Danube Delta Bird Sound Recognition

Bird populations have recorded important changes in number and distribution since the 1970s, due mainly to land management policy and climate changes. So, bird populations monitoring and scientific research became important for environmental preservation. Automatic bird sound recognition starts receiving more attention since the midnineties due to technological advances and the importance of the discipline, as many bird species are often better detectable by their sounds than by vision. The underlying methodology takes advantage of similarities with speaker recognition and has evolved on the tracks of speech-based technologies. With the With the evolution of the deep learning systems, many attempts using these methods were made. The Danube Delta and its close geographical areas are home to 327 bird species and have received over time national and international recognition regarding nature protection. The paper presents the research in the field of bird sound recognition applied to some bird species specific to the Danube Delta. We have tested several settings for the Gaussian Mixture Modelling, Convolutional Neural Networks and Feed Forward Neural Networks. The experiments consider five to fifteen bird sound classes.