Sebastián Aguilera Novoa

In this study, we introduce a physical-computational model designed to automate the generation of birdsongs from recorded samples. By integrating concepts from numerical analysis, signal processing, and numerical optimization, we formulate the problem as a minimization task to identify optimal parameters that accurately replicate the tempo-spectral features of original birdsongs. Our model undergoes extensive testing and evaluation, primarily focusing on the Rufous-collared Sparrow songs, achieving impressive results with relative errors in fundamental frequency below 2%.

Additionally, we evaluate the model on the Ocellated Tapaculo and Mimus Gilvus species, known for their simpler songs. The results demonstrate the model’s robustness and versatility. To showcase its capabilities, we generate multiple syllables from a single recording, highlighting the model’s potential as a highly effective data augmentation technique for birdsongs. This novel approach addresses the challenge of data scarcity in birdsong analysis, contributing significantly to various fields such as biology, neuroscience, computer science, and ecology.