoder_weights.pth` - reference CNN-LSTM encoder weights",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "You will be evaluated based on the Levenshtein distance between your output sequences and the ground truth sequences for the test dataset of LibriSpeech.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "- Don't cheat!\n - Apply yourself!\n - Have fun!",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "**Attention-based Speech Recognition**\n\nIn this competition you will work on a sequence-to-sequence conversion problem, in which you must\ntrain attention-based (transformer) models to transduce speech recordings into word sequences, spelled out alphabetically. This competition is a part of Carnegie Mellon's Introduction to Deep Learning (11-785) course.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}