Learnt formant modulation via upper vocal tract movements in a marine mammal

Formants are resonance frequencies shaped by the upper vocal tract. Across vertebrates, they have a key role in acoustic communication. Humans show fine-grained control of formant modulation; little is known about such control in other species. We trained a male harbor seal (Phoca vitulina) to modify a baseline vocalization via upper vocal tract movements, resulting in a conditioned vocalization. The seal also had to maintain the baseline throughout the experiment. Over 150 days, we collected 455 baseline and 640 conditioned vocalizations. We extracted the first three formants (F1, F2, F3) and quantified their within-vocalization modulation using complementary measures. First, the times series of formant contours showed that F1 and F3 were similar at the start of the experiment but diverged over time. Second, coefficients of variation for F1 decreased, with higher values in the conditioned vocalizations at the end of the experiment. Third, F1 and F3 modulation depth, measuring frequency variation between adjacent time points, was higher in conditioned vocalizations. Fourth, formants' spectral entropy decreased only in conditioned vocalizations, indicating a more predictable energy distribution over the course of the experiment. Fifth, machine learning techniques confirmed that vocal types became more distinguishable at the end than at the start of the experiment solely based on modulation parameters. Our findings show that (1) an adult harbor seal can be trained to open and close its mouth while phonating; (2) the resulting vocal output contains modulated formants and (3) differs from a baseline vocalization; (4) both vocal types' formant features changed over time as an interconnected system.

The online version contains supplementary material available at 10.1007/s44338-025-00145-z.