Bird Sounds Recognition Based on Optoelectric Reservoir Computing

This work explores bird sounds recognition using an optoelectric reservoir computing (RC) system to address challenges in traditional methods, such as high computational costs and hardware demands. The system employs a MachZehnder Modulator (MZM) with a feedback loop as the reservoir. A dataset of 1357 samples from five bird species was used, with Mel-frequency cepstral coefficients (MFCCs) extracted as features. The system achieves a minimum error rate of 0.01, demonstrating that appropriate feedback strength <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(0&lt;\beta&lt;0.6)$</tex> is crucial for optimal nonlinear transformation and memory capabilities for the RC system. Excessive feedback strength (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\boldsymbol{\beta}&gt;\boldsymbol{0.6}$</tex>) causes system oscillations and increases error rate. In addition, the variation of the recognition error rate with respect to the offset phase has also been analyzed. The results reveal that as the compensation phase varies from 0 to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2 \pi$</tex>, the error rate exhibits minor fluctuations without a clear or consistent pattern. The results of this work highlight the potential of optoelectric RC for efficient, low-power bird sound recognition in ecological applications.