well, my first instinct is to take Mel spectrogram of the 5 sec audio and run the image classification on the spectrogram. is this a good approach? any suggestions, I am just getting started.
", "messages": [ { "id": "2796176", "postDate": "05/06/2024 06:29:10", "content": "well, my first instinct is to take Mel spectrogram of the 5 sec audio and run the image classification on the spectrogram. is this a good approach? any suggestions, I am just getting started.
", "rawMarkdown": "well, my first instinct is to take Mel spectrogram of the 5 sec audio and run the image classification on the spectrogram. is this a good approach? any suggestions, I am just getting started.", "votes": null }, { "id": "2796469", "postDate": "05/06/2024 08:41:25", "content": "Yes, previous top solutions have a similar pipeline, audio -> mel spectrogram -> multi-labels image classification.
", "rawMarkdown": "Yes, previous top solutions have a similar pipeline, audio -> mel spectrogram -> multi-labels image classification.", "votes": null }, { "id": "2796519", "postDate": "05/06/2024 09:28:16", "content": "Mel spectrogram has been used a lot in previous birdsong competitions.
\nIf you want to try other methods then you can find various ways to transform waveforms into 2D images in the HMS solutions: https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/leaderboard
\nPeople also used 1D models there.
", "rawMarkdown": "Mel spectrogram has been used a lot in previous birdsong competitions.\n\nIf you want to try other methods then you can find various ways to transform waveforms into 2D images in the HMS solutions: https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/leaderboard\n\nPeople also used 1D models there.", "votes": null }, { "id": "2797817", "postDate": "05/07/2024 00:48:30", "content": "Mel spectrograms seem to be the dominant approach in this competition historically. The mel scale is approximately logarithmic, and in general I'd expect spectrograms that use a log scale for frequency would work better than a linear scale, since we humans tend to sense logarithmically, and birds probably do too.
", "rawMarkdown": "Mel spectrograms seem to be the dominant approach in this competition historically. The mel scale is approximately logarithmic, and in general I'd expect spectrograms that use a log scale for frequency would work better than a linear scale, since we humans [tend to sense logarithmically](https://rss.onlinelibrary.wiley.com/doi/full/10.1111/j.1740-9713.2013.00636.x), and birds probably do too.", "votes": null } ], "comments": [ { "id": 2796469, "author_name": "zz1zzi", "author_url": "", "post_date": "05/06/2024 08:41:25", "content": "Yes, previous top solutions have a similar pipeline, audio -> mel spectrogram -> multi-labels image classification.
", "votes": null, "replies": [] }, { "id": 2796519, "author_name": "cpmpml", "author_url": "", "post_date": "05/06/2024 09:28:16", "content": "Mel spectrogram has been used a lot in previous birdsong competitions.
\nIf you want to try other methods then you can find various ways to transform waveforms into 2D images in the HMS solutions: https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/leaderboard
\nPeople also used 1D models there.
", "votes": null, "replies": [] }, { "id": 2797817, "author_name": "robbynevels", "author_url": "", "post_date": "05/07/2024 00:48:30", "content": "Mel spectrograms seem to be the dominant approach in this competition historically. The mel scale is approximately logarithmic, and in general I'd expect spectrograms that use a log scale for frequency would work better than a linear scale, since we humans tend to sense logarithmically, and birds probably do too.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2796176": "well, my first instinct is to take Mel spectrogram of the 5 sec audio and run the image classification on the spectrogram. is this a good approach? any suggestions, I am just getting started.", "2796469": "Yes, previous top solutions have a similar pipeline, audio -> mel spectrogram -> multi-labels image classification.", "2796519": "Mel spectrogram has been used a lot in previous birdsong competitions.\n\nIf you want to try other methods then you can find various ways to transform waveforms into 2D images in the HMS solutions: https://www.kaggle.com/competitions/hms-harmful-brain-activity-classification/leaderboard\n\nPeople also used 1D models there.", "2797817": "Mel spectrograms seem to be the dominant approach in this competition historically. The mel scale is approximately logarithmic, and in general I'd expect spectrograms that use a log scale for frequency would work better than a linear scale, since we humans [tend to sense logarithmically](https://rss.onlinelibrary.wiley.com/doi/full/10.1111/j.1740-9713.2013.00636.x), and birds probably do too." }, "source": "meta" }