New Technique to Identify Birdsongs
May. 10 de 2012
By: Ana María Escobar Jiménez, Unimedios
Making an inventory of the birds in an ecosystem is no easy task; generally, it is done through visual inspection in the field that involves arduous workdays of travel and observation and also requires auditory training.
Sergio Tobón Ocampo, president of the National Network of Bird Observers of Colombia, says that the best hours of the day for observation and study are dawn and dusk. Specialized long–range binoculars are needed along with lenses able to capture the smallest details of the birds.
In countries such as Puerto Rico, where conservation and protection of birds is regulated by the Wildlife Law of 1976, a computer program is being implemented to facilitate the recognition of species.
Inspired by the Automated Remote Biodiversity Monitoring Network (Arbimon), a network for monitoring of biodiversity, the Signal Processing and Recognition Group (Grupo de Procesamiento y Reconocimiento de Señales) at the Universidad Nacional de Colombia in Manizales developed an acoustical characterization study to produce software that is adapted to local conditions.
"Based on Arbimon –which was created at the Universidad de Puerto Rico–, and in collaboration with professionals in the fields of computer science, electronics, biology and ecology, we wanted to implement new automatic recognition and signal processing techniques as representations of dissimilarities, which was the method used in this case", according to research coordinator Mauricio Orozco Alzate.
The system operates through birdsong characterization. Traditional techniques analyze various acoustic signal measurements or particularities of the sounds –such as wave frequency and duration, among others–, whereas the UN study classifies the signals using methods based on dissimilarities.
This is a recent procedure in the theory of automatic or pattern recognition, which consists in directly comparing two elements without measuring their particular properties.
It is similar to the way human beings reason, when for example, they distinguish a particular person through general characteristics such as the face or voice; in other words, based on a complete pattern that we already have in mind, without stopping to determine if their skin is white or dark, without discriminating the color of their eyes or hair, or if the timbre of their voice is high or low.
"The methodology works because the program is trained to recognize the birdsongs from a database so that it is able to predict what animal it is. Very good performance is obtained, but enlarging the database of sounds is the key to increasing the precision", according to José Francisco Ruiz Muñoz, who is developing the technique as his master's thesis in industrial automation – engineering.
Application of this method is the starting point for consolidation of a larger scale process in automatic monitoring of diverse specimens.
According to the engineer, the procedure would be adapted to a natural environment by placing microphones in strategic sites; the sounds captured are sent over the Internet to a computer that has previously been fed the systemized songs, which then makes the identification; in this way, the researchers do not need to travel and temporary coverage is achieved over a larger space, in that the recording and analysis is permanent, with the added advantage that it can be carried out in a number of places simultaneously.
The tests were performed using 538 birdsongs, corresponding to 11 species from the Río Blanco Reserve in Manizales. The researchers then processed information, extracted the portions of the audio track containing the required segments, marked the samples with the label corresponding to the bird to which the sound pertained, and inserted this information into the system.
The material analyzed on the UN computers resulted in 97.87% certainty in the species identified. "This figure shows the good performance of the process, because it only erred with two species whose sounds are very similar", says Ruiz Muñoz.
Applications of this process include studying volcanic seismic signals, although this is not the core function but rather another way of proving its performance. Professor Orozco Alzate has been doing this since 2006 in collaboration with the Volcanological and Seismological Observatory (Observatorio Vulcanológico y Sismológico) of Manizales.
José Francisco Ruiz’s masters project can be implemented in this field because the nature of acoustic and seismic signals is similar: both waves are elastic with the difference that audio signals are propagated through air while vibrations are propagated through the earth.
In this manner, their experiments contribute towards the objective of the primary research, aimed at improving the identification of seismic signals and the delivery of already classified data to geologists, according to its typology.
In Colombia, where there are an estimated 2,000 species of birds –a huge resource in terms of biodiversity–, this new technology would have a number of uses, ranging from classification of species in determined ecosystems to the identification of new specimens as yet unknown in our territory.