Real-time Birdcall Classification (Machine Learning)

We trained an AI system to identify bird calls in an aviary at the San Diego Zoo Safari Park, showing images on a display in real-time whenever particular bird species vocalized.

Using a custom-built acoustic annotation tool, hundreds of positive and negative recordings for each species were used to construct training datasets. Utilizing convolutional neural networks, the resulting prototypes could reliably classify human speech, crested oropendolas, and greater sunbitterns.

This method would be useful for automatic bird classification in field settings as well as for assistive identification systems in wildlife sanctuaries. Furthermore, such systems could be used for real-time security alerts for acoustic events like chainsaws or vehicles in restricted wildlife areas.

with G. Cano (SD Zoo).