{
  "id": 500853,
  "title": "with few dataset and heavy augmentation is there a hope?",
  "url": "/competitions/birdclef-2024/discussion/500853",
  "author_name": "",
  "post_date": "2024-05-07T06:47:16.145245Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I am wondering if with a clean audio sample like 30 seconds for each bird and heavy augmentation. I have manually cleaned up 30-sec samples of 31 birds (I am tired, I need help.) I choose this approach because of the inconsistent dataset.</p>\n<p>Using Audacity I removed bg noise and spectrogram delete to remove multiple birds noise.<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14899462%2F8597f1b2810528c26ffcf257ac6f489a%2FScreenshot%20from%202024-05-07%2012-11-23.png?generation=1715064139285816&amp;alt=media\"></p>\n<p>tbh, I don't know if I should continue with this approach.</p>",
  "messages": [
    {
      "id": "2798219",
      "postDate": "05/07/2024 06:47:16",
      "content": "<p>I am wondering if with a clean audio sample like 30 seconds for each bird and heavy augmentation. I have manually cleaned up 30-sec samples of 31 birds (I am tired, I need help.) I choose this approach because of the inconsistent dataset.</p>\n<p>Using Audacity I removed bg noise and spectrogram delete to remove multiple birds noise.<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14899462%2F8597f1b2810528c26ffcf257ac6f489a%2FScreenshot%20from%202024-05-07%2012-11-23.png?generation=1715064139285816&amp;alt=media\"></p>\n<p>tbh, I don't know if I should continue with this approach.</p>",
      "rawMarkdown": "I am wondering if with a clean audio sample like 30 seconds for each bird and heavy augmentation. I have manually cleaned up 30-sec samples of 31 birds (I am tired, I need help.) I choose this approach because of the inconsistent dataset.\n\n\nUsing Audacity I removed bg noise and spectrogram delete to remove multiple birds noise.![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14899462%2F8597f1b2810528c26ffcf257ac6f489a%2FScreenshot%20from%202024-05-07%2012-11-23.png?generation=1715064139285816&alt=media)\n\ntbh, I don't know if I should continue with this approach.",
      "votes": null
    },
    {
      "id": "2798349",
      "postDate": "05/07/2024 07:34:06",
      "content": "<p>I think that it is possible to obtain good results in cross-valuation, but in hidden testing, obtaining the noise profile maybe difficult.</p>",
      "rawMarkdown": "I think that it is possible to obtain good results in cross-valuation, but in hidden testing, obtaining the noise profile maybe difficult.",
      "votes": null
    },
    {
      "id": "2798506",
      "postDate": "05/07/2024 08:56:26",
      "content": "<p>noise profile can be obtained from unlabled_soundscapes</p>",
      "rawMarkdown": "noise profile can be obtained from unlabled_soundscapes",
      "votes": null
    },
    {
      "id": "2798660",
      "postDate": "05/07/2024 10:37:04",
      "content": "<p>Is there any difficulty to determine the one-to-one relationship with the test set?</p>",
      "rawMarkdown": "Is there any difficulty to determine the one-to-one relationship with the test set?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2798349,
      "author_name": "tanxxx",
      "author_url": "",
      "post_date": "05/07/2024 07:34:06",
      "content": "<p>I think that it is possible to obtain good results in cross-valuation, but in hidden testing, obtaining the noise profile maybe difficult.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2798506,
          "author_name": "jewinv",
          "author_url": "",
          "post_date": "05/07/2024 08:56:26",
          "content": "<p>noise profile can be obtained from unlabled_soundscapes</p>",
          "votes": null,
          "replies": [
            {
              "id": 2798660,
              "author_name": "tanxxx",
              "author_url": "",
              "post_date": "05/07/2024 10:37:04",
              "content": "<p>Is there any difficulty to determine the one-to-one relationship with the test set?</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2798219": "I am wondering if with a clean audio sample like 30 seconds for each bird and heavy augmentation. I have manually cleaned up 30-sec samples of 31 birds (I am tired, I need help.) I choose this approach because of the inconsistent dataset.\n\n\nUsing Audacity I removed bg noise and spectrogram delete to remove multiple birds noise.![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14899462%2F8597f1b2810528c26ffcf257ac6f489a%2FScreenshot%20from%202024-05-07%2012-11-23.png?generation=1715064139285816&alt=media)\n\ntbh, I don't know if I should continue with this approach.",
    "2798349": "I think that it is possible to obtain good results in cross-valuation, but in hidden testing, obtaining the noise profile maybe difficult.",
    "2798506": "noise profile can be obtained from unlabled_soundscapes",
    "2798660": "Is there any difficulty to determine the one-to-one relationship with the test set?"
  },
  "source": "meta"
}