Sea bream sound recordings

This dataset contains multi-window segmented passive acoustic recordings of Gilthead Sea Bream (Sparus aurata) feeding behaviour collected in controlled recirculating aquaculture system (RAS) environments. Continuous raw audio (44.1 kHz, 16‑bit PCM) was recorded across three experimental tanks using an Aquarian Audio H2dM hydrophone. To support machine‑learning and deep‑learning research, the recordings were processed through a nine‑stage pipeline including band‑limited filtering (50–1000 Hz), RMS-based activity validation, and fixed‑length segmentation into five temporal resolutions: 1 s, 2 s, 5 s, 10 s, and 20 s. Each segment is annotated into one of four behavioural classes—Background, Pre-Feeding, Feeding, and Post-Feeding—based on synchronised behavioural logs collected during multiple feeding cycles. The repository includes the segmented audio files, metadata (tank conditions, hydrophone placement, timestamps, and behavioural labels), and documentation describing the preprocessing workflow. This dataset is intended to support reproducible research in bioacoustics, underwater signal processing, behavioural inference, and precision aquaculture.

本数据集包含在可控循环水养殖系统（Recirculating Aquaculture System, RAS）环境中采集的金头鲷（Gilthead Sea Bream, Sparus aurata）摄食行为的多窗口分段被动声学记录。研究人员采用Aquarian Audio H2dM水听器，在三个实验养殖池中录制了连续原始音频（44.1 kHz、16位脉冲编码调制（PCM））。为支撑机器学习与深度学习研究，该音频经九阶段流水线处理：涵盖带限滤波（50–1000 Hz）、基于均方根（Root Mean Square, RMS）的活性验证，以及固定时长分段，生成5种时间分辨率的音频片段：1秒、2秒、5秒、10秒及20秒。基于多次投喂周期中同步采集的行为日志，每个片段被标注为四类行为之一：背景噪声、投喂前、投喂中及投喂后。本数据集仓库包含分段音频文件、元数据（养殖池环境参数、水听器部署位置、时间戳及行为标签），以及描述预处理流程的文档。本数据集旨在为生物声学、水下信号处理、行为推断以及精准水产养殖领域的可复现研究提供支撑。