Apes, Acoustics, and Algorithms: Machine Learning to Identify Chimpanzee Callers

Project Description:

Like humans, chimpanzees exhibit a fluid social organisation. Individuals from the same community never congregate and instead, form temporally and spatially ephemeral sub-groups, sometimes going days or weeks without seeing each other. When they do re-group, they rely on their long call – the pant hoot – to coordinate reunions. Whilst there is evidence of referentially specific calls (e.g. a “travel call” and “food call”), there is also evidence to suggest that chimpanzees exhibit signature calls, i.e. those that are individually specific. However, this has not been assessed using machine learning, a powerful and increasingly common tool in signal processing and pattern recognition.
Identifying individually specific call patterns has implications for behaviour, conservation, and our reconstructions of human evolution. First, caller-specificity allows researchers to simultaneously monitor the movements of multiple chimpanzees across space and time, something otherwise logistically prohibitive. This will, in turn, provide insight into the nature of ‘coordinated movement.’ Second, by demonstrating that we can identify individuals remotely, a new census method will be established, that of counting callers to assess population abundance using passive acoustic monitoring. Already density estimates are possible with PAM, but they rely on indirect metrics. Such metrics are critical in conservation when establishing baseline figures to monitor population dynamics. Finally, chimpanzees in western Tanzania from where the vocalisation data will originate live in a mosaic woodland habitat that closely resembles reconstructions of those of early human ancestors. Investigation into behavioural adaptations of low-density, wind ranging apes has implications for our modelling of early hominin behaviour.

Policy Impact of Research:

The research has implications to influence not only great ape survey methods, but also those for any other loud calling species, e.g. wild dogs, lion, gibbons, etc.

Research Relating to this Project:

1. Automatic acoustic identification of individuals in multiple species: improving identification across recording conditions
2. Machine Learning Algorithms for Automatic Classification of Marmoset Vocalizations
3. Individuality in a Long‐Range Vocalization of Wild Chimpanzees