CRNN-BASED SPLASH AUDIO EVENT DETECTION FOR FISH MONITORING

Abstract

Monitoring migratory fish species provides a good indicator for rivers' health. Migratory fish as alosa (Alosa fallax also known as twait shad), swims up the rivers to reproduce if the dam infrastructure allows it. During spawning, some species of alosa produce during a few seconds a characteristic splash sound, that enables them to perceive their presence. Stakeholders involved in the rehabilitation of freshwater ecosystems rely on staff to aurally count the bulls during spring nights and then estimate the alosa population at different sites. In order to reduce the human costs and expand the scope of the analysis, we propose a deep learning approach for audio event detection from dozens of GB of audio files recorded from the riverbanks. An automatic detection system consisting of a Recurrent Convolutional Neural Network (CRNN) is presented. Encouraging results enable us to aim for an automated implementation on sites.