Houston Toad and Other Chorusing Amphibian Species Call Detection Using Deep Learning Architectures

The preservation of species from extinction using automatic animal voice detection from audio recordings is a topic of high interest in bioacoustics. The Houston Toad is an endangered amphibian and researchers of the Biology Department at Texas State University are working on a project to rescue this species. The initial approach used by the researchers is called Automated Recording Device (ARD) that only detects the Houston Toad calls. However, it has shown limited success identifying toad calls. If a species is not a Houston Toad but has a frequency spectrum close to a Houston Toad, then the ARD falsely identifies it as a Houston Toad. Hence, the current ARD solution produces high false-positives. This paper proposes a modified ARD solution to detect not only the Houston Toad, but also the Gulf Coast Toad, ther Crawfish Frog, and the Woodhouse's Toad experimenting deep learning architectures: Recurrent Neural Network (RNN), Convolutional neural network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRUs) with dominating audio features such as MFCC, LPC, PLP, Mel-Filterbanks, Spectrogram and takes an attempt to improve the performance of the toad calls identification system with reduced false-postive rate. This work also proposes a new single board computing platform NVIDIA Jetson Nano for on-filed deployment of the model.