{
  "id": 213867,
  "title": "Denoise data",
  "url": "/competitions/rfcx-species-audio-detection/discussion/213867",
  "author_name": "",
  "post_date": "2021-01-24T15:14:30.464700Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Are you using some denoise to preprocess your data? Now a lot of preprocessing comes down to <code>augmentations</code> (moreover, both to <code>sound</code> files directly, and to <code>images</code>, for example, for <code>ResNet</code>). </p>\n<p>But maybe there is a more interesting way to clean up the data like denoise? At first glance, it seems like a pretty good preprocessing method. Therefore, if anyone uses this, please share links to articles / old notebooks / library repositories. Has anyone used denoise from <a href=\"https://github.com/facebookresearch/denoiser\" target=\"_blank\">facebook</a>?</p>",
  "messages": [
    {
      "id": "1167911",
      "postDate": "01/24/2021 15:14:30",
      "content": "<p>Are you using some denoise to preprocess your data? Now a lot of preprocessing comes down to <code>augmentations</code> (moreover, both to <code>sound</code> files directly, and to <code>images</code>, for example, for <code>ResNet</code>). </p>\n<p>But maybe there is a more interesting way to clean up the data like denoise? At first glance, it seems like a pretty good preprocessing method. Therefore, if anyone uses this, please share links to articles / old notebooks / library repositories. Has anyone used denoise from <a href=\"https://github.com/facebookresearch/denoiser\" target=\"_blank\">facebook</a>?</p>",
      "rawMarkdown": "Are you using some denoise to preprocess your data? Now a lot of preprocessing comes down to `augmentations` (moreover, both to `sound` files directly, and to `images`, for example, for `ResNet`). \n\nBut maybe there is a more interesting way to clean up the data like denoise? At first glance, it seems like a pretty good preprocessing method. Therefore, if anyone uses this, please share links to articles / old notebooks / library repositories. Has anyone used denoise from [facebook](https://github.com/facebookresearch/denoiser)?",
      "votes": null
    },
    {
      "id": "1168095",
      "postDate": "01/24/2021 17:53:20",
      "content": "<p>I tried, but it removed a lot of details since the pretarined Facebook's denoiser was trained using speech data. If you're able to train your own denoiser using sound from Mother Nature, I think it will help</p>",
      "rawMarkdown": "I tried, but it removed a lot of details since the pretarined Facebook's denoiser was trained using speech data. If you're able to train your own denoiser using sound from Mother Nature, I think it will help",
      "votes": null
    },
    {
      "id": "1168479",
      "postDate": "01/25/2021 01:14:36",
      "content": "<p>I'm trying to use denoise like <a href=\"https://www.kaggle.com/c/birdsong-recognition/discussion/179290\" target=\"_blank\">previous competition</a>.<br>\nI plan to share datasets and notebooks soon.</p>",
      "rawMarkdown": "I'm trying to use denoise like [previous competition](https://www.kaggle.com/c/birdsong-recognition/discussion/179290).\nI plan to share datasets and notebooks soon.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1168095,
      "author_name": "jihangz",
      "author_url": "",
      "post_date": "01/24/2021 17:53:20",
      "content": "<p>I tried, but it removed a lot of details since the pretarined Facebook's denoiser was trained using speech data. If you're able to train your own denoiser using sound from Mother Nature, I think it will help</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1168479,
      "author_name": "takamichitoda",
      "author_url": "",
      "post_date": "01/25/2021 01:14:36",
      "content": "<p>I'm trying to use denoise like <a href=\"https://www.kaggle.com/c/birdsong-recognition/discussion/179290\" target=\"_blank\">previous competition</a>.<br>\nI plan to share datasets and notebooks soon.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1167911": "Are you using some denoise to preprocess your data? Now a lot of preprocessing comes down to `augmentations` (moreover, both to `sound` files directly, and to `images`, for example, for `ResNet`). \n\nBut maybe there is a more interesting way to clean up the data like denoise? At first glance, it seems like a pretty good preprocessing method. Therefore, if anyone uses this, please share links to articles / old notebooks / library repositories. Has anyone used denoise from [facebook](https://github.com/facebookresearch/denoiser)?",
    "1168095": "I tried, but it removed a lot of details since the pretarined Facebook's denoiser was trained using speech data. If you're able to train your own denoiser using sound from Mother Nature, I think it will help",
    "1168479": "I'm trying to use denoise like [previous competition](https://www.kaggle.com/c/birdsong-recognition/discussion/179290).\nI plan to share datasets and notebooks soon."
  },
  "source": "meta"
}