Identity-resolved vocal exchanges in multi-marmoset settings

Acoustic communication is tightly coupled to collective behaviour and social network structure in many animal societies. Yet continuous, identity-resolved recordings of multiple concurrent signalers are challenging. We evaluated a commercial acoustic camera for marmoset (Callithrix jacchus) vocal behavior across controlled bench tests and naturalistic family interactions. Spectral characterization showed the system is closest to neutral and most sensitive in 5-10 kHz, supporting reliable detection without correction in this range. Static localization accuracy was between 1-3 cm across the experimental enclosure. In simultaneous two-source tests, separation depended strongly on geometry and intensity level: very small spacing (20 cm) produced a single merged hotspot, frame-wise co-representation of multiple sources was infrequent and biased by sub-decibel level asymmetries. In natural 15-min family recordings (N=519 calls), the camera detected 96.5% of calls, among detected events, caller assignment was 97.6% correct (overall 94.2%). Ground-truth playback confirmed near-ceiling detection ([~]100%) and high assignment with two emitters ([~]98%), with reduced attribution in crowded four-emitter layouts. Together, these results indicate that acoustic imaging can deliver high detection and accurate caller attribution for typical marmoset vocalizations, with primary limitations arising from spatial crowding and small inter-source level differences. Importantly, this limitation can be mitigated in the native software by sequentially silencing suspected sound sources. The approach thus provides a practical and reliable path toward continuous, identity-resolved vocal monitoring in freely behaving primates.