AgmoNET

Overview

What this project explores

Passive acoustic monitoring produces more audio than researchers can label by hand. AgmoNET was developed in the context of my master's research to classify local bird vocalizations and connect model predictions with questions about ecological activity over time.

Contribution

My role

Research and software development across audio preparation, augmentation, CNN and ResNet models, evaluation, experiments, long-term activity analysis and visualization.

Current state

What exists today

• Research code covers audio segmentation, augmentation, Mel spectrograms, model training and evaluation.
• Activity analysis aggregates predictions by species and station, accounts for missing dates, and aligns activity with sunrise and sunset.
• The codebase is not a packaged application or a public demo; it relies on research-specific data and configuration.
• Classification of isolated calls and detection in overlapping field soundscapes are distinct evaluation problems.

Toolkit

Tools and methods

Next

Where it goes from here

No new development milestone is currently set. Research figures and evaluation context will be added to this project presentation as they are consolidated.