{
  "id": 451543,
  "title": "92th Place Solution",
  "url": "/competitions/bengaliai-speech/discussion/451543",
  "author_name": "KhanhVD",
  "post_date": "2023-10-29T13:40:44.661000",
  "votes": 4,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Thank you Bengali.AI &amp; Kaggle for organizing this competition. </p>\n<h1>Overview of the Approach</h1>\n<h2>ASR Model</h2>\n<ol>\n<li>Dataset: I only use small subset of competition dataset with filter audios which duration is less then 1 sec. as outlier. Removing all kinds of punctuation. Removing all kinds of punctuation then Normalization</li>\n<li>Augmentation: BackgroundNoise, Gain</li>\n<li>Pretrain model: arijitx-full-model/wav2vec2-xls-r-300m-bengali</li>\n</ol>\n<p>I don't have the time or resources to train more epochs (the model only train on 1 epoch). If possible the results could be even better!</p>\n<h2>Language Model</h2>\n<p>Same as public notebook: 5-gram LM (arijitx-full-model/wav2vec2-xls-r-300m-bengali)</p>",
  "messages": [
    {
      "id": 2503831,
      "postDate": "2023-10-29T13:40:44.660Z",
      "content": "<p>Thank you Bengali.AI &amp; Kaggle for organizing this competition. </p>\n<h1>Overview of the Approach</h1>\n<h2>ASR Model</h2>\n<ol>\n<li>Dataset: I only use small subset of competition dataset with filter audios which duration is less then 1 sec. as outlier. Removing all kinds of punctuation. Removing all kinds of punctuation then Normalization</li>\n<li>Augmentation: BackgroundNoise, Gain</li>\n<li>Pretrain model: arijitx-full-model/wav2vec2-xls-r-300m-bengali</li>\n</ol>\n<p>I don't have the time or resources to train more epochs (the model only train on 1 epoch). If possible the results could be even better!</p>\n<h2>Language Model</h2>\n<p>Same as public notebook: 5-gram LM (arijitx-full-model/wav2vec2-xls-r-300m-bengali)</p>",
      "rawMarkdown": "Thank you Bengali.AI & Kaggle for organizing this competition. \n# Overview of the Approach\n## ASR Model\n1. Dataset: I only use small subset of competition dataset with filter audios which duration is less then 1 sec. as outlier. Removing all kinds of punctuation. Removing all kinds of punctuation then Normalization\n2. Augmentation: BackgroundNoise, Gain\n3. Pretrain model: arijitx-full-model/wav2vec2-xls-r-300m-bengali\n\nI don't have the time or resources to train more epochs (the model only train on 1 epoch). If possible the results could be even better!\n## Language Model\nSame as public notebook: 5-gram LM (arijitx-full-model/wav2vec2-xls-r-300m-bengali)",
      "votes": 4
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2503831": "Thank you Bengali.AI & Kaggle for organizing this competition. \n# Overview of the Approach\n## ASR Model\n1. Dataset: I only use small subset of competition dataset with filter audios which duration is less then 1 sec. as outlier. Removing all kinds of punctuation. Removing all kinds of punctuation then Normalization\n2. Augmentation: BackgroundNoise, Gain\n3. Pretrain model: arijitx-full-model/wav2vec2-xls-r-300m-bengali\n\nI don't have the time or resources to train more epochs (the model only train on 1 epoch). If possible the results could be even better!\n## Language Model\nSame as public notebook: 5-gram LM (arijitx-full-model/wav2vec2-xls-r-300m-bengali)"
  }
}