Here’s a scientific breakthrough that amateur and expert bird-watchers would be glad to hear: Bird sounds can now be identified from enormous audio collections.
The study conducted by scientists from the Queen Mary University of London (QMUL) made use of an approach combining feature-learning automatic procedure and classification algorithm for them to develop a system for differentiating which bird exists in a huge dataset.
“Automatic classification of bird sounds is useful when trying to understand how many and what type of birds you might have in one location,” lead author Dr. Dan Stowell says in a statement. Stowell is from the School of Electronic Engineering and Computer Science and Centre for Digital Music at QMUL.
The British Library Sound Archive provided the dataset of sound recordings. The sound recordings were from individual birds and from dawn choruses. The scientists also used online sources for their research. For instance is the Xeno Canto, a Dutch archive.
Stowell says that birdsong has much in common with the language of humans, regardless of having evolved separately.
Stowell claims that man can understand further how the human language progressed as well as the social organization of animal groups from these songbirds.
The scientists’ classification system did good in a public competition through the use of a set of thousands of recordings that have more than 500 species of birds from Brazil. Their system was considered as the "best performing audio only classifier" and as second placer in all entries coming from 10 groups of researchers.
Stowell reveals he is trying to work on techniques that could transcribe all bird sounds in an audio scene that goes beyond identifying who is talking, but also the question of when, to whom and the relationships reflected in the identified sound, and for instance who dominates the conversation.
The scientists are looking at further examining the details in their succeeding project. As for now, their study has proven to be a significant advance in understanding bird sounds.
The study, titled Automatic large-scale classification of bird sounds is strongly improved by unsupervised feature learning, was published in PeerJ.
