A framework for anuran monitoring, species classification and bioacoustical studies

This research work presents an infrastructure testbed to conduct bioacoustical studies and a novel application for the automatic identification of species from their vocalizations. The infrastructure designed has as purpose being able to deploy applications in disaster relief and environmental monitoring scenarios. Here, three different solar powered testbed are presented. The first one is a standalone unit to record audio, the second one is based on several processing units that can be used to form a sensor array processing (SAP), and the third one is based on a single robust unit with the capacity of record audio simultaneously from different microphones to form an array. An algorithm for automatic identification of species from their vocalizations is also presented. In this algorithm, an adaptive filtering process is conducted to enhance the audio signal, a segmentation process is performed using a temporal analysis and the Energy function, a new feature extraction stage based on Mel-frequency cepstrum coefficients (MFCC) and principal components analysis (PCA) is exposed. Finally, a k-nearest neighbors (k-NN) classification process is carried on, letting an adequate identification of the species existent in the audio sample, as can be inferred from the quantified performance evaluations which are presented.