Segmentation and annotation of birdsong with a hybrid recurrent-convolutional neural network
Date:
Songbirds provide an excellent model system for investigating learned motor skills. Because they sing spontaneously, songbirds produce terabytes of behavioral data. Many analyses require labeling the elements of song, called syllables. Labeling syllables by hand consumes many hours, and labeling all the song is often infeasible, preventing full analysis of this data.
This short tutorial was part of the data science workshop organized by David Nicholson in the annual SFN satellite birdsong meeting. The tutorial presented TweetyNet, a deep neural network algorithm I invented for automating the annotation of complex birdsong and developed in collaboration with David. The talk was aimed for songbird neuroscientists of all taxa.
The talk was recorded (Using Sam Sober’s iPhone)