{
  "id": 216661,
  "title": "Does raw audio augmentation work for you?",
  "url": "/competitions/rfcx-species-audio-detection/discussion/216661",
  "author_name": "",
  "post_date": "2021-02-03T15:05:32.127949Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I used AddBackgroundNoise and AddGaussianSNR with max snr of 1.0 and min snr of 0.001, and it worsened the performance. Did I used the wrong range of snr or is it a general trend? Judging by the previous competitions, heavy raw audio augmentations seemed to work pretty well. </p>",
  "messages": [
    {
      "id": "1184525",
      "postDate": "02/03/2021 15:05:32",
      "content": "<p>I used AddBackgroundNoise and AddGaussianSNR with max snr of 1.0 and min snr of 0.001, and it worsened the performance. Did I used the wrong range of snr or is it a general trend? Judging by the previous competitions, heavy raw audio augmentations seemed to work pretty well. </p>",
      "rawMarkdown": "I used AddBackgroundNoise and AddGaussianSNR with max snr of 1.0 and min snr of 0.001, and it worsened the performance. Did I used the wrong range of snr or is it a general trend? Judging by the previous competitions, heavy raw audio augmentations seemed to work pretty well.",
      "votes": null
    },
    {
      "id": "1195613",
      "postDate": "02/11/2021 00:29:44",
      "content": "<p>I'm new to these competitions, but I do have a working knowledge of audio processing.</p>\n<p>What an SNR of only 1.0?  This means the level of the target signal is at the same level as the noise.</p>\n<p>Did you try SNR values larger than 1.0?</p>",
      "rawMarkdown": "I'm new to these competitions, but I do have a working knowledge of audio processing.\n\nWhat an SNR of only 1.0?  This means the level of the target signal is at the same level as the noise.\n\nDid you try SNR values larger than 1.0?",
      "votes": null
    },
    {
      "id": "1197692",
      "postDate": "02/12/2021 10:33:56",
      "content": "<p>In my experiments (SED) I've seen improvements on my score using (audiomentations library):</p>\n<ul>\n<li>GaussianNoise(p=0.5),</li>\n<li>GaussianSNR(p=0.5),</li>\n<li>Shift(p=0.5),</li>\n<li>Gain(p=0.5)</li>\n</ul>\n<p>Standard parameters so yeah, max SNR = 1.0, min SNR = 0.001<br>\nNo luck with other augmentations but I haven't tried the background noise</p>",
      "rawMarkdown": "In my experiments (SED) I've seen improvements on my score using (audiomentations library):\n- GaussianNoise(p=0.5),\n- GaussianSNR(p=0.5),\n- Shift(p=0.5),\n- Gain(p=0.5)\n\nStandard parameters so yeah, max SNR = 1.0, min SNR = 0.001\nNo luck with other augmentations but I haven't tried the background noise",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1195613,
      "author_name": "michaelrw",
      "author_url": "",
      "post_date": "02/11/2021 00:29:44",
      "content": "<p>I'm new to these competitions, but I do have a working knowledge of audio processing.</p>\n<p>What an SNR of only 1.0?  This means the level of the target signal is at the same level as the noise.</p>\n<p>Did you try SNR values larger than 1.0?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1197692,
      "author_name": "framoni",
      "author_url": "",
      "post_date": "02/12/2021 10:33:56",
      "content": "<p>In my experiments (SED) I've seen improvements on my score using (audiomentations library):</p>\n<ul>\n<li>GaussianNoise(p=0.5),</li>\n<li>GaussianSNR(p=0.5),</li>\n<li>Shift(p=0.5),</li>\n<li>Gain(p=0.5)</li>\n</ul>\n<p>Standard parameters so yeah, max SNR = 1.0, min SNR = 0.001<br>\nNo luck with other augmentations but I haven't tried the background noise</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1184525": "I used AddBackgroundNoise and AddGaussianSNR with max snr of 1.0 and min snr of 0.001, and it worsened the performance. Did I used the wrong range of snr or is it a general trend? Judging by the previous competitions, heavy raw audio augmentations seemed to work pretty well.",
    "1195613": "I'm new to these competitions, but I do have a working knowledge of audio processing.\n\nWhat an SNR of only 1.0?  This means the level of the target signal is at the same level as the noise.\n\nDid you try SNR values larger than 1.0?",
    "1197692": "In my experiments (SED) I've seen improvements on my score using (audiomentations library):\n- GaussianNoise(p=0.5),\n- GaussianSNR(p=0.5),\n- Shift(p=0.5),\n- Gain(p=0.5)\n\nStandard parameters so yeah, max SNR = 1.0, min SNR = 0.001\nNo luck with other augmentations but I haven't tried the background noise"
  },
  "source": "meta"
}