Welcome Your IET account
Zebra finches on a branch

Finch horny-o-meter highlights ‘bottom-up’ approach to AI research

Image credit: Dreamstime

McGill University researchers have developed an algorithm for identifying specific features of zebra finch songs sung by males, which may underlie the distinction between birdsong for courtship purposes and birdsong for other purposes.

Like many other animals, male zebra finches adjust their vocal signals depending on their audience. They sing the same sequence of 'syllables' (a motif) during courtship interactions with female finches as when singing alone, but with some subtle modifications to acoustic structure. These modifications are undetectable to humans and, for a long time, it was not clear whether these modifications could also be picked up by female zebra finches.

A team of Canadian researchers demonstrated through behavioural studies that female zebra finches are very capable of discriminating between motifs recorded in amorous versus platonic settings. The females started at chance levels (accurate 50 per cent of the time) but significantly improved their ability to correctly classify motifs with training.

The researchers confirmed that the birds were using acoustic properties to distinguish between courtship and non-courtship motifs by testing their responses to new samples not included in the original training sets (excluding the possibility that they had memorised the motifs).

Next, the researchers developed an algorithm capable of extracting features from male zebra finch vocalisations that may mark the distinction between a short phrase sung during courtship and the same phrase sung in a non-courtship context. They took an expansive “bottom-up” approach to their work, using highly comparative time-series analyses (HCTSA): a novel toolbox combining approximately 1,000 analytical techniques to compute thousands of individual 'features' from time-series data. They extracted over 5,000 features from recordings of songs from 17 males and trained an algorithm to use these features to distinguish between courtship and non-courtship phrases.

The algorithm was capable of uncovering features that may be key to song perception, including some which had not previously been identified. It also made predictions about the distinction capabilities of female zebra finches which aligned well with the behavioural experiments.

According to the researchers, their findings highlight the potential for bottom-up approaches to revealing acoustic features important for communication.

“As vocal communicators ourselves, we have a tendency to focus on aspects of communication signals that are salient to us,” said Professor Sarah Woolley, the McGill University neurobiologist who led the study. “Using our bottom-up approach, we identified features that might never have been on our radar.”

Next, Woolley and her colleagues plan to test whether manipulating the features they discovered may alter how female zebra finches perceive the song samples, as well as evaluating how well their findings may be generalised to courtship vocalisations in other species.

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles

Info Message

We use cookies to give you the best online experience. Please let us know if you agree to all of these cookies.


Learn more about IET cookies and how to control them