People & Nature Lab Seminar : The Search for Squawk by Tom Denton (Google DeepMind)

Where and When: 22nd November 14.00-15.00 at UCL East, Room 512, Marshgate

Abstract: Bioacoustic monitoring can collect thousands or even millions of hours of data, but it has historically been difficult to make sense to these massive datasets. However, we have found that embeddings from bird song classification can help surface novel signals efficiently, allowing experts to use their data to answer questions we could only dream about before. In this talk, we’ll explain how the system works, and look at case studies including threatened animal monitoring, population monitoring, and ecosystem health assessment. We’ll also highlight joint work with Ben Williams (UCL) applying bird song embeddings to coral reef monitoring.

Bio: Tom Denton leads Perch, a bioacoustics project in Google Deepmind. In bioacoustics, he has worked on unsupervised source separation, tracking forest health through bioacoustic signals, and co-hosts the yearly BirdClef Kaggle competition. He has also worked on neural voice compression, math education in Sub-Saharan Africa, and plays a pretty mean ukulele. He obtained a PhD in mathematics from the University of California, Davis, in 2011 with a specialization in algebraic combinatorics.  

A flavour of Tom’s work can also be found on his Google Scholar page https://scholar.google.com/citations?user=M1nVuUgAAAAJ&hl=en.

If you are not able to attend in person, please contact Raphaele to get the Teams link.