Researchers at the University of Seville have used artificial intelligence to create an automatic classifier of the thousands of sounds of frogs and toads that can be recorded in a natural environment.
Amphibian sounds are known to be altered by increased ambient temperature, a phenomenon that, in addition to interfering with reproductive behavior, is useful as an indicator of global warming.
Climate change has consequences on the physiological functions of animals, frogs and toads do not escape it and their croaking is affected. The singing of these amphibians plays a key role in mating and sexual selection, which is why global warming is interfering with their reproduction.
When this exceeds a certain threshold, the physiological processes associated with the production of sound are restricted, and some songs are even inhibited. In fact, the onset, duration, and intensity of male-to-female calls are changed, influencing reproductive activity.
Taking this phenomenon into account, the analysis and classification of the sounds produced by certain species of amphibians have turned out to be a key indicator of temperature variations and, therefore, of the existence and evolution of global warming.
Wireless Audio Sensor Networks
To capture the sounds of the frogs, networks of audio sensors are placed, connected wirelessly in areas that can reach several hundred square kilometers. The problem is that a huge amount of bioacoustic information is collected in environments as noisy as a jungle, and this makes it difficult to identify the species and their songs.
To solve it, engineers from the University of Seville have turned to artificial intelligence. “We have segmented the sound into time windows orframes of audio and we have classified them using decision trees, a machine learning technique that is used in computing, ”explains Amalia Luque Sendra, co-author of the work.
MPEG-7 audio descriptors
To perform the classification, the researchers have relied on MPEG-7 audio descriptors and parameters, a standard way of representing audiovisual information. Details are published in the magazineExpert Systems with Applications.
This technique has been tested with real sounds of amphibians recorded in the middle of nature and provided by the National Museum of Natural Sciences. Specifically, 868 records with 369 mating calls sung by the male and 63 release songs issued by the female runner toad (Epidalea calamita), Along with 419 mating calls and 17 distress calls from the common midwife toad (Alytes obstetricans).
"In this case we obtained a success rate close to 90% when classifying the sounds," highlights Luque Sendra, who recalls that, in addition to the types of songs, the number of individuals of certain species of amphibians that are heard in a geographic region over time can also be used as an indicator of climate change.
“An increase in temperature affects the singing patterns,” he emphasizes, “but as these in most cases have a sexual call character, they end up also affecting the number of individuals. Our method is not yet capable of directly determining the exact number of specimens in an area, but it is a first approximation ”.
Amalia Luque, Javier Romero-Lemos, Alejandro Carrasco, Julio Barbancho. “Non-sequential automatic classification of anuran sounds for the estimation of climate-change indicators”.Expert Systems With Applications 95: 248–260, 2018.
With information from: