Deep Convolutional Network for Animal Sound Classification and Source Attribution using Dual Audio Recordings

Oikarinen, T. P.; Srinivasan, K.; Meisner, O.; Hyman, J. B.; Parmar, S.; Desimone, R.; Landman, R.; Feng, G.

doi:10.1101/437004

Home/Archive/bioRxiv

bioRxiv · 10.1101/437004

Deep Convolutional Network for Animal Sound Classification and Source Attribution using Dual Audio Recordings

Oikarinen, T. P.·Srinivasan, K.·Meisner, O.·Hyman, J. B.·Parmar, S.·Desimone, R.·Landman, R.·Feng, G.

View original source ↗Export metadata ↓

Abstract

We introduce an end-to-end feedforward convolutional neural network that is able to reliably classify the source and type of animal calls in a noisy environment using two streams of audio data after being trained on a dataset of modest size and imperfect labels. The data consists of audio recordings from captive marmoset monkeys housed in pairs, with several other cages nearby. Our network can classify both the call type and which animal made it with a single pass through a single network using raw spectrogram images as input. The network vastly increases data analysis capacity for researchers interested in studying marmoset vocalizations, and allows data collection in the home cage, in group housed animals.

Source connections

Published article ↗

Explore related subjects

animal behavior and cognition Preprint

Keep this discovery

BibTeX RIS

Oikarinen, T. P., Srinivasan, K., Meisner, O., Hyman, J. B., Parmar, S., Desimone, R., Landman, R., Feng, G.. 2018-10-07. Deep Convolutional Network for Animal Sound Classification and Source Attribution using Dual Audio Recordings. https://doi.org/10.1101/437004

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Who rests with whom? Sex composition and group demography shape resting associations in free-ranging dogs

Free-ranging dogs frequently rest near conspecifics, but the demographic factors structuring their resting associations remain poorly understood. We quantified dyadic resting associations in 26 free-ranging dog groups in West Bengal, India, observed between 2019 and 2023. Association strength was estimated from scan based resting co-occurrences using the Half-Weight Index. We tested whether dyadic association strength varied with dyad sex composition, dyad life stage composition, group size, and group sex ratio using a generalised additive model for location, scale and shape that accounted for group identity and repeated occurrence of individuals across dyads. Male-male dyads had lower association strengths than female-female dyads, whereas mixed-sex dyads did not differ from female-female dyads. Association strength decreased with increasing group size but increased as the male-to-female ratio within the group increased, while life-stage composition had no detectable effect. Individual level network metrics, including strength, reach, clustering coefficient, affinity, and eigenvector centrality, did not vary with sex or season. Mixed-sex pairs were also frequently represented among the strongest dyadic associations within groups. These findings indicate that resting associations in free-ranging dogs vary with dyad sex composition and group demography. Further opportunity-controlled analyses are required to determine whether the prominence of mixed-sex dyads reflects preferential association rather than group composition alone.

animal behavior and cognition↗

bioRxiv2026

Exploring rhythmic and melodic preferences in budgerigars: Individual and possible sex-related variation

Budgerigars (Melopsittacus undulatus) are vocal-learning birds with well-developed auditory abilities, but how they behaviorally evaluate melodic and rhythmic structure in sound sequences remains unclear. We examined whether budgerigars show preferences for these acoustic features and whether such preferences differ between the sexes. Three male and three female budgerigars were presented with four 8-s sound sequences in a preference apparatus: Simple (no pitch or temporal variation), Melody (pitch variation only), Rhythm (temporal variation only), and Complex (both pitch and temporal variation). Preference was quantified as the time spent in the area associated with each stimulus. No statistically significant differences among the four stimuli were detected within individuals. However, effect-size estimates indicated that two females spent more time with sequences containing rhythmic structure, whereas males showed no consistent preference related to either melodic or rhythmic components. Multidimensional scaling further suggested greater separation among stimulus conditions in females than in males. Consistent with this pattern, condition differentiation indices were higher in all three females than in all three males, although the sex difference was not statistically significant. These results suggest a possible sex-related difference in how budgerigars behaviorally weight temporal structure, with females showing greater differentiation among auditory sequence types under the present testing conditions.

animal behavior and cognition↗

bioRxiv2026

Copulation calls indicate fertility but do not reflect female mate competition in wild Guinea baboons

Across different modalities, signals play a core role in attracting mates and influencing mating success. In several non-human primate species, females produce calls during mating that are thought to promote male competition over receptive females. The extent to which social system characteristics modulate the function of copulation calls remains less clear. We studied copulation calls in wild Guinea baboons (Papio papio), who live in a multilevel society structured around units in which females associate and mate almost exclusively with a single male. We hypothesised that females use copulation calls as an indirect form of mate competition, with competition increasing in larger units. In addition, we hypothesised that females are more likely to mate again after calling. We analysed 6116 copulations between 2014 and 2025, involving 99 reproductively active females and 78 subadult and adult males. Females produced copulation calls in 72.7% of copulations, with large inter-individual variation. Neither unit size nor its interaction with the female's swelling size or the presence of simultaneously receptive females affected the probability of calling. A survival analysis with a subset of the data (2353 copulations) revealed no effect of calling on the latency to the next mating. Our results render the hypothesis that female Guinea baboons use calls in indirect mate competition unlikely. Yet, the probability of calling varied with sexual swelling size, suggesting that calls signal female fertility. Possibly, Guinea baboon copulation calls represent an evolutionary remnant, no longer under selective pressure, and can be considered index signals of female fertility.

animal behavior and cognition↗