Ecosystem sounds are plentiful, especially when all mixed together, and a new computer system has been designed to sort them out. Ecologists need to be able to discern the individual sounds of various bird species, for example, when they are studying an ecosystem. It is important that the sound analysis system is accurate and can handle a large amount of data. A recent report has announced development of a new system that employs machine learning rather than manual feature detection. The new system employs what has been named “unsupervised feature learning” and it can be used to study ecosystem sounds.
Dan Stowell and Mark D. Plumbley, who are from the Centre for Digital Music at Queen Mary University of London, are the authors of the paper that describes how this new system works when performing large-scale classification of bird sounds. The description was published in the journal PeerJ.
This new system provides an example for how the emerging field of soundscape ecology works. It involves use of automated recorders and very powerful sound analysis software. Soundscape ecology is about recording sounds from an environment, describing the sounds from the life forms in that environment, defining patterns of ecosystem sounds and tracking how they change over time.
The technology was based on human voice recognition systems. Human voice recognition systems are the computer programs that listen to a voice on the phone when someone is answering questions asked by a computer voice. When one answers yes or no, the human voice recognition system discerns what was said and then moves to the next question. Development of these voice recognition systems has the potential to have someone say their name, and have the program understand what was said.
When this new system is at work, it is able to pick out a single species of bird song from a cacophony of sounds. Once the system has identified an individual species it can then classify a single bird species. The soundscape ecology system can be applied with other animals as well. It has been reported that there is a bioacoustics scientist at Cornell University who has six million hours of underwater recordings and he needs a method to analyze all this information. He intends to pick out the signature sounds of different marine mammals.
Soundscape ecology is used to analyze the makeup of an ecosystem over long periods of time. It can be used to track the movements of various animals, determine changes in the population and identify habitat preferences. The effect of humans on a habitat can also be studied.
It is much more efficient to have a computer system that can store data about many different animals and then automatically, through programming, analyze the animal of concern. Otherwise, a scientist would normally set up a system to recognize one targeted animal, and that is where the information gathering would cease. Also, many different species of birds sing a song that is particular to their species but sounds very similar to the song of another bird species. A computerized system may perform this task much more efficiently and accurately. In the report, the authors said that they were able to achieve accuracy at about 85 percent when testing their system with the British Library Sound Archive. This level of accuracy means the computer system is quite good at sorting out ecosystem sounds.
By Margaret Lutze