Deep Learning Recognition of Acoustic Adaptation in Traffic-Exposed Frogs

Abstract

Introduction: Urbanization and anthropogenic traffic noise impose significant selective pressures on the vocal communication systems of wildlife, particularly anurans (frogs), which rely heavily on acoustic signals for mating and territorial defense. To cope with low-frequency masking caused by traffic, urban frog populations frequently undergo acoustic adaptation, shifting call parameters such as dominant frequency, call duration, and pulse rate. Materials & Methods: This study investigates the automated recognition of these adaptive vocal shifts using advanced deep learning architectures. Audio recordings of frog calls were collected from high-traffic roadside habitats and pristine control wetlands during the breeding seasons. Mel-frequency cepstral coefficients (MFCCs) and spectrogram features extracted from the recordings served as input for a convolutional neural network (CNN) and recurrent neural network (RNN) hybrid pipeline designed to classify adaptation patterns and quantify frequency shifts. Results: The deep learning model successfully recognized and classified traffic-exposed acoustic adaptations with high precision, achieving superior accuracy compared to traditional statistical feature-matching methods. The results demonstrate that traffic-exposed frogs consistently elevate their call dominant frequencies to escape ambient low-frequency noise interference. Conclusion: Integrating deep learning into bioacoustic monitoring provides a scalable and robust approach for assessing how wildlife adapts to anthropogenic noise pollution, offering valuable insights into urban wildlife resilience and evolutionary ecology.