I do not know if some \"recordings\" of the train data are the right ones to get good results. Are they really intended?
\nFor example
pandas pd\n librosa\n IPython.display Audio\n\ndf_train = pd.read_csv()\nf = df_train.iloc[][]\nwav, sr = librosa.load(path=, sr=)\nAudio(wav, rate=)\nreveals a 1:30 min recording of a human talking about: \"Successful recording on April 4, 2022, at 2:12:44 AM. This is a recording of a single adult male of the genus Rhogeessa, species gracilis. Omnidirectional recording in the ultrasonic range. Distance from microphone 0.5 meters, temperature 31 degrees Celsius. There is no information on the sound pressure level. A Wildlife Accoustics bat detector was used for this recording…\" (transcripted and translated with gemini on mobile phone)
\nAlso sample 1,2,3 and 570 are like this one.
", "messages": [ { "id": 3146415, "postDate": "2025-03-10T20:35:07.273Z", "content": "I do not know if some \"recordings\" of the train data are the right ones to get good results. Are they really intended?
\nFor example
pandas pd\n librosa\n IPython.display Audio\n\ndf_train = pd.read_csv()\nf = df_train.iloc[][]\nwav, sr = librosa.load(path=, sr=)\nAudio(wav, rate=)\nreveals a 1:30 min recording of a human talking about: \"Successful recording on April 4, 2022, at 2:12:44 AM. This is a recording of a single adult male of the genus Rhogeessa, species gracilis. Omnidirectional recording in the ultrasonic range. Distance from microphone 0.5 meters, temperature 31 degrees Celsius. There is no information on the sound pressure level. A Wildlife Accoustics bat detector was used for this recording…\" (transcripted and translated with gemini on mobile phone)
\nAlso sample 1,2,3 and 570 are like this one.
", "rawMarkdown": "I do not know if some \"recordings\" of the train data are the right ones to get good results. Are they really intended?\nFor example\n```python\nimport pandas as pd\nimport librosa\nfrom IPython.display import Audio\n\ndf_train = pd.read_csv('/kaggle/input/birdclef-2025/train.csv')\nf = df_train.iloc[0]['filename']\nwav, sr = librosa.load(path=f'/kaggle/input/birdclef-2025/train_audio/{f}', sr=None)\nAudio(wav, rate=32000)\n```\nreveals a 1:30 min recording of a human talking about: \"Successful recording on April 4, 2022, at 2:12:44 AM. This is a recording of a single adult male of the genus Rhogeessa, species gracilis. Omnidirectional recording in the ultrasonic range. Distance from microphone 0.5 meters, temperature 31 degrees Celsius. There is no information on the sound pressure level. A Wildlife Accoustics bat detector was used for this recording...\" (transcripted and translated with gemini on mobile phone)\n\nAlso sample 1,2,3 and 570 are like this one.", "votes": 20 }, { "id": 3149123, "postDate": "2025-03-13T20:11:14.667Z", "content": "Thanks for noticing this!
\nFabio's voice can be easily filtered by a simple algorithm: actual insect sound is surrounded by short inserts of silence. I have created a notebook that uses this idea.
\nAlso, I want to confirm, that there is voice in almost all of the recordings in CSA collection. However, it's not so easy to delete it since different authors follow different patterns.
", "rawMarkdown": "Thanks for noticing this!\n\nFabio's voice can be easily filtered by a simple algorithm: actual insect sound is surrounded by short inserts of silence. I have created [**a notebook that uses this idea**](https://www.kaggle.com/code/kdmitrie/bc25-separation-voice-from-data/notebook). \n\nAlso, I want to confirm, that there is voice in almost all of the recordings in CSA collection. However, it's not so easy to delete it since different authors follow different patterns.", "votes": 9 }, { "id": 3148665, "postDate": "2025-03-13T11:56:27.817Z", "content": "All recordings of the author Fabio A. Sarria-S contain a human voice.
", "rawMarkdown": "All recordings of the author Fabio A. Sarria-S contain a human voice.", "votes": 7 }, { "id": 3146731, "postDate": "2025-03-11T07:44:00.347Z", "content": "Yes, we expect the training data to be noisy to some degree. We download data from public libraries and it's not feasible to manually curate ~28K files. However, through smart preprocessing, you should be able to clean the data to an extent where you can use it for training.
", "rawMarkdown": "Yes, we expect the training data to be noisy to some degree. We download data from public libraries and it's not feasible to manually curate ~28K files. However, through smart preprocessing, you should be able to clean the data to an extent where you can use it for training.", "votes": 5, "replies": [ { "id": 3200927, "postDate": "2025-05-13T08:25:44.030Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 3146512, "postDate": "2025-03-11T01:09:58.350Z", "content": "I noticed that unexpected (at least for me) train_audio in spanish which is a description of the recordings.
\nThen, I wrote on my Notebook: \"description of the recordings, not an expected animal.\"
The translation with Gemini on your mobile is amazing! Thanks for the description and its translation.
", "rawMarkdown": "I noticed that unexpected (at least for me) train_audio in spanish which is a description of the recordings. \nThen, I wrote on my Notebook: \"description of the recordings, not an expected animal.\" \n\nThe translation with Gemini on your mobile is amazing! Thanks for the description and its translation.", "votes": 3, "replies": [ { "id": 3200930, "postDate": "2025-05-13T08:27:41.183Z", "rawMarkdown": "", "isDeleted": true } ] }, { "id": 3149290, "postDate": "2025-03-14T03:08:03.103Z", "content": "Just add one more class to the train data - \"human\" - ...,Homo sapiens,Human,Mammalia 😀 . I think the data is easy to find
i was thinking that too atleast the model can verify that its from human and cant be any animal? 😀
", "rawMarkdown": "i was thinking that too atleast the model can verify that its from human and cant be any animal? 😀", "votes": 1 } ] } ], "comments": [ { "id": 3149123, "author_name": "Konstantin Dmitriev", "author_url": "", "post_date": "2025-03-13T20:11:14.667000", "content": "Thanks for noticing this!
\nFabio's voice can be easily filtered by a simple algorithm: actual insect sound is surrounded by short inserts of silence. I have created a notebook that uses this idea.
\nAlso, I want to confirm, that there is voice in almost all of the recordings in CSA collection. However, it's not so easy to delete it since different authors follow different patterns.
", "votes": 9, "replies": [] }, { "id": 3148665, "author_name": "Mart Preusse", "author_url": "", "post_date": "2025-03-13T11:56:27.817000", "content": "All recordings of the author Fabio A. Sarria-S contain a human voice.
", "votes": 7, "replies": [] }, { "id": 3146731, "author_name": "Stefan Kahl", "author_url": "", "post_date": "2025-03-11T07:44:00.347000", "content": "Yes, we expect the training data to be noisy to some degree. We download data from public libraries and it's not feasible to manually curate ~28K files. However, through smart preprocessing, you should be able to clean the data to an extent where you can use it for training.
", "votes": 5, "replies": [ { "id": 3200927, "author_name": "", "author_url": "", "post_date": "2025-05-13T08:25:44.030000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3146512, "author_name": "Marília Prata", "author_url": "", "post_date": "2025-03-11T01:09:58.350000", "content": "I noticed that unexpected (at least for me) train_audio in spanish which is a description of the recordings.
\nThen, I wrote on my Notebook: \"description of the recordings, not an expected animal.\"
The translation with Gemini on your mobile is amazing! Thanks for the description and its translation.
", "votes": 3, "replies": [ { "id": 3200930, "author_name": "", "author_url": "", "post_date": "2025-05-13T08:27:41.183000", "content": "", "votes": 0, "replies": [] } ] }, { "id": 3149290, "author_name": "Pavel Orlov", "author_url": "", "post_date": "2025-03-14T03:08:03.103000", "content": "Just add one more class to the train data - \"human\" - ...,Homo sapiens,Human,Mammalia 😀 . I think the data is easy to find
i was thinking that too atleast the model can verify that its from human and cant be any animal? 😀
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "3146415": "I do not know if some \"recordings\" of the train data are the right ones to get good results. Are they really intended?\nFor example\n```python\nimport pandas as pd\nimport librosa\nfrom IPython.display import Audio\n\ndf_train = pd.read_csv('/kaggle/input/birdclef-2025/train.csv')\nf = df_train.iloc[0]['filename']\nwav, sr = librosa.load(path=f'/kaggle/input/birdclef-2025/train_audio/{f}', sr=None)\nAudio(wav, rate=32000)\n```\nreveals a 1:30 min recording of a human talking about: \"Successful recording on April 4, 2022, at 2:12:44 AM. This is a recording of a single adult male of the genus Rhogeessa, species gracilis. Omnidirectional recording in the ultrasonic range. Distance from microphone 0.5 meters, temperature 31 degrees Celsius. There is no information on the sound pressure level. A Wildlife Accoustics bat detector was used for this recording...\" (transcripted and translated with gemini on mobile phone)\n\nAlso sample 1,2,3 and 570 are like this one.", "3149123": "Thanks for noticing this!\n\nFabio's voice can be easily filtered by a simple algorithm: actual insect sound is surrounded by short inserts of silence. I have created [**a notebook that uses this idea**](https://www.kaggle.com/code/kdmitrie/bc25-separation-voice-from-data/notebook). \n\nAlso, I want to confirm, that there is voice in almost all of the recordings in CSA collection. However, it's not so easy to delete it since different authors follow different patterns.", "3148665": "All recordings of the author Fabio A. Sarria-S contain a human voice.", "3146731": "Yes, we expect the training data to be noisy to some degree. We download data from public libraries and it's not feasible to manually curate ~28K files. However, through smart preprocessing, you should be able to clean the data to an extent where you can use it for training.", "3146512": "I noticed that unexpected (at least for me) train_audio in spanish which is a description of the recordings. \nThen, I wrote on my Notebook: \"description of the recordings, not an expected animal.\" \n\nThe translation with Gemini on your mobile is amazing! Thanks for the description and its translation.", "3149290": "Just add one more class to the train data - \"human\" - `...,Homo sapiens,Human,Mammalia` 😀 . I think the data is easy to find" } }