I was going through some of the discussions and fantastic notebooks here, like:
\nhttps://www.kaggle.com/andradaolteanu/g2net-searching-the-sky-pytorch-effnet-w-meta
\nand https://www.kaggle.com/mistag/data-preprocessing-with-gwpy
From these, I understand that nnAudio's \"cqt1992v2\" method is much faster by comparison. However, I find the CQT image from GWPy's \"q_transform\" method to be more interpretable to visualize a potential event. Is the y-axis of nnAudio's \"cqt1992v2\" returning frequency bin numbers?
\nMy question is, are these two methods returning similar things? If yes, then is there a way to transform nnAudio's \"cqt1992v2\" output to look something similar to that of GWPy's \"q_transform\" (log scale on frequency maybe)?
\nThanks!
I was going through some of the discussions and fantastic notebooks here, like:
\nhttps://www.kaggle.com/andradaolteanu/g2net-searching-the-sky-pytorch-effnet-w-meta
\nand https://www.kaggle.com/mistag/data-preprocessing-with-gwpy
From these, I understand that nnAudio's \"cqt1992v2\" method is much faster by comparison. However, I find the CQT image from GWPy's \"q_transform\" method to be more interpretable to visualize a potential event. Is the y-axis of nnAudio's \"cqt1992v2\" returning frequency bin numbers?
\nMy question is, are these two methods returning similar things? If yes, then is there a way to transform nnAudio's \"cqt1992v2\" output to look something similar to that of GWPy's \"q_transform\" (log scale on frequency maybe)?
\nThanks!
Actually, nnAudio's CQT() doesn't support frequency linear scaling.
", "rawMarkdown": "~~nnAudio's CQT() doesn't support frequency log scaling - you'll have to modify the original code.~~ Actually, nnAudio's CQT() doesn't support frequency linear scaling.", "votes": null }, { "id": "1490620", "postDate": "08/25/2021 18:21:03", "content": "Thanks for your reply. So is log scaling of frequency axis the the only difference between the two?
", "rawMarkdown": "Thanks for your reply. So is log scaling of frequency axis the the only difference between the two?", "votes": null }, { "id": "1490669", "postDate": "08/25/2021 19:08:57", "content": "\n\nnnAudio's CQT() doesn't support frequency log scaling
\n
It's the opposite, it does not support linear frequency scale. It uses a log scale.
", "rawMarkdown": "> nnAudio's CQT() doesn't support frequency log scaling\n\nIt's the opposite, it does not support linear frequency scale. It uses a log scale.", "votes": null }, { "id": "1490730", "postDate": "08/25/2021 20:13:14", "content": "I've checked the code, and you are right. Thanks for pointing out my mistake.
", "rawMarkdown": "I've checked the code, and you are right. Thanks for pointing out my mistake.", "votes": null }, { "id": "1491799", "postDate": "08/26/2021 16:25:37", "content": "Thank you. Unfortunately, I am unable to upload images here to point out my doubts.
\nSo using nnAudio's \"cqt1992v2\" function, I get a spectogram that seems to have features on the extremes of the time axis. Is there some kind of reflection involved?
FYI. nnAudio doens't support in TPU :(
", "rawMarkdown": "FYI. nnAudio doens't support in TPU :(", "votes": null }, { "id": "1494966", "postDate": "08/29/2021 07:49:15", "content": "Regarding the images upload (which is very annoying but I guess Kaggle is working on an improvement), you can use this service: https://imgbb.com/. Very easy to use and it gives you a link to your image.
", "rawMarkdown": "Regarding the images upload (which is very annoying but I guess Kaggle is working on an improvement), you can use this service: https://imgbb.com/. Very easy to use and it gives you a link to your image.", "votes": null }, { "id": "1494983", "postDate": "08/29/2021 07:59:16", "content": "Regarding the visual aspect, I think I see what you mean. Indeed, it seems there is some wrapping around the end of the plot for nnAudio's CQT function. I will explore the function's output further and post what I find.
\nBy the way, notice that the links you have provided aren't correctly formatted: when clicking on them they lead to the main competition page.
", "rawMarkdown": "Regarding the visual aspect, I think I see what you mean. Indeed, it seems there is some wrapping around the end of the plot for nnAudio's CQT function. I will explore the function's output further and post what I find.\n\nBy the way, notice that the links you have provided aren't correctly formatted: when clicking on them they lead to the main competition page.", "votes": null }, { "id": "1495079", "postDate": "08/29/2021 09:43:01", "content": "Thanks for confirming the suspected wrapping. Also, fixed the URLs to the notebooks.
", "rawMarkdown": "Thanks for confirming the suspected wrapping. Also, fixed the URLs to the notebooks.", "votes": null }, { "id": "1495391", "postDate": "08/29/2021 14:27:57", "content": "I also tried to use nnAudio with PyTorch XLA, but it doesn't work for me as well.
", "rawMarkdown": "I also tried to use nnAudio with PyTorch XLA, but it doesn't work for me as well.", "votes": null }, { "id": "1495432", "postDate": "08/29/2021 14:56:35", "content": "Awesome! yes, the two outputs aren't the same for sure so will need some investigation. 👀
", "rawMarkdown": "Awesome! yes, the two outputs aren't the same for sure so will need some investigation. 👀", "votes": null }, { "id": "1561066", "postDate": "10/27/2021 09:53:56", "content": "Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris
Actually, nnAudio's CQT() doesn't support frequency linear scaling.
", "votes": null, "replies": [ { "id": 1490620, "author_name": "sbpdev", "author_url": "", "post_date": "08/25/2021 18:21:03", "content": "Thanks for your reply. So is log scaling of frequency axis the the only difference between the two?
", "votes": null, "replies": [] }, { "id": 1490669, "author_name": "cpmpml", "author_url": "", "post_date": "08/25/2021 19:08:57", "content": "\n\nnnAudio's CQT() doesn't support frequency log scaling
\n
It's the opposite, it does not support linear frequency scale. It uses a log scale.
", "votes": null, "replies": [] }, { "id": 1490730, "author_name": "atamazian", "author_url": "", "post_date": "08/25/2021 20:13:14", "content": "I've checked the code, and you are right. Thanks for pointing out my mistake.
", "votes": null, "replies": [] }, { "id": 1491799, "author_name": "sbpdev", "author_url": "", "post_date": "08/26/2021 16:25:37", "content": "Thank you. Unfortunately, I am unable to upload images here to point out my doubts.
\nSo using nnAudio's \"cqt1992v2\" function, I get a spectogram that seems to have features on the extremes of the time axis. Is there some kind of reflection involved?
Regarding the images upload (which is very annoying but I guess Kaggle is working on an improvement), you can use this service: https://imgbb.com/. Very easy to use and it gives you a link to your image.
", "votes": null, "replies": [] }, { "id": 1494983, "author_name": "yassinealouini", "author_url": "", "post_date": "08/29/2021 07:59:16", "content": "Regarding the visual aspect, I think I see what you mean. Indeed, it seems there is some wrapping around the end of the plot for nnAudio's CQT function. I will explore the function's output further and post what I find.
\nBy the way, notice that the links you have provided aren't correctly formatted: when clicking on them they lead to the main competition page.
", "votes": null, "replies": [] }, { "id": 1495079, "author_name": "sbpdev", "author_url": "", "post_date": "08/29/2021 09:43:01", "content": "Thanks for confirming the suspected wrapping. Also, fixed the URLs to the notebooks.
", "votes": null, "replies": [] }, { "id": 1495432, "author_name": "yassinealouini", "author_url": "", "post_date": "08/29/2021 14:56:35", "content": "Awesome! yes, the two outputs aren't the same for sure so will need some investigation. 👀
", "votes": null, "replies": [] } ] }, { "id": 1494851, "author_name": "leemop", "author_url": "", "post_date": "08/29/2021 06:06:09", "content": "FYI. nnAudio doens't support in TPU :(
", "votes": null, "replies": [ { "id": 1495391, "author_name": "atamazian", "author_url": "", "post_date": "08/29/2021 14:27:57", "content": "I also tried to use nnAudio with PyTorch XLA, but it doesn't work for me as well.
", "votes": null, "replies": [] } ] }, { "id": 1561066, "author_name": "zerafachris", "author_url": "", "post_date": "10/27/2021 09:53:56", "content": "Hey All,
\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.
\nThank you all,
\nRegards,
\nChris