SEABAD: Southeast Asian Bird Activity Detection Dataset

SEABAD (Southeast Asian Bird Activity Detection) is a large-scale audio dataset for Edge AI bird detection. It includes 25,000 positive (bird-present) and 25,000 negative (bird-absent) 3-second WAV clips, sampled at 16 kHz. The dataset is designed for automated bird activity detection, biodiversity monitoring, and real-time acoustic sensing on low-power devices. It is fully machine-learning ready. Positive samples were extracted from Xeno-canto recordings in Malaysia and neighboring countries (Thailand, Indonesia, Singapore, and Brunei) through December 2025. Samples were obtained by extracting the highest-energy 3-second segment from each Xeno-canto recording. Negative samples were drawn primarily from the BirdCLEF “BAD” competition datasets (BirdVox, Freefield1010, and Warblr) and were further supplemented with non-bird environmental audio from the FSC22, ESC50, and DataSEC datasets. Released under CC BY 4.0, MyBAD provides an open, reproducible resource for developing Edge AI models for tropical bird monitoring and acoustic biodiversity assessment.SEABAD（东南亚鸟类活动检测，Southeast Asian Bird Activity Detection）是一款面向边缘AI（Edge AI）鸟类检测任务的大规模音频数据集。其包含25000条正样本（存在鸟类）与25000条负样本（无鸟类），均为时长3秒的WAV格式音频片段，采样率为16 kHz。 该数据集旨在支撑自动化鸟类活动检测、生物多样性监测，以及在低功耗设备上开展实时声学感知任务，且已完全适配机器学习开发流程。 正样本提取自2025年12月前马来西亚及周边国家（泰国、印度尼西亚、新加坡、文莱）的Xeno-canto录音数据，具体提取方式为从每条Xeno-canto录音中截取能量最高的3秒片段作为样本。 负样本主要取自BirdCLEF "BAD"竞赛数据集（包含BirdVox、Freefield1010与Warblr），并额外补充了来自FSC22、ESC50及DataSEC数据集的非鸟类环境音频。 该数据集以CC BY 4.0协议发布，MyBAD可为热带鸟类监测与声学生物多样性评估的边缘AI模型开发提供一套开放、可复现的研究资源。