Automatic segmentation and classification of the Thamnophilidae's family of the western andean region of Colombia

Colombia has the greatest diversity of birds in the world, approximately 1900 species, and therefore it is important to promote their protection and conservation. Monitoring the vocalization of the birds allows a sampling that does not affect the species or the habitat, helps to monitor migrations and estimate the populations present in the determined area, in addition to strengthening different fields of research. In the Thamnophilidae family, their vocalization is key to find differences in speciation. This paper propose a preprocessing process where the files first are separate from the environment noise with a RPCA technique and then is apply a segmentation process, when is calculated the Hilbert enveloping that find the importance segments in the audio. These segments contain the syllables that are saved separately to feed the classification system. Different techniques of Machine Learning and Deep Learning are applied and were obtained high accuracies, like the support vector machine that obtained a 96, $96\pm 0,64$.