期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2021
卷号：99
期号：16
语种：English
出版社：Journal of Theoretical and Applied
摘要：An innate quality of living beings is the expression of their emotions in sound form. Birds stand out within living beings that perform sound communication, because they modulate their song to indicate moods or emotions as a protection or survival mechanism. This mechanism and the sound produced changes according to the taxonomic distribution of birds, since each species has its own physiognomy and even within the same species there can be different types of singing. Therefore, the study of the characteristic tonalities of each bird species and its song has become a topic of general interest in certain areas of biology, to determine the possible geographical locations or survival habits of each species. However, the classification of the tones generated in the song of a bird is a subject under study, due to the large number of sounds generated by a single species of bird. Therefore, this article proposes a strategy of classification and identification of the species of a bird by extracting characteristics of the tone of its song with a computational learning algorithm. Our scheme proposes the use of digitized bird sounds, which are processed by digital filters to extract the acoustic characteristics of interest. This filtering is performed in a two-stage scheme, which allows us to narrow down the region of interest of each digital file, which in the end constitutes the dataset of the learning system. To learn the typical characteristics of the sounds, different deep neural network structures are evaluated to identify the topology capable of replicating the characteristics of the samples. The results show a high performance of the identifier, which is linked to the characteristics of the network architecture.
