{
  "id": 313271,
  "title": "torchaudio load the resample dataset 16K .wav in < 2ms 😮 ",
  "url": "/competitions/kaggle-pog-series-s01e02/discussion/313271",
  "author_name": "",
  "post_date": "2022-03-16T10:13:12.426932Z",
  "votes": 3,
  "comment_count": 2,
  "views": 0,
  "content": "<p>After improving loading audio by using torchaudio instead of librosa (<a href=\"https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312747)\" target=\"_blank\">https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312747)</a>, I still find it is very slow (it took me 20min to finish an epoch.</p>\n<p>Then I tried the Resampled 16K Dataset <a href=\"https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312458\" target=\"_blank\">https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312458</a> from <a href=\"https://www.kaggle.com/harveenchadha\" target=\"_blank\">@harveenchadha</a> and it took &lt; 2ms for each audio 😮 </p>\n<p>%%timeit<br>\ntorchaudio.load('../input/kaggle-pog-series-s01e02/train/011395.ogg')<br>\n65.8 ms ± 10.6 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)</p>\n<p>%%**timeit<br>\ntorchaudio.load('../input/pogmusicclassification/resampled_train/011395_16k.wav')<br>\n1.17 ms ± 81.7 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)</p>\n<p>Cheers !</p>",
  "messages": [
    {
      "id": "1724548",
      "postDate": "03/16/2022 10:13:12",
      "content": "<p>After improving loading audio by using torchaudio instead of librosa (<a href=\"https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312747)\" target=\"_blank\">https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312747)</a>, I still find it is very slow (it took me 20min to finish an epoch.</p>\n<p>Then I tried the Resampled 16K Dataset <a href=\"https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312458\" target=\"_blank\">https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312458</a> from <a href=\"https://www.kaggle.com/harveenchadha\" target=\"_blank\">@harveenchadha</a> and it took &lt; 2ms for each audio 😮 </p>\n<p>%%timeit<br>\ntorchaudio.load('../input/kaggle-pog-series-s01e02/train/011395.ogg')<br>\n65.8 ms ± 10.6 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)</p>\n<p>%%**timeit<br>\ntorchaudio.load('../input/pogmusicclassification/resampled_train/011395_16k.wav')<br>\n1.17 ms ± 81.7 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)</p>\n<p>Cheers !</p>",
      "rawMarkdown": "After improving loading audio by using torchaudio instead of librosa (https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312747), I still find it is very slow (it took me 20min to finish an epoch.\n\nThen I tried the Resampled 16K Dataset https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312458 from @harveenchadha and it took < 2ms for each audio 😮 \n\n%%timeit\ntorchaudio.load('../input/kaggle-pog-series-s01e02/train/011395.ogg')\n65.8 ms ± 10.6 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)\n\n%%**timeit\ntorchaudio.load('../input/pogmusicclassification/resampled_train/011395_16k.wav')\n1.17 ms ± 81.7 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)\n\nCheers !",
      "votes": null
    },
    {
      "id": "1724555",
      "postDate": "03/16/2022 10:21:15",
      "content": "<p><a href=\"https://www.kaggle.com/robikscube\" target=\"_blank\">@robikscube</a> I hope it doesn't violate the competition rules if I use this dataset :D</p>",
      "rawMarkdown": "robikscube I hope it doesn't violate the competition rules if I use this dataset :D",
      "votes": null
    },
    {
      "id": "1724815",
      "postDate": "03/16/2022 14:34:01",
      "content": "<p>No, this is just a processed version of the same dataset. Unless it includes additional audio data, it's fine to use.</p>",
      "rawMarkdown": "No, this is just a processed version of the same dataset. Unless it includes additional audio data, it's fine to use.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1724555,
      "author_name": "dienhoa",
      "author_url": "",
      "post_date": "03/16/2022 10:21:15",
      "content": "<p><a href=\"https://www.kaggle.com/robikscube\" target=\"_blank\">@robikscube</a> I hope it doesn't violate the competition rules if I use this dataset :D</p>",
      "votes": null,
      "replies": [
        {
          "id": 1724815,
          "author_name": "robikscube",
          "author_url": "",
          "post_date": "03/16/2022 14:34:01",
          "content": "<p>No, this is just a processed version of the same dataset. Unless it includes additional audio data, it's fine to use.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1724548": "After improving loading audio by using torchaudio instead of librosa (https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312747), I still find it is very slow (it took me 20min to finish an epoch.\n\nThen I tried the Resampled 16K Dataset https://www.kaggle.com/c/kaggle-pog-series-s01e02/discussion/312458 from @harveenchadha and it took < 2ms for each audio 😮 \n\n%%timeit\ntorchaudio.load('../input/kaggle-pog-series-s01e02/train/011395.ogg')\n65.8 ms ± 10.6 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)\n\n%%**timeit\ntorchaudio.load('../input/pogmusicclassification/resampled_train/011395_16k.wav')\n1.17 ms ± 81.7 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)\n\nCheers !",
    "1724555": "robikscube I hope it doesn't violate the competition rules if I use this dataset :D",
    "1724815": "No, this is just a processed version of the same dataset. Unless it includes additional audio data, it's fine to use."
  },
  "source": "meta"
}