Abstract
ABSTRACT
Understanding how animals communicate while moving helps explain social dynamics, but simultaneously tracking vocal and spatial behaviour in freely moving groups remains challenging. Animal-borne audio and GPS recordings can place naturally occurring vocalisations in their spatial context, but continuous recordings generate large datasets that require automated call detection. We used this approach to investigate whether alpaca hums, a poorly characterised vocalisation, contribute to group cohesion. We combined automated detection of hums from continuous collar-borne audio with synchronised GPS data from a partially collared herd to test whether humming is associated with changes in spacing and movement consistent with a contact-call function. We combined deep-learning-assisted detection with manual validation to build a large curated hum corpus from 2,011.1 h of continuous recordings from a partially collared herd. We then tested how humming was associated with spacing and movement at both individual and group levels. Following hum production, distance between focal alpaca and the group shifted toward closer contact relative to matched non-hum periods. Hums were also more likely to occur when the collared alpaca moved faster. These findings support a cohesion-related role of hums and suggest that this call type may help alpaca maintain social contact while moving. More broadly, our study shows how automated call detection combined with movement data can place poorly characterised vocalisations in a behavioural context, narrow their plausible functions, and generate targeted hypotheses even when caller identity remains uncertain and only part of a group can be collared. This approach extends functional research on vocal communication to species whose calls remain poorly understood.