{
  "id": 43632,
  "title": "Other labeled audio sets",
  "url": "/competitions/tensorflow-speech-recognition-challenge/discussion/43632",
  "author_name": "",
  "post_date": "2017-11-17T03:24:14.220840100Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Just a quick question\nWhy is there some other labeled  audio sets exist under train/audio folder?\nI know it already mentioned it has 30 different types but we're only looking for 10 of them as it is said in evaluation instruction. </p>\n\n<p>Maybe it is a dumb question after I hear what it is meant for but for now, I can't think of usage well.</p>",
  "messages": [
    {
      "id": "244861",
      "postDate": "11/17/2017 03:24:14",
      "content": "<p>Just a quick question\nWhy is there some other labeled  audio sets exist under train/audio folder?\nI know it already mentioned it has 30 different types but we're only looking for 10 of them as it is said in evaluation instruction. </p>\n\n<p>Maybe it is a dumb question after I hear what it is meant for but for now, I can't think of usage well.</p>",
      "rawMarkdown": "Just a quick question\nWhy is there some other labeled  audio sets exist under train/audio folder?\nI know it already mentioned it has 30 different types but we're only looking for 10 of them as it is said in evaluation instruction. \n\nMaybe it is a dumb question after I hear what it is meant for but for now, I can't think of usage well.",
      "votes": null
    },
    {
      "id": "244878",
      "postDate": "11/17/2017 04:16:07",
      "content": "<p>That is a good question! The data set is designed to be a broadly useful set of short speech commands, and have the ability to distinguish between somebody saying a word we cared about (one of the actual ten commands) and some other random word. The non-command words are sampled from to produce the \"Unknown\" category, so we can test the model's ability to reject words that shouldn't be recognized.</p>\n\n<p>Does that make sense?</p>",
      "rawMarkdown": "That is a good question! The data set is designed to be a broadly useful set of short speech commands, and have the ability to distinguish between somebody saying a word we cared about (one of the actual ten commands) and some other random word. The non-command words are sampled from to produce the \"Unknown\" category, so we can test the model's ability to reject words that shouldn't be recognized.\n\nDoes that make sense?",
      "votes": null
    },
    {
      "id": "244884",
      "postDate": "11/17/2017 04:38:37",
      "content": "<p>Ah now I understand. That sounds very clear to me. Thanks!</p>",
      "rawMarkdown": "Ah now I understand. That sounds very clear to me. Thanks!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 244878,
      "author_name": "petewarden",
      "author_url": "",
      "post_date": "11/17/2017 04:16:07",
      "content": "<p>That is a good question! The data set is designed to be a broadly useful set of short speech commands, and have the ability to distinguish between somebody saying a word we cared about (one of the actual ten commands) and some other random word. The non-command words are sampled from to produce the \"Unknown\" category, so we can test the model's ability to reject words that shouldn't be recognized.</p>\n\n<p>Does that make sense?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 244884,
      "author_name": "gilgarad",
      "author_url": "",
      "post_date": "11/17/2017 04:38:37",
      "content": "<p>Ah now I understand. That sounds very clear to me. Thanks!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "244861": "Just a quick question\nWhy is there some other labeled  audio sets exist under train/audio folder?\nI know it already mentioned it has 30 different types but we're only looking for 10 of them as it is said in evaluation instruction. \n\nMaybe it is a dumb question after I hear what it is meant for but for now, I can't think of usage well.",
    "244878": "That is a good question! The data set is designed to be a broadly useful set of short speech commands, and have the ability to distinguish between somebody saying a word we cared about (one of the actual ten commands) and some other random word. The non-command words are sampled from to produce the \"Unknown\" category, so we can test the model's ability to reject words that shouldn't be recognized.\n\nDoes that make sense?",
    "244884": "Ah now I understand. That sounds very clear to me. Thanks!"
  },
  "source": "meta"
}