Sea bream sound recordings
Description
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.
Files
Steps to reproduce
Steps to Reproduce the Dataset Collection The dataset can be fully reproduced by following the same controlled recording procedure used in the study. The complete steps are: 1. Experimental Tank Setup Prepare three independent aquaculture tanks with the following conditions: NE tank: Non‑Environmentally Enriched (6 fish) EE tank: Environmentally Enriched (7 fish) TT tank: Test Tank (used for background profiling and individual fish recording) Maintain environmental parameters: Temperature: 18 ± 1 °C Photoperiod: 12 h light : 12 h dark Volume: ~197 L Water flow + aeration: Match standard RAS conditions 2. Hydrophone Hardware and Placement Use an Aquarian Audio H2dM hydrophone connected via a 3.5 mm TRS → USB ADC adapter. Position the hydrophone: Depth: 0.48–0.50 m Offset from tank centre to avoid altering fish behaviour Secure cable to avoid contact with surfaces 3. Continuous Audio Recording Procedure Record using: Sampling rate: 44,100 Hz Resolution: 16‑bit PCM Uncompressed WAV format Capture: Multiple full feeding cycles for each tank Pre‑feeding, feeding, and post‑feeding periods Background‑only recordings in the TT tank Store all raw audio with timestamps. 4. Behaviour Annotation Synchronise recording times with behavioural observations: Anticipatory movement → Pre‑Feeding Surface strikes/pellet competition → Feeding Decline of activity → Post‑Feeding Pump/aeration noise only → Background Create a timeline annotation file for each tank. 5. Signal Conditioning and Filtering Process raw recordings using: Band‑pass filtering: 10–1000 Hz sampling: 44.1 kHz Energy‑based screening: Remove segments with RMS < 0.005 6. Segmentation Into Fixed Window Lengths Segment the filtered audio into: 1 s, 2 s, 5 s, 10 s, 20 s windows No overlap Assign behavioural labels using annotation timelines Maintain separate folders for each window length and class.
Institutions
- Nottingham Trent UniversityEngland, Nottingham