Hey everyone, \nI noticed the mp3 files are of variable bit rate. I thought it could be a good start to resample all the mp3 files and create a separate dataset of mp3s with same bitrate. However, when a lot of files are produced as output in kaggle web browser stops responding. Can anybody help me what can I do to handle that? can i zipp them in the output???
\n\nHere is my notebook that converts all mp3s into 44.1kHz. Audio Conversion [44.1 kHz]
", "messages": [ { "id": "916794", "postDate": "07/06/2020 03:32:23", "content": "Hey everyone, \nI noticed the mp3 files are of variable bit rate. I thought it could be a good start to resample all the mp3 files and create a separate dataset of mp3s with same bitrate. However, when a lot of files are produced as output in kaggle web browser stops responding. Can anybody help me what can I do to handle that? can i zipp them in the output???
\n\nHere is my notebook that converts all mp3s into 44.1kHz. Audio Conversion [44.1 kHz]
", "rawMarkdown": "Hey everyone, \nI noticed the mp3 files are of variable bit rate. I thought it could be a good start to resample all the mp3 files and create a separate dataset of mp3s with same bitrate. However, when a lot of files are produced as output in kaggle web browser stops responding. Can anybody help me what can I do to handle that? can i zipp them in the output???\n\nHere is my notebook that converts all mp3s into 44.1kHz. [Audio Conversion [44.1 kHz]](https://www.kaggle.com/redwankarimsony/audio-conversion-44-1-khz)", "votes": null }, { "id": "916816", "postDate": "07/06/2020 04:07:36", "content": "I think kaggle notebooks have output limit of 5 GB. I don't think it's enough to output all the resampled audio files even if they could be zipped.
", "rawMarkdown": "I think kaggle notebooks have output limit of 5 GB. I don't think it's enough to output all the resampled audio files even if they could be zipped.", "votes": null }, { "id": "917204", "postDate": "07/06/2020 10:14:15", "content": "The test set is sampled at 32000Hz (as said here by one of the organizers) so I would reccomend first of all to start sampling at 32000 instead of 44100Hz, secondly there are some discussions where people have already created new datasets with the resampled data so it might be faster for you to download and start using them instead (see here). Since there is a disc memory limit of 5 GB, if you want to do it by yourself anyway, you need to split the task in different notebooks for example in the first one you do birds from a to d, in the second one from e-h etc. and zip the file either using ! zip -r dest.zip source_folder and delete all the mp3 after the zip (and keep only one zip file), or you can use the python library for zip files (you might need to google the commands but they were pretty simple and ituitive if I remember correctly).
pkl format should be OK
", "rawMarkdown": "pkl format should be OK", "votes": null }, { "id": "917477", "postDate": "07/06/2020 14:42:14", "content": "Thanks a lot @pranavkasela for the solution. i hope I will use the dataset sampled at 32kHz uploaded by Radek Osmulski. And of course at the inference time, data is going to be of 32kHz.. I hope it works.. !
", "rawMarkdown": "Thanks a lot @pranavkasela for the solution. i hope I will use the dataset sampled at 32kHz uploaded by [Radek Osmulski](https://www.kaggle.com/radek1). And of course at the inference time, data is going to be of 32kHz.. I hope it works.. !", "votes": null } ], "comments": [ { "id": 916816, "author_name": "jonykarki", "author_url": "", "post_date": "07/06/2020 04:07:36", "content": "I think kaggle notebooks have output limit of 5 GB. I don't think it's enough to output all the resampled audio files even if they could be zipped.
", "votes": null, "replies": [ { "id": 917442, "author_name": "servietsky", "author_url": "", "post_date": "07/06/2020 14:12:44", "content": "pkl format should be OK
", "votes": null, "replies": [] } ] }, { "id": 917204, "author_name": "pranavkasela", "author_url": "", "post_date": "07/06/2020 10:14:15", "content": "The test set is sampled at 32000Hz (as said here by one of the organizers) so I would reccomend first of all to start sampling at 32000 instead of 44100Hz, secondly there are some discussions where people have already created new datasets with the resampled data so it might be faster for you to download and start using them instead (see here). Since there is a disc memory limit of 5 GB, if you want to do it by yourself anyway, you need to split the task in different notebooks for example in the first one you do birds from a to d, in the second one from e-h etc. and zip the file either using ! zip -r dest.zip source_folder and delete all the mp3 after the zip (and keep only one zip file), or you can use the python library for zip files (you might need to google the commands but they were pretty simple and ituitive if I remember correctly).
Thanks a lot @pranavkasela for the solution. i hope I will use the dataset sampled at 32kHz uploaded by Radek Osmulski. And of course at the inference time, data is going to be of 32kHz.. I hope it works.. !
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "916794": "Hey everyone, \nI noticed the mp3 files are of variable bit rate. I thought it could be a good start to resample all the mp3 files and create a separate dataset of mp3s with same bitrate. However, when a lot of files are produced as output in kaggle web browser stops responding. Can anybody help me what can I do to handle that? can i zipp them in the output???\n\nHere is my notebook that converts all mp3s into 44.1kHz. [Audio Conversion [44.1 kHz]](https://www.kaggle.com/redwankarimsony/audio-conversion-44-1-khz)", "916816": "I think kaggle notebooks have output limit of 5 GB. I don't think it's enough to output all the resampled audio files even if they could be zipped.", "917204": "The test set is sampled at 32000Hz (as said [here](https://www.kaggle.com/c/birdsong-recognition/discussion/159943) by one of the organizers) so I would reccomend first of all to start sampling at 32000 instead of 44100Hz, secondly there are some discussions where people have already created new datasets with the resampled data so it might be faster for you to download and start using them instead (see [here](https://www.kaggle.com/c/birdsong-recognition/discussion/164197)). Since there is a disc memory limit of 5 GB, if you want to do it by yourself anyway, you need to split the task in different notebooks for example in the first one you do birds from a to d, in the second one from e-h etc. and zip the file either using `! zip -r dest.zip source_folder` and delete all the mp3 after the zip (and keep only one zip file), or you can use the python library for zip files (you might need to google the commands but they were pretty simple and ituitive if I remember correctly).", "917442": "pkl format should be OK", "917477": "Thanks a lot @pranavkasela for the solution. i hope I will use the dataset sampled at 32kHz uploaded by [Radek Osmulski](https://www.kaggle.com/radek1). And of course at the inference time, data is going to be of 32kHz.. I hope it works.. !" }, "source": "meta" }