{
  "id": 450682,
  "title": "My 78th Position Solution.",
  "url": "/competitions/bengaliai-speech/discussion/450682",
  "author_name": "Pritam Sinha",
  "post_date": "2023-10-25T08:54:54.189000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>First of all, thanks for this wonderful competition hosting. I do not have a GPU or any service I can use to train the data fully. So, I basically fine-tuned other's trained models on a small portion of the data.</p>\n<p>I used Joseph Josia's trained model <a href=\"https://www.kaggle.com/datasets/takanashihumbert/bengali-ex006\" target=\"_blank\">bengali-ex006</a> to finetune the model.</p>\n<p>For decoding, I used the 5-gram model.</p>\n<p>My notebook: <a href=\"https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-v6/notebook\" target=\"_blank\">https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-v6/notebook</a><br>\nMy train notebook: <a href=\"https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-train-v4/notebook\" target=\"_blank\">https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-train-v4/notebook</a></p>",
  "messages": [
    {
      "id": 2498351,
      "postDate": "2023-10-25T08:54:54.190Z",
      "content": "<p>First of all, thanks for this wonderful competition hosting. I do not have a GPU or any service I can use to train the data fully. So, I basically fine-tuned other's trained models on a small portion of the data.</p>\n<p>I used Joseph Josia's trained model <a href=\"https://www.kaggle.com/datasets/takanashihumbert/bengali-ex006\" target=\"_blank\">bengali-ex006</a> to finetune the model.</p>\n<p>For decoding, I used the 5-gram model.</p>\n<p>My notebook: <a href=\"https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-v6/notebook\" target=\"_blank\">https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-v6/notebook</a><br>\nMy train notebook: <a href=\"https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-train-v4/notebook\" target=\"_blank\">https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-train-v4/notebook</a></p>",
      "rawMarkdown": "First of all, thanks for this wonderful competition hosting. I do not have a GPU or any service I can use to train the data fully. So, I basically fine-tuned other's trained models on a small portion of the data.\n\nI used Joseph Josia's trained model [bengali-ex006] (https://www.kaggle.com/datasets/takanashihumbert/bengali-ex006) to finetune the model.\n\nFor decoding, I used the 5-gram model.\n\nMy notebook: https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-v6/notebook\nMy train notebook: https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-train-v4/notebook",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2498351": "First of all, thanks for this wonderful competition hosting. I do not have a GPU or any service I can use to train the data fully. So, I basically fine-tuned other's trained models on a small portion of the data.\n\nI used Joseph Josia's trained model [bengali-ex006] (https://www.kaggle.com/datasets/takanashihumbert/bengali-ex006) to finetune the model.\n\nFor decoding, I used the 5-gram model.\n\nMy notebook: https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-v6/notebook\nMy train notebook: https://www.kaggle.com/code/pritamsinha23/bengali-speech-kaggle-wav2vec2-train-v4/notebook"
  }
}