{
  "id": 76102,
  "title": "Analogy to Speech Recognition",
  "url": "/competitions/vsb-power-line-fault-detection/discussion/76102",
  "author_name": "",
  "post_date": "2018-12-29T10:40:07.983429300Z",
  "votes": 15,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Looking at the data I see strong analogies to audio signals and hence recommend looking at the previous speech recognition challenge </p>\n\n<p><a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/\">https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/</a> </p>\n\n<p>You might be interested in striding through the forum and pick up a thing or two how to use 1D and 2D Convolutions to solve the problem. </p>",
  "messages": [
    {
      "id": "447195",
      "postDate": "12/29/2018 10:40:07",
      "content": "<p>Looking at the data I see strong analogies to audio signals and hence recommend looking at the previous speech recognition challenge </p>\n\n<p><a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/\">https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/</a> </p>\n\n<p>You might be interested in striding through the forum and pick up a thing or two how to use 1D and 2D Convolutions to solve the problem. </p>",
      "rawMarkdown": "Looking at the data I see strong analogies to audio signals and hence recommend looking at the previous speech recognition challenge \n\n[https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/][1] \n\nYou might be interested in striding through the forum and pick up a thing or two how to use 1D and 2D Convolutions to solve the problem. \n\n\n  [1]: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/",
      "votes": null
    },
    {
      "id": "450765",
      "postDate": "01/05/2019 17:46:20",
      "content": "<p>The analogy is strong indeed, one should be careful with feature and model selection though.</p>\n\n<p>For instance, such signal representations as mel spectrograms or mfcc features were designed specifically for speech processing (primarily for speech recognition), thereby they throw away some irrelevant information (for example, by removing non-informative frequencies from the spectra). However, that sort of information might still be important for the fault detection problem. </p>",
      "rawMarkdown": "The analogy is strong indeed, one should be careful with feature and model selection though.\n\nFor instance, such signal representations as mel spectrograms or mfcc features were designed specifically for speech processing (primarily for speech recognition), thereby they throw away some irrelevant information (for example, by removing non-informative frequencies from the spectra). However, that sort of information might still be important for the fault detection problem.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 450765,
      "author_name": "ddanevskyi",
      "author_url": "",
      "post_date": "01/05/2019 17:46:20",
      "content": "<p>The analogy is strong indeed, one should be careful with feature and model selection though.</p>\n\n<p>For instance, such signal representations as mel spectrograms or mfcc features were designed specifically for speech processing (primarily for speech recognition), thereby they throw away some irrelevant information (for example, by removing non-informative frequencies from the spectra). However, that sort of information might still be important for the fault detection problem. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "447195": "Looking at the data I see strong analogies to audio signals and hence recommend looking at the previous speech recognition challenge \n\n[https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/][1] \n\nYou might be interested in striding through the forum and pick up a thing or two how to use 1D and 2D Convolutions to solve the problem. \n\n\n  [1]: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/",
    "450765": "The analogy is strong indeed, one should be careful with feature and model selection though.\n\nFor instance, such signal representations as mel spectrograms or mfcc features were designed specifically for speech processing (primarily for speech recognition), thereby they throw away some irrelevant information (for example, by removing non-informative frequencies from the spectra). However, that sort of information might still be important for the fault detection problem."
  },
  "source": "meta"
}