\n\nDeep learning (DL) has greatly advanced audio classification, yet the field is limited by the scarcity of large-scale benchmark datasets that have propelled progress in other domains. While AudioSet is a pivotal step to bridge this gap as a universal domain dataset, its restricted accessibility and limited range of evaluation use cases challenge its role as the sole resource.
\n
\n Therefore, we introduce BirdSet, a large-scale benchmark dataset for audio classification focusing on avian bioacoustics. BirdSet surpasses AudioSet with:\n
\n- Over 6,800 recording hours (↑ 17%) from nearly 10,000 classes (↑ 18×) for training
\n- More than 400 hours (↑ 7×) across eight strongly labeled evaluation datasets.
\n
\n It serves as a versatile resource for use cases such as:- Multi-label classification
\n- Covariate shift
\n- Self-supervised learning
\n
\n We benchmark six well-known DL models in multi-label classification across three distinct training scenarios and outline further evaluation use cases in audio classification.
\n We host our dataset on Hugging Face for easy accessibility and offer an extensive codebase to reproduce our results.
\n\nBirdSet Dataset
\n\n
\n\n\n
\n\n\n
\n\nNotebook - 01 BirdSet Data Pipeline Tutorial
\nNotebook - 02 BirdSet Tutorial Augmentations
\nNotebook - 03 BirdSet Fine-Tuning Tutorial
\n
\n\nDeep learning (DL) has greatly advanced audio classification, yet the field is limited by the scarcity of large-scale benchmark datasets that have propelled progress in other domains. While AudioSet is a pivotal step to bridge this gap as a universal domain dataset, its restricted accessibility and limited range of evaluation use cases challenge its role as the sole resource.
\n
\n Therefore, we introduce BirdSet, a large-scale benchmark dataset for audio classification focusing on avian bioacoustics. BirdSet surpasses AudioSet with:\n
\n- Over 6,800 recording hours (↑ 17%) from nearly 10,000 classes (↑ 18×) for training
\n- More than 400 hours (↑ 7×) across eight strongly labeled evaluation datasets.
\n
\n It serves as a versatile resource for use cases such as:- Multi-label classification
\n- Covariate shift
\n- Self-supervised learning
\n
\n We benchmark six well-known DL models in multi-label classification across three distinct training scenarios and outline further evaluation use cases in audio classification.
\n We host our dataset on Hugging Face for easy accessibility and offer an extensive codebase to reproduce our results.
\n\nBirdSet Dataset
\n\n
\n\n\n
\n\n\n
\n\nNotebook - 01 BirdSet Data Pipeline Tutorial
\nNotebook - 02 BirdSet Tutorial Augmentations
\nNotebook - 03 BirdSet Fine-Tuning Tutorial
\n
This is a very useful summary.
", "rawMarkdown": "This is a very useful summary.", "votes": null } ], "comments": [ { "id": 3160541, "author_name": "", "author_url": "", "post_date": "03/26/2025 22:20:20", "content": "This is a very useful summary.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3146509": "# [BirdSet: A Large-Scale Dataset for Audio Classification in Avian Bioacoustics](https://arxiv.org/pdf/2403.10380)\n> **Deep learning (DL)** has greatly advanced audio classification, yet the field is limited by the scarcity of large-scale benchmark datasets that have propelled progress in other domains. While **AudioSet** is a pivotal step to bridge this gap as a universal domain dataset, its restricted accessibility and limited range of evaluation use cases challenge its role as the sole resource.\nTherefore, we introduce **BirdSet**, a large-scale benchmark dataset for audio classification focusing on **avian bioacoustics**. **BirdSet** surpasses **AudioSet** with:\n- Over **6,800 recording hours** (↑ 17%) from nearly **10,000 classes** (↑ 18×) for training\n- More than **400 hours** (↑ 7×) across **eight strongly labeled evaluation datasets**.\nIt serves as a versatile resource for use cases such as:\n- **Multi-label classification**\n- **Covariate shift**\n- **Self-supervised learning**\nWe benchmark six well-known DL models in multi-label classification across three distinct training scenarios and outline further evaluation use cases in audio classification.\nWe host our dataset on **Hugging Face** for easy accessibility and offer an extensive **codebase** to reproduce our results.\n\n---\n\n> # [BirdSet Dataset](https://github.com/DBD-research-group/BirdSet)\n> \n---\n\n# [Models](https://huggingface.co/collections/DBD-research-group/birdset-dataset-and-models-665ef710a28cbe70dfaa028a)\n> \n\n---\n\n# [Leaderboard](https://huggingface.co/spaces/DBD-research-group/BirdSet-Leaderboard)\n> \n\n---\n\n# Notebooks\n> #[Notebook - 01 BirdSet Data Pipeline Tutorial](https://www.kaggle.com/code/seshurajup/birdset-data-pipeline-tutorial?scriptVersionId=226898347)\n#[Notebook - 02 BirdSet Tutorial Augmentations](https://www.kaggle.com/code/seshurajup/02-birdset-tutorial-augmentations)\n#[Notebook - 03 BirdSet Fine-Tuning Tutorial](https://www.kaggle.com/code/seshurajup/03-birdset-fine-tuning-tutorial)\n\n---", "3160541": "This is a very useful summary." }, "source": "meta" }![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2Fd0325d6acd604756c5213e960e0e26ff%2FScreenshot%202025-03-11%20at%206.31.23AM.png?generation=1741654919107608&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2F57e618e5ee8c2b03649a86455395021c%2FScreenshot%202025-03-11%20at%206.32.20AM.png?generation=1741654955455985&alt=media\")
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F761268%2F07844bfae60f4f931df3d182ef6be46e%2FScreenshot%202025-03-11%20at%205.53.59AM.png?generation=1741652659278906&alt=media\")