Field recordings of the Cuban tree-frog Eleutherodactylus eileenae (“Colines”) capture rich nocturnal chorus behavior, but ambient noise, overlapping calls and multi-mic delays make manual analysis slow and error-prone. No Cuban study to date has combined passive acoustic monitoring with statistical interaction models to quantify how individual frogs coordinate their calling.
- Noise Filtering
Applied percentile-based spectral thresholds to each Mel-spectrogram, suppressing broadband (wind, traffic) and transient (other animals) noise. - Multi-Mic Synchronization
Aligned nine simultaneous tracks via cross-correlation of histograms. - Heuristic Call Detection
• Energy-Based Algorithm: Selected highest-energy events per channel.
• Spectro-Temporal Clustering: Grouped local spectrogram peaks into CO/LIN pairs via agglomerative clustering. - Interaction Modeling
Mapped each frog’s call series to binary “spin” states and inferred pairwise couplings (J_{ij}) using an exact maximum-likelihood Ising framework.
- Detection Performance: Clustering outperformed energy-thresholding when chorusing activity was sparse; both algorithms achieved near-identical timing (9/9 microphones, (R^2 \ge 0.99)).
- Interaction Inference: The Ising model revealed a handful of significant couplings, but a simpler independent‐spin model better predicted call‐patterns, suggesting either weak acoustic interactions or limits of the equilibrium assumption.
Our fully automated pipeline delivers reproducible call‐detection and quantifies chorus coordination in E. eileenae. Although the standard Ising formulation showed limited predictive power, our approach lays the groundwork for scalable ecoacoustic studies and points toward dynamic, non-equilibrium models to capture frog chorus behavior more faithfully.