Shazam for Birds: Building an Offline, Real Time Machine Listening Tool for Bird Vocalizations (via Zoom)

Abstract: In this presentation I will discuss how we built Merlin Sound ID, a machine listening tool for the Merlin Bird ID app ( I will dive into how we collected audio recordings, how we annotated those recordings, and how we trained the model. This project followed a similar trajectory to most image recognition projects, with a few key differences. I will highlight these differences and discuss their impact. I will conclude with an update on our current research to improve Sound ID and our plans for expansions to other regions of the world. 

Bio: Grant Van Horn is a machine learning researcher at the Cornell Lab of Ornithology. His research interests include fine-grained categorization in images, audio, and video, as well as expertise estimation and crowdsourcing. Grant has integrated machine learning models into several popular citizen science apps including Merlin Bird ID, iNaturalist, and Seek. Prior to joining Cornell, Grant was a senior scientist at Amazon Web Services. Grant received his PhD in computer vision from Caltech in 2018, advised by Pietro Perona.