Abstract—Effcient neural coding is a theoretical model in sensory neuroscience, positing that biological systems maximize information transfer to the brain while minimizing neural resources. While this concept has been extensively studied in the context of human speech perception and the human brain, its applicability to non-human vocalizations remains relatively unexplored. This study applies sparse coding to bat echolocation calls and demonstrates that the resulting kernel representations exhibit properties consistent with effcient coding principles, namely high compactness, sparsity, and functional specialization. Distinct kernel activation profles were found to encode different echolocation call shapes and identify anomalous, irregular calls, indicating
that the model captures biologically relevant features and exhibits sensitivity to deviations from stereotyped call structure.
These fndings underscore the advantages of sparse coding over traditional signal representations for modeling bat vocalizations and align with evidence that effcient coding strategies are shared across mammals, tuned to species-specifc vocal patterns and conspecifc vocalizations. This work improves the interpretability of animal auditory processing and provides a computational basis for modeling mammalian vocalizations, thereby supporting further research in decoding animal
signals and interspecies communication.