{
  "id": 49712,
  "title": "Any good solutions to the special TensorFlow competition?",
  "url": "/competitions/tensorflow-speech-recognition-challenge/discussion/49712",
  "author_name": "",
  "post_date": "2018-02-14T15:30:31.589135300Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi</p>\n\n<p>I have seen a lot of topics on people sharing their solution for the main competition and it have helped me so much. However Im trying to build a model that could run on a small MCU so it would be super useful if anyone had any tips on how they solved that part of the competition. </p>\n\n<p>Thanks!</p>",
  "messages": [
    {
      "id": "282782",
      "postDate": "02/14/2018 15:30:31",
      "content": "<p>Hi</p>\n\n<p>I have seen a lot of topics on people sharing their solution for the main competition and it have helped me so much. However Im trying to build a model that could run on a small MCU so it would be super useful if anyone had any tips on how they solved that part of the competition. </p>\n\n<p>Thanks!</p>",
      "rawMarkdown": "Hi\n\nI have seen a lot of topics on people sharing their solution for the main competition and it have helped me so much. However Im trying to build a model that could run on a small MCU so it would be super useful if anyone had any tips on how they solved that part of the competition. \n\nThanks!",
      "votes": null
    },
    {
      "id": "282900",
      "postDate": "02/14/2018 19:01:05",
      "content": "<p>Hi Elias,</p>\n\n<p>I am part of the team that won the special price. If you just want to predict the 12 classes you can try our <a href=\"https://github.com/see--/speech_recognition/blob/master/tf_files/frozen_195.pb\">frozen graph</a> and you could use <a href=\"https://github.com/see--/speech_recognition/blob/master/make_submission_on_rpi.py\">this script</a> to run the graph. The model gets a score of 0.908 on the private leaderboard. The training is a bit clumsy as it requires hard targets from our ensemble model (this video gives some good insights <a href=\"https://www.youtube.com/watch?v=EK61htlw8hY\">https://www.youtube.com/watch?v=EK61htlw8hY</a>). The model targets the raspberry pi 3 and I don't know anything about microcontrollers but there was a nice discussion and paper on this topic: <a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/45037\">https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/45037</a>.</p>",
      "rawMarkdown": "Hi Elias,\n\nI am part of the team that won the special price. If you just want to predict the 12 classes you can try our [frozen graph](https://github.com/see--/speech_recognition/blob/master/tf_files/frozen_195.pb) and you could use [this script](https://github.com/see--/speech_recognition/blob/master/make_submission_on_rpi.py) to run the graph. The model gets a score of 0.908 on the private leaderboard. The training is a bit clumsy as it requires hard targets from our ensemble model (this video gives some good insights https://www.youtube.com/watch?v=EK61htlw8hY). The model targets the raspberry pi 3 and I don't know anything about microcontrollers but there was a nice discussion and paper on this topic: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/45037.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 282900,
      "author_name": "seesee",
      "author_url": "",
      "post_date": "02/14/2018 19:01:05",
      "content": "<p>Hi Elias,</p>\n\n<p>I am part of the team that won the special price. If you just want to predict the 12 classes you can try our <a href=\"https://github.com/see--/speech_recognition/blob/master/tf_files/frozen_195.pb\">frozen graph</a> and you could use <a href=\"https://github.com/see--/speech_recognition/blob/master/make_submission_on_rpi.py\">this script</a> to run the graph. The model gets a score of 0.908 on the private leaderboard. The training is a bit clumsy as it requires hard targets from our ensemble model (this video gives some good insights <a href=\"https://www.youtube.com/watch?v=EK61htlw8hY\">https://www.youtube.com/watch?v=EK61htlw8hY</a>). The model targets the raspberry pi 3 and I don't know anything about microcontrollers but there was a nice discussion and paper on this topic: <a href=\"https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/45037\">https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/45037</a>.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "282782": "Hi\n\nI have seen a lot of topics on people sharing their solution for the main competition and it have helped me so much. However Im trying to build a model that could run on a small MCU so it would be super useful if anyone had any tips on how they solved that part of the competition. \n\nThanks!",
    "282900": "Hi Elias,\n\nI am part of the team that won the special price. If you just want to predict the 12 classes you can try our [frozen graph](https://github.com/see--/speech_recognition/blob/master/tf_files/frozen_195.pb) and you could use [this script](https://github.com/see--/speech_recognition/blob/master/make_submission_on_rpi.py) to run the graph. The model gets a score of 0.908 on the private leaderboard. The training is a bit clumsy as it requires hard targets from our ensemble model (this video gives some good insights https://www.youtube.com/watch?v=EK61htlw8hY). The model targets the raspberry pi 3 and I don't know anything about microcontrollers but there was a nice discussion and paper on this topic: https://www.kaggle.com/c/tensorflow-speech-recognition-challenge/discussion/45037."
  },
  "source": "meta"
}