{
  "id": 89108,
  "title": "sound 1d44b0bd.wav",
  "url": "/competitions/freesound-audio-tagging-2019/discussion/89108",
  "author_name": "",
  "post_date": "2019-04-11T15:04:51.080139900Z",
  "votes": 6,
  "comment_count": 7,
  "views": 0,
  "content": "<p>There is no sound or I have some issue in the file?</p>",
  "messages": [
    {
      "id": "514310",
      "postDate": "04/11/2019 15:04:51",
      "content": "<p>There is no sound or I have some issue in the file?</p>",
      "rawMarkdown": "There is no sound or I have some issue in the file?",
      "votes": null
    },
    {
      "id": "514580",
      "postDate": "04/11/2019 18:56:33",
      "content": "<p>It looks like 1d44b0bd.wav in the curated train set happens to be digital silence (all frames are exactly zero). <a href=\"/eduardofonseca\">@eduardofonseca</a> </p>\n\n<p>Ideally, your data processing pipeline should be able to handle files like this (which could easily show up in a real data set if you take random subclips of a large clip).</p>",
      "rawMarkdown": "It looks like 1d44b0bd.wav in the curated train set happens to be digital silence (all frames are exactly zero). @eduardofonseca \n\nIdeally, your data processing pipeline should be able to handle files like this (which could easily show up in a real data set if you take random subclips of a large clip).",
      "votes": null
    },
    {
      "id": "514882",
      "postDate": "04/12/2019 02:07:59",
      "content": "<p>Thanks for pointing that out! Will check it out. In theory, the original file shouldn't be digital silence as it was annotated with a label by a human :)</p>\n\n<p>Perhaps something went wrong in the format conversion... (although there was no error). Will look into it.</p>",
      "rawMarkdown": "Thanks for pointing that out! Will check it out. In theory, the original file shouldn't be digital silence as it was annotated with a label by a human :)\n\nPerhaps something went wrong in the format conversion... (although there was no error). Will look into it.",
      "votes": null
    },
    {
      "id": "515352",
      "postDate": "04/12/2019 13:37:17",
      "content": "<p>I just stumbled as well over this. In general the quality of the data set seems not to be optimal to say it politely. For example the \"Gasp,Sigh\" labeled clips. I checked the english dictionary if i maybe misunderstood something.</p>\n\n<p>train_noisy/ff97b092.wav \ntrain_noisy/e140e930.wav\nI don't hear a single gasp or sigh in either of those. And i haven't heared much into the data.</p>",
      "rawMarkdown": "I just stumbled as well over this. In general the quality of the data set seems not to be optimal to say it politely. For example the \"Gasp,Sigh\" labeled clips. I checked the english dictionary if i maybe misunderstood something.\n\ntrain_noisy/ff97b092.wav \ntrain_noisy/e140e930.wav\nI don't hear a single gasp or sigh in either of those. And i haven't heared much into the data.",
      "votes": null
    },
    {
      "id": "515380",
      "postDate": "04/12/2019 14:27:30",
      "content": "<p>The noisy training data is intended to have unreliable labels. As we tried to explain in the challenge description, we're trying to simulate a real-world scenario where you have a small amount of relatively well-labeled data and a much larger amount of weakly labeled data.</p>",
      "rawMarkdown": "The noisy training data is intended to have unreliable labels. As we tried to explain in the challenge description, we're trying to simulate a real-world scenario where you have a small amount of relatively well-labeled data and a much larger amount of weakly labeled data.",
      "votes": null
    },
    {
      "id": "515413",
      "postDate": "04/12/2019 15:14:04",
      "content": "<p>Thanks, Manoj. I will try to filter out the ones, that seem to have benefit.</p>\n\n<p>The following files from \"train_noisy\" are completely silent as well. They are all exaclty 15 seconds long.:\n<code>\n02f274b2.wav\n08b34136.wav\n1af3bd88.wav\n1fd4f275.wav\n2f503375.wav\n3496256e.wav\n551a4b3b.wav\n5a5761c9.wav\n6d062e59.wav\n769d131d.wav\n8c712129.wav\n988cf8f2.wav\n9f4fa2df.wav\nb1d2590c.wav\nbe273a3c.wav\nd527dcf0.wav\ne4faa2e1.wav\nfa659a71.wav\nfba392d8.wav\n</code></p>",
      "rawMarkdown": "Thanks, Manoj. I will try to filter out the ones, that seem to have benefit.\n\nThe following files from \"train_noisy\" are completely silent as well. They are all exaclty 15 seconds long.:\n```\n02f274b2.wav\n08b34136.wav\n1af3bd88.wav\n1fd4f275.wav\n2f503375.wav\n3496256e.wav\n551a4b3b.wav\n5a5761c9.wav\n6d062e59.wav\n769d131d.wav\n8c712129.wav\n988cf8f2.wav\n9f4fa2df.wav\nb1d2590c.wav\nbe273a3c.wav\nd527dcf0.wav\ne4faa2e1.wav\nfa659a71.wav\nfba392d8.wav\n```",
      "votes": null
    },
    {
      "id": "522607",
      "postDate": "04/24/2019 17:56:21",
      "content": "<p>i hear no sound , also in plot all signal is zero !</p>",
      "rawMarkdown": "i hear no sound , also in plot all signal is zero !",
      "votes": null
    },
    {
      "id": "529449",
      "postDate": "05/09/2019 23:21:11",
      "content": "<p>I confirm that the clip 1d44b0bd.wav contained a (rather scary) perfectly audible <em>Whispering</em> in its original format. But in the version uploaded to Kaggle the file is indeed corrupted (empty). My guess is this happened during format conversion to 44.1 kHz @16bit without ffmpeg complaining. </p>\n\n<p>thanks for letting us know. Will double check format conversions in the future!</p>",
      "rawMarkdown": "I confirm that the clip 1d44b0bd.wav contained a (rather scary) perfectly audible *Whispering* in its original format. But in the version uploaded to Kaggle the file is indeed corrupted (empty). My guess is this happened during format conversion to 44.1 kHz @16bit without ffmpeg complaining. \n\nthanks for letting us know. Will double check format conversions in the future!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 514580,
      "author_name": "plakal",
      "author_url": "",
      "post_date": "04/11/2019 18:56:33",
      "content": "<p>It looks like 1d44b0bd.wav in the curated train set happens to be digital silence (all frames are exactly zero). <a href=\"/eduardofonseca\">@eduardofonseca</a> </p>\n\n<p>Ideally, your data processing pipeline should be able to handle files like this (which could easily show up in a real data set if you take random subclips of a large clip).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 514882,
      "author_name": "eduardofonseca",
      "author_url": "",
      "post_date": "04/12/2019 02:07:59",
      "content": "<p>Thanks for pointing that out! Will check it out. In theory, the original file shouldn't be digital silence as it was annotated with a label by a human :)</p>\n\n<p>Perhaps something went wrong in the format conversion... (although there was no error). Will look into it.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 515352,
      "author_name": "marekwyborski",
      "author_url": "",
      "post_date": "04/12/2019 13:37:17",
      "content": "<p>I just stumbled as well over this. In general the quality of the data set seems not to be optimal to say it politely. For example the \"Gasp,Sigh\" labeled clips. I checked the english dictionary if i maybe misunderstood something.</p>\n\n<p>train_noisy/ff97b092.wav \ntrain_noisy/e140e930.wav\nI don't hear a single gasp or sigh in either of those. And i haven't heared much into the data.</p>",
      "votes": null,
      "replies": [
        {
          "id": 515380,
          "author_name": "plakal",
          "author_url": "",
          "post_date": "04/12/2019 14:27:30",
          "content": "<p>The noisy training data is intended to have unreliable labels. As we tried to explain in the challenge description, we're trying to simulate a real-world scenario where you have a small amount of relatively well-labeled data and a much larger amount of weakly labeled data.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 515413,
          "author_name": "marekwyborski",
          "author_url": "",
          "post_date": "04/12/2019 15:14:04",
          "content": "<p>Thanks, Manoj. I will try to filter out the ones, that seem to have benefit.</p>\n\n<p>The following files from \"train_noisy\" are completely silent as well. They are all exaclty 15 seconds long.:\n<code>\n02f274b2.wav\n08b34136.wav\n1af3bd88.wav\n1fd4f275.wav\n2f503375.wav\n3496256e.wav\n551a4b3b.wav\n5a5761c9.wav\n6d062e59.wav\n769d131d.wav\n8c712129.wav\n988cf8f2.wav\n9f4fa2df.wav\nb1d2590c.wav\nbe273a3c.wav\nd527dcf0.wav\ne4faa2e1.wav\nfa659a71.wav\nfba392d8.wav\n</code></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 522607,
      "author_name": "",
      "author_url": "",
      "post_date": "04/24/2019 17:56:21",
      "content": "<p>i hear no sound , also in plot all signal is zero !</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 529449,
      "author_name": "eduardofonseca",
      "author_url": "",
      "post_date": "05/09/2019 23:21:11",
      "content": "<p>I confirm that the clip 1d44b0bd.wav contained a (rather scary) perfectly audible <em>Whispering</em> in its original format. But in the version uploaded to Kaggle the file is indeed corrupted (empty). My guess is this happened during format conversion to 44.1 kHz @16bit without ffmpeg complaining. </p>\n\n<p>thanks for letting us know. Will double check format conversions in the future!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "514310": "There is no sound or I have some issue in the file?",
    "514580": "It looks like 1d44b0bd.wav in the curated train set happens to be digital silence (all frames are exactly zero). @eduardofonseca \n\nIdeally, your data processing pipeline should be able to handle files like this (which could easily show up in a real data set if you take random subclips of a large clip).",
    "514882": "Thanks for pointing that out! Will check it out. In theory, the original file shouldn't be digital silence as it was annotated with a label by a human :)\n\nPerhaps something went wrong in the format conversion... (although there was no error). Will look into it.",
    "515352": "I just stumbled as well over this. In general the quality of the data set seems not to be optimal to say it politely. For example the \"Gasp,Sigh\" labeled clips. I checked the english dictionary if i maybe misunderstood something.\n\ntrain_noisy/ff97b092.wav \ntrain_noisy/e140e930.wav\nI don't hear a single gasp or sigh in either of those. And i haven't heared much into the data.",
    "515380": "The noisy training data is intended to have unreliable labels. As we tried to explain in the challenge description, we're trying to simulate a real-world scenario where you have a small amount of relatively well-labeled data and a much larger amount of weakly labeled data.",
    "515413": "Thanks, Manoj. I will try to filter out the ones, that seem to have benefit.\n\nThe following files from \"train_noisy\" are completely silent as well. They are all exaclty 15 seconds long.:\n```\n02f274b2.wav\n08b34136.wav\n1af3bd88.wav\n1fd4f275.wav\n2f503375.wav\n3496256e.wav\n551a4b3b.wav\n5a5761c9.wav\n6d062e59.wav\n769d131d.wav\n8c712129.wav\n988cf8f2.wav\n9f4fa2df.wav\nb1d2590c.wav\nbe273a3c.wav\nd527dcf0.wav\ne4faa2e1.wav\nfa659a71.wav\nfba392d8.wav\n```",
    "522607": "i hear no sound , also in plot all signal is zero !",
    "529449": "I confirm that the clip 1d44b0bd.wav contained a (rather scary) perfectly audible *Whispering* in its original format. But in the version uploaded to Kaggle the file is indeed corrupted (empty). My guess is this happened during format conversion to 44.1 kHz @16bit without ffmpeg complaining. \n\nthanks for letting us know. Will double check format conversions in the future!"
  },
  "source": "meta"
}