https://ai.googleblog.com/2022/01/separating-birdsong-in-wild-for.html
\nSeparating Birdsong in the Wild for Classification
https://ai.googleblog.com/2022/01/separating-birdsong-in-wild-for.html
\nSeparating Birdsong in the Wild for Classification
And in case you haven't noticed, the lead author of this publication is our competition host @tomdenton :)
", "rawMarkdown": "And in case you haven't noticed, the lead author of this publication is our competition host @tomdenton :)", "votes": null }, { "id": "1737188", "postDate": "03/28/2022 07:38:05", "content": "there is tf code in google research website:
\nhttps://github.com/google-research/sound-separation/tree/master/models/bird_mixit
for pytorch user, you can try to use this: https://asteroid-team.github.io/
\nfollow the tutorial documentation and i think you can setup bird mixIT within hours.
if you are interested in trends in audio source separation try:
\nhttps://www.youtube.com/watch?v=AB-F2JmI9U4
\nhttps://sigsep.github.io/AES2020_CurrentTrendsInSourceSeparation.pdf
Current Trends in Audio Source Separation
\nAES Virtual Symposium: Applications of Machine Learning in Audio September 28-29 2020
\nSpeaker: Fabian-Robert Stöter (Inria), Stefan Uhlich (Sony)
@hengck23 mixIT recipe is already provided in asteroid under WHAM example (egs/wham/MixIT)
\nHow can we adapt it for birds since WHAM is synthetic mixture (the s1 and s2 source is available) and it won't be the case for birds
@hengck23 I checked this blog and I am wondering how could I use this for the competition. If I understand correctly we can feed the model with audio files and we get back 4 or 8 separated files. But it should be labeled somehow but how should we know what is the label of the separated files? Do you have any idea?
", "rawMarkdown": "hengck23 I checked this blog and I am wondering how could I use this for the competition. If I understand correctly we can feed the model with audio files and we get back 4 or 8 separated files. But it should be labeled somehow but how should we know what is the label of the separated files? Do you have any idea?", "votes": null }, { "id": "1745882", "postDate": "04/05/2022 11:45:36", "content": "Sorry for interrupt.
\nDo you also checked the original paper?
\nIf you don’t, check chapter 2.5.
\n\nTo combine the separation and classification models, we apply a sep- aration model to an input audio window to obtain M output chan- nels. We then apply the classification model to each separated chan- nel and the original audio and take the maximum probability for each species.
\n
https://arxiv.org/pdf/2110.03209.pdf
", "rawMarkdown": "Sorry for interrupt.\nDo you also checked the original paper?\nIf you don’t, check chapter 2.5.\n\n> To combine the separation and classification models, we apply a sep- aration model to an input audio window to obtain M output chan- nels. We then apply the classification model to each separated chan- nel and the original audio and take the maximum probability for each species.\n\nhttps://arxiv.org/pdf/2110.03209.pdf", "votes": null }, { "id": "1745886", "postDate": "04/05/2022 11:50:52", "content": "In the original paper, it introduces other techniques to improve the performance of sound event detection model. I think this thesis could be a baseline solution for the competition.
", "rawMarkdown": "In the original paper, it introduces other techniques to improve the performance of sound event detection model. I think this thesis could be a baseline solution for the competition.", "votes": null }, { "id": "1746372", "postDate": "04/05/2022 17:43:51", "content": "Thank you! I haven't read the paper yet, but I will! Thank you for the help. :)
", "rawMarkdown": "Thank you! I haven't read the paper yet, but I will! Thank you for the help. :)", "votes": null }, { "id": "1748501", "postDate": "04/07/2022 16:38:20", "content": "I did a fast implementation here with some visualization for those interested:
\n", "rawMarkdown": "I did a fast implementation here with some visualization for those interested:\n - https://www.kaggle.com/code/dschettler8845/birdclef-22-sound-separation-experiment", "votes": null } ], "comments": [ { "id": 1737180, "author_name": "stefankahl", "author_url": "", "post_date": "03/28/2022 07:28:27", "content": "And in case you haven't noticed, the lead author of this publication is our competition host @tomdenton :)
", "votes": null, "replies": [] }, { "id": 1737188, "author_name": "hengck23", "author_url": "", "post_date": "03/28/2022 07:38:05", "content": "there is tf code in google research website:
\nhttps://github.com/google-research/sound-separation/tree/master/models/bird_mixit
for pytorch user, you can try to use this: https://asteroid-team.github.io/
\nfollow the tutorial documentation and i think you can setup bird mixIT within hours.
if you are interested in trends in audio source separation try:
\nhttps://www.youtube.com/watch?v=AB-F2JmI9U4
\nhttps://sigsep.github.io/AES2020_CurrentTrendsInSourceSeparation.pdf
Current Trends in Audio Source Separation
\nAES Virtual Symposium: Applications of Machine Learning in Audio September 28-29 2020
\nSpeaker: Fabian-Robert Stöter (Inria), Stefan Uhlich (Sony)
@hengck23 mixIT recipe is already provided in asteroid under WHAM example (egs/wham/MixIT)
\nHow can we adapt it for birds since WHAM is synthetic mixture (the s1 and s2 source is available) and it won't be the case for birds
@hengck23 I checked this blog and I am wondering how could I use this for the competition. If I understand correctly we can feed the model with audio files and we get back 4 or 8 separated files. But it should be labeled somehow but how should we know what is the label of the separated files? Do you have any idea?
", "votes": null, "replies": [ { "id": 1745882, "author_name": "tatamikenn", "author_url": "", "post_date": "04/05/2022 11:45:36", "content": "Sorry for interrupt.
\nDo you also checked the original paper?
\nIf you don’t, check chapter 2.5.
\n\nTo combine the separation and classification models, we apply a sep- aration model to an input audio window to obtain M output chan- nels. We then apply the classification model to each separated chan- nel and the original audio and take the maximum probability for each species.
\n
https://arxiv.org/pdf/2110.03209.pdf
", "votes": null, "replies": [] }, { "id": 1745886, "author_name": "tatamikenn", "author_url": "", "post_date": "04/05/2022 11:50:52", "content": "In the original paper, it introduces other techniques to improve the performance of sound event detection model. I think this thesis could be a baseline solution for the competition.
", "votes": null, "replies": [] }, { "id": 1746372, "author_name": "kucsikz", "author_url": "", "post_date": "04/05/2022 17:43:51", "content": "Thank you! I haven't read the paper yet, but I will! Thank you for the help. :)
", "votes": null, "replies": [] } ] }, { "id": 1748501, "author_name": "dschettler8845", "author_url": "", "post_date": "04/07/2022 16:38:20", "content": "I did a fast implementation here with some visualization for those interested:
\n", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1734358": "https://ai.googleblog.com/2022/01/separating-birdsong-in-wild-for.html\nSeparating Birdsong in the Wild for Classification", "1737180": "And in case you haven't noticed, the lead author of this publication is our competition host @tomdenton :)", "1737188": "there is tf code in google research website:\nhttps://github.com/google-research/sound-separation/tree/master/models/bird_mixit\n\nfor pytorch user, you can try to use this: https://asteroid-team.github.io/\nfollow the tutorial documentation and i think you can setup bird mixIT within hours.\n\n---\n\nif you are interested in trends in audio source separation try:\nhttps://www.youtube.com/watch?v=AB-F2JmI9U4\nhttps://sigsep.github.io/AES2020_CurrentTrendsInSourceSeparation.pdf\n\nCurrent Trends in Audio Source Separation\nAES Virtual Symposium: Applications of Machine Learning in Audio September 28-29 2020\nSpeaker: Fabian-Robert Stöter (Inria), Stefan Uhlich (Sony)", "1738843": "hengck23 mixIT recipe is already provided in asteroid under WHAM example (egs/wham/MixIT)\nHow can we adapt it for birds since WHAM is synthetic mixture (the s1 and s2 source is available) and it won't be the case for birds", "1745767": "hengck23 I checked this blog and I am wondering how could I use this for the competition. If I understand correctly we can feed the model with audio files and we get back 4 or 8 separated files. But it should be labeled somehow but how should we know what is the label of the separated files? Do you have any idea?", "1745882": "Sorry for interrupt.\nDo you also checked the original paper?\nIf you don’t, check chapter 2.5.\n\n> To combine the separation and classification models, we apply a sep- aration model to an input audio window to obtain M output chan- nels. We then apply the classification model to each separated chan- nel and the original audio and take the maximum probability for each species.\n\nhttps://arxiv.org/pdf/2110.03209.pdf", "1745886": "In the original paper, it introduces other techniques to improve the performance of sound event detection model. I think this thesis could be a baseline solution for the competition.", "1746372": "Thank you! I haven't read the paper yet, but I will! Thank you for the help. :)", "1748501": "I did a fast implementation here with some visualization for those interested:\n - https://www.kaggle.com/code/dschettler8845/birdclef-22-sound-separation-experiment" }, "source": "meta" }