{
  "id": 491225,
  "title": "Mel Spectrogram Notebook / Dataset",
  "url": "/competitions/birdclef-2024/discussion/491225",
  "author_name": "Rich Olson",
  "post_date": "2024-04-05T04:13:00.460000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hey all -</p>\n<p>Just made my very simple Mel Spectrogram Generator notebook public:<br>\n<a href=\"https://www.kaggle.com/code/richolson/birdclef2024-simple-mel-spectrogram-generator/\" target=\"_blank\">https://www.kaggle.com/code/richolson/birdclef2024-simple-mel-spectrogram-generator/</a></p>\n<p>There is also a \"small\" dataset here (might be enough to get an ImageNet training on):<br>\n<a href=\"https://www.kaggle.com/datasets/richolson/birdclef-2024-mel-spectrograms\" target=\"_blank\">https://www.kaggle.com/datasets/richolson/birdclef-2024-mel-spectrograms</a><br>\n(about 14,522 224x224 images - took about 45 minutes to generate)</p>\n<p>The notebook takes each audio file - and converts it into 1 or more PNG files containing a Mel Spectrogram.</p>\n<p>It's intended to be very configurable… (dataset was created from these parameters)</p>\n<pre><code>\n\naudio_duration = \n\n\n\nmax_clips_per_audio_file = \n\n\nimage_size = \n\n\nmax_audio_input_per_species = \n\n\n\nstart_at_folder_offset = \n</code></pre>\n<p>Could easily use this to generate a larger dataset.</p>\n<p>-Rich</p>",
  "messages": [
    {
      "id": 2736162,
      "postDate": "2024-04-05T04:13:00.460Z",
      "content": "<p>Hey all -</p>\n<p>Just made my very simple Mel Spectrogram Generator notebook public:<br>\n<a href=\"https://www.kaggle.com/code/richolson/birdclef2024-simple-mel-spectrogram-generator/\" target=\"_blank\">https://www.kaggle.com/code/richolson/birdclef2024-simple-mel-spectrogram-generator/</a></p>\n<p>There is also a \"small\" dataset here (might be enough to get an ImageNet training on):<br>\n<a href=\"https://www.kaggle.com/datasets/richolson/birdclef-2024-mel-spectrograms\" target=\"_blank\">https://www.kaggle.com/datasets/richolson/birdclef-2024-mel-spectrograms</a><br>\n(about 14,522 224x224 images - took about 45 minutes to generate)</p>\n<p>The notebook takes each audio file - and converts it into 1 or more PNG files containing a Mel Spectrogram.</p>\n<p>It's intended to be very configurable… (dataset was created from these parameters)</p>\n<pre><code>\n\naudio_duration = \n\n\n\nmax_clips_per_audio_file = \n\n\nimage_size = \n\n\nmax_audio_input_per_species = \n\n\n\nstart_at_folder_offset = \n</code></pre>\n<p>Could easily use this to generate a larger dataset.</p>\n<p>-Rich</p>",
      "rawMarkdown": "Hey all -\n\nJust made my very simple Mel Spectrogram Generator notebook public:\nhttps://www.kaggle.com/code/richolson/birdclef2024-simple-mel-spectrogram-generator/\n\nThere is also a \"small\" dataset here (might be enough to get an ImageNet training on):\nhttps://www.kaggle.com/datasets/richolson/birdclef-2024-mel-spectrograms\n(about 14,522 224x224 images - took about 45 minutes to generate)\n\nThe notebook takes each audio file - and converts it into 1 or more PNG files containing a Mel Spectrogram.\n\nIt's intended to be very configurable... (dataset was created from these parameters)\n\n```python\n#generates spectrograms are generated on audio clips this long\n#audio clips shorter than this are ignored\naudio_duration = 5.0\n\n#generates up to this many images per file (multiples of audio_duration)\n#remainders of audio clip length / audio_duration are ignored\nmax_clips_per_audio_file = 3\n\n#image size (image_size x image_size)\nimage_size = 224\n\n#processes up to this many audio files per species\nmax_audio_input_per_species = 50\n\n#setting this above 0 will cause processing to start at a folder after the first\n#useful for restarting / continuing in case of running out of time / crash (folders are processed alphabetically)\nstart_at_folder_offset = 0\n\n```\nCould easily use this to generate a larger dataset.\n\n-Rich",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2736162": "Hey all -\n\nJust made my very simple Mel Spectrogram Generator notebook public:\nhttps://www.kaggle.com/code/richolson/birdclef2024-simple-mel-spectrogram-generator/\n\nThere is also a \"small\" dataset here (might be enough to get an ImageNet training on):\nhttps://www.kaggle.com/datasets/richolson/birdclef-2024-mel-spectrograms\n(about 14,522 224x224 images - took about 45 minutes to generate)\n\nThe notebook takes each audio file - and converts it into 1 or more PNG files containing a Mel Spectrogram.\n\nIt's intended to be very configurable... (dataset was created from these parameters)\n\n```python\n#generates spectrograms are generated on audio clips this long\n#audio clips shorter than this are ignored\naudio_duration = 5.0\n\n#generates up to this many images per file (multiples of audio_duration)\n#remainders of audio clip length / audio_duration are ignored\nmax_clips_per_audio_file = 3\n\n#image size (image_size x image_size)\nimage_size = 224\n\n#processes up to this many audio files per species\nmax_audio_input_per_species = 50\n\n#setting this above 0 will cause processing to start at a folder after the first\n#useful for restarting / continuing in case of running out of time / crash (folders are processed alphabetically)\nstart_at_folder_offset = 0\n\n```\nCould easily use this to generate a larger dataset.\n\n-Rich"
  }
}