What about this object? There is the sound of footsteps in it but not in label.
", "messages": [ { "id": 535914, "postDate": "2019-05-23T16:11:02.437Z", "content": "The labels are not guaranteed to be exhaustive or 100% correct. All you can assume is that there is a good chance that the labels we provide exist somewhere in the clip. It is possible that one or more of the labels we provide are wrong (because of human error), and it is certainly possible that there are sound events in the clip that we have not labeled (and these events could be in our list of 80 classes or not).
", "rawMarkdown": "The labels are not guaranteed to be exhaustive or 100% correct. All you can assume is that there is a good chance that the labels we provide exist somewhere in the clip. It is possible that one or more of the labels we provide are wrong (because of human error), and it is certainly possible that there are sound events in the clip that we have not labeled (and these events could be in our list of 80 classes or not). ", "votes": 1 }, { "id": 534698, "postDate": "2019-05-21T17:56:16.470Z", "content": "What about this object? There is the sound of footsteps in it but not in label.
", "rawMarkdown": "What about this object? There is the sound of footsteps in it but not in label.", "votes": 2 }, { "id": 536409, "postDate": "2019-05-24T12:12:51.277Z", "content": "Hi m1ckyro5a,
\n\nYou just found a behaviour that sometimes occurs in the curated set. The clip 900e6ee9.wav is labeled with \"Keys_jangling\", and the clip contains such an event, hence the label was curated properly.
\n\nHowever, as Manoj says, it can happen that a few of these audio clips present additional acoustic material beyond the provided ground truth label(s). This is explained in the Data section, where we describe the Curated Train Set\nhttps://www.kaggle.com/c/freesound-audio-tagging-2019/data
\n\nThis particular case means that the annotation \"Keys_jangling\" is correct but incomplete, and the missing label (\"Walk, footsteps\") happens to be within the vocabulary of the dataset (ie belongs to another class within the list of 80). \nMore info about label noise types here: https://arxiv.org/abs/1901.01189
\n\nThis phenomenon can happen sometimes in the curated train set, but it happens a lot more in the noisy set. And if I may, this happens in nearly every large audio dataset nowadays.
\n\nhope this helps
", "rawMarkdown": "Hi m1ckyro5a,\n\nYou just found a behaviour that sometimes occurs in the curated set. The clip 900e6ee9.wav is labeled with \"Keys_jangling\", and the clip contains such an event, hence the label was curated properly.\n\nHowever, as Manoj says, it can happen that a few of these audio clips present additional acoustic material beyond the provided ground truth label(s). This is explained in the Data section, where we describe the Curated Train Set\n[https://www.kaggle.com/c/freesound-audio-tagging-2019/data](https://www.kaggle.com/c/freesound-audio-tagging-2019/data)\n\nThis particular case means that the annotation \"Keys_jangling\" is correct but incomplete, and the missing label (\"Walk, footsteps\") happens to be within the vocabulary of the dataset (ie belongs to another class within the list of 80). \nMore info about label noise types here: [https://arxiv.org/abs/1901.01189](https://arxiv.org/abs/1901.01189)\n\nThis phenomenon can happen sometimes in the curated train set, but it happens a lot more in the noisy set. And if I may, this happens in nearly every large audio dataset nowadays.\n\nhope this helps" }, { "id": 535665, "postDate": "2019-05-23T09:59:09.030Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 535914, "author_name": "Manoj Plakal", "author_url": "", "post_date": "2019-05-23T16:11:02.437000", "content": "The labels are not guaranteed to be exhaustive or 100% correct. All you can assume is that there is a good chance that the labels we provide exist somewhere in the clip. It is possible that one or more of the labels we provide are wrong (because of human error), and it is certainly possible that there are sound events in the clip that we have not labeled (and these events could be in our list of 80 classes or not).
", "votes": 1, "replies": [] }, { "id": 536409, "author_name": "Eduardo Fonseca", "author_url": "", "post_date": "2019-05-24T12:12:51.277000", "content": "Hi m1ckyro5a,
\n\nYou just found a behaviour that sometimes occurs in the curated set. The clip 900e6ee9.wav is labeled with \"Keys_jangling\", and the clip contains such an event, hence the label was curated properly.
\n\nHowever, as Manoj says, it can happen that a few of these audio clips present additional acoustic material beyond the provided ground truth label(s). This is explained in the Data section, where we describe the Curated Train Set\nhttps://www.kaggle.com/c/freesound-audio-tagging-2019/data
\n\nThis particular case means that the annotation \"Keys_jangling\" is correct but incomplete, and the missing label (\"Walk, footsteps\") happens to be within the vocabulary of the dataset (ie belongs to another class within the list of 80). \nMore info about label noise types here: https://arxiv.org/abs/1901.01189
\n\nThis phenomenon can happen sometimes in the curated train set, but it happens a lot more in the noisy set. And if I may, this happens in nearly every large audio dataset nowadays.
\n\nhope this helps
", "votes": 0, "replies": [] }, { "id": 535665, "author_name": "", "author_url": "", "post_date": "2019-05-23T09:59:09.030000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "535914": "The labels are not guaranteed to be exhaustive or 100% correct. All you can assume is that there is a good chance that the labels we provide exist somewhere in the clip. It is possible that one or more of the labels we provide are wrong (because of human error), and it is certainly possible that there are sound events in the clip that we have not labeled (and these events could be in our list of 80 classes or not). ", "534698": "What about this object? There is the sound of footsteps in it but not in label.", "536409": "Hi m1ckyro5a,\n\nYou just found a behaviour that sometimes occurs in the curated set. The clip 900e6ee9.wav is labeled with \"Keys_jangling\", and the clip contains such an event, hence the label was curated properly.\n\nHowever, as Manoj says, it can happen that a few of these audio clips present additional acoustic material beyond the provided ground truth label(s). This is explained in the Data section, where we describe the Curated Train Set\n[https://www.kaggle.com/c/freesound-audio-tagging-2019/data](https://www.kaggle.com/c/freesound-audio-tagging-2019/data)\n\nThis particular case means that the annotation \"Keys_jangling\" is correct but incomplete, and the missing label (\"Walk, footsteps\") happens to be within the vocabulary of the dataset (ie belongs to another class within the list of 80). \nMore info about label noise types here: [https://arxiv.org/abs/1901.01189](https://arxiv.org/abs/1901.01189)\n\nThis phenomenon can happen sometimes in the curated train set, but it happens a lot more in the noisy set. And if I may, this happens in nearly every large audio dataset nowadays.\n\nhope this helps", "535665": "" } }