Come mating season, fishes off the California coast sing songs of love in the evenings and before sunrise. They vocalize not so much as lone crooners but in choruses, in some cases loud enough to be heard from land. It's a technique of romance shared by frogs, insects, whales, and other animals when the time is right. For most of these vocal arrangements, the choruses are low-frequency. They're hard to distinguish from the sounds of ships passing in the night among others.

This paper presents a novel supervised approach to detecting the chorus segments in popular music. Traditional approaches to this task are mostly unsupervised, with pipelines designed to target some quality that is assumed to define "chorusness," which usually means seeking the loudest or most frequently repeated sections. We propose to use a convolutional neural network with a multi-task learning objective, which simultaneously fits two temporal activation curves: one indicating "chorusness" as a function of time, and the other the location of the boundaries. We also propose a post-processing method that jointly takes into account the chorus and boundary predictions to produce binary output. In experiments using three datasets, we compare our system to a set of public implementations of other segmentation and chorus-detection algorithms, and find our approach performs significantly better.