{
  "id": 130860,
  "title": "Audio Classification",
  "url": "/competitions/deepfake-detection-challenge/discussion/130860",
  "author_name": "",
  "post_date": "2020-02-16T22:25:20.585177600Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Did anyone manage to get any improvement to their networks performance by also using the audio tracks?\nI've tried several approaches and see no improvement at all when comparing the performance against a images only model.</p>",
  "messages": [
    {
      "id": "747817",
      "postDate": "02/16/2020 22:25:20",
      "content": "<p>Did anyone manage to get any improvement to their networks performance by also using the audio tracks?\nI've tried several approaches and see no improvement at all when comparing the performance against a images only model.</p>",
      "rawMarkdown": "Did anyone manage to get any improvement to their networks performance by also using the audio tracks?\nI've tried several approaches and see no improvement at all when comparing the performance against a images only model.",
      "votes": null
    },
    {
      "id": "747827",
      "postDate": "02/16/2020 22:50:31",
      "content": "<p>I've not had any luck with it, see this post <a href=\"https://www.kaggle.com/c/deepfake-detection-challenge/discussion/130502#746479\">https://www.kaggle.com/c/deepfake-detection-challenge/discussion/130502#746479</a></p>",
      "rawMarkdown": "I've not had any luck with it, see this post https://www.kaggle.com/c/deepfake-detection-challenge/discussion/130502#746479",
      "votes": null
    },
    {
      "id": "747829",
      "postDate": "02/16/2020 22:55:56",
      "content": "<p>Weird. Or we haven't yet found the right network architecture to extract patterns from the audio, or the percentage of movies, that only have fake audio, and no fake faces is so small that it really doesn't affect the overall performance.</p>",
      "rawMarkdown": "Weird. Or we haven't yet found the right network architecture to extract patterns from the audio, or the percentage of movies, that only have fake audio, and no fake faces is so small that it really doesn't affect the overall performance.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 747827,
      "author_name": "jamesphoward",
      "author_url": "",
      "post_date": "02/16/2020 22:50:31",
      "content": "<p>I've not had any luck with it, see this post <a href=\"https://www.kaggle.com/c/deepfake-detection-challenge/discussion/130502#746479\">https://www.kaggle.com/c/deepfake-detection-challenge/discussion/130502#746479</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 747829,
          "author_name": "ngcferreira",
          "author_url": "",
          "post_date": "02/16/2020 22:55:56",
          "content": "<p>Weird. Or we haven't yet found the right network architecture to extract patterns from the audio, or the percentage of movies, that only have fake audio, and no fake faces is so small that it really doesn't affect the overall performance.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "747817": "Did anyone manage to get any improvement to their networks performance by also using the audio tracks?\nI've tried several approaches and see no improvement at all when comparing the performance against a images only model.",
    "747827": "I've not had any luck with it, see this post https://www.kaggle.com/c/deepfake-detection-challenge/discussion/130502#746479",
    "747829": "Weird. Or we haven't yet found the right network architecture to extract patterns from the audio, or the percentage of movies, that only have fake audio, and no fake faces is so small that it really doesn't affect the overall performance."
  },
  "source": "meta"
}