Lobster Sound Dataset
Description
The following lobster sounds dataset was collected from Johnshaven in Scotland between June 16 to June 18, 2024.
Files
Steps to reproduce
Underwater Bioacoustics Dataset Collection This dataset was collected using a comprehensive methodology for passive bioacoustic monitoring of European lobsters. The process combined specialized hardware, software, and analytical workflows to ensure data quality and reproducibility. Methodology Overview Hardware: Underwater sounds were recorded using hydrophones connected to embedded computing units, such as a Raspberry Pi 3 Model B, for in-situ data acquisition. Data Collection: Recordings were conducted in a controlled tank environment to capture and classify specific lobster sounds, including rasping and clicking. Lobsters were identified by sex and age with the help of domain experts to create a well-labeled dataset. Analytical Workflow Data Processing: Raw audio recordings were processed using Python (version 3.12.3) and libraries like PyAudio. Feature Extraction & Modeling: Key acoustic features, such as Mel-Frequency Cepstral Coefficients (MFCCs), were extracted. Multiple AI models, including classical classifiers (e.g., SVM, KNN) and deep learning architectures, were trained and optimized using grid search and Principal Component Analysis (PCA) for dimensionality reduction. Validation: All models were validated using cross-validation techniques to ensure accuracy and to confirm the robustness of the results. Reproducibility To ensure reproducibility, the detailed protocols, code, and data are documented according to ISO standards. The public availability of code can be found at git@github.com:N0736086/lobster_soundscodes.git to facilitate validation and further research.
Institutions
- Nottingham Trent University - Clifton Campus