Joël Fagot | Searcharxiv

arXiv2023

Detection and classification of vocal productions in large scale audio recordings

We propose an automatic data processing pipeline to extract vocal productions from large-scale natural audio recordings and classify these vocal productions. The pipeline is based on a deep neural network and adresses both issues simultaneously. Though a series of computationel steps (windowing, creation of a noise class, data augmentation, re-sampling, transfer learning, Bayesian optimisation), it automatically trains a neural network without requiring a large sample of labeled data and important computing resources. Our end-to-end methodology can handle noisy recordings made under different recording conditions. We test it on two different natural audio data sets, one from a group of Guinea baboons recorded from a primate research center and one from human babies recorded at home. The pipeline trains a model on 72 and 77 minutes of labeled audio recordings, with an accuracy of 94.58% and 99.76%. It is then used to process 443 and 174 hours of natural continuous recordings and it creates two new databases of 38.8 and 35.2 hours, respectively. We discuss the strengths and limitations of this approach that can be applied to any massive audio recording.

cs.SD↗

arXiv2019

Detecting social (in)stability in primates from their temporal co-presence network

The stability of social relationships is important to animals living in groups, and social network analysis provides a powerful tool to help characterize and understand their (in)stability and the consequences at the group level. However, the use of dynamic social networks is still limited in this context as it requires long-term social data and new analytical tools. Here, we study the dynamic evolution of a group of 29 Guinea baboons using a dataset of automatically collected cognitive tests comprising more than 16M records collected over 3 years. We first built a monthly aggregated temporal network describing the baboon's co-presence in the cognitive testing booths. We then used a null model, considering the heterogeneity in the baboons' activity, to define both positive (association) and negative (avoidance) monthly networks. We tested social balance theory by combining these positive and negative social networks. The results showed that the networks were structurally balanced and that newly created edges also tended to preserve social balance. We investigated several network metrics to gain insights into the individual level and group level social networks long-term temporal evolution. A measure of similarity between successive monthly networks was able to pinpoint periods of stability and instability and to show how some baboons' ego-networks remained stable while others changed radically. Our study confirms the prediction of social balance theory but also shows that large fluctuations in the numbers of triads may limit its applicability to study the dynamic evolution of animal social networks. In contrast, the use of the similarity measure proved to be very versatile and sensitive in detecting relationships' (in)stabilities at different levels. The changes we identified can be linked, at least in some cases, to females changing primary male, as observed in the wild.

physics.soc-ph↗