Multiscale Signal Manifold Modeling for Whale Communication

Multiscale Signal Manifold Modeling for Whale Communication Whales, and sperm whales in particular, communicate through structured click sequences known as codas. These signals display remarkable richness in tempo, rhythm, and ornamentation, suggesting that they may form a complex communication system with symbolic or semantic properties. Conventional machine learning approaches—such as clustering, CNNs, or RNNs—can classify or group codas, but they face key limitations. Hierarchical structure is often flattened into static features, temporal elasticity is treated as noise, and micro-variations are discarded. Furthermore, links between acoustic form and behavioral context remain poorly understood. This technical note introduces a multiscale signal manifold framework. The method treats codas not as flat sequences but as entities embedded in a scale–time manifold. By incorporating: scale-conditioned metrics (to handle tempo elasticity), hierarchical graph representations (click → coda → phrase → dialogue), and resonance-based alignment with behavioral context, the framework enables separation of form from meaning and provides a principled path toward semantic interpretation. The paper outlines: a modular software architecture for preprocessing, representation, modeling, and alignment, a phased plan of experimental studies, expected outcomes such as robust semantic units, cross-clan generalization, and conservation insights, and an extension to dolphin communication, showing how the same principles could apply to multimodal signaling in other marine mammals. This open proposal is intended as a practical resource for marine biologists, acousticians, and data scientists. It aims to bridge the gap between pattern recognition and semantic-level understanding in whale communication. License: CC BY (open access, free to use with attribution).