Abstract: In longitudinal observations of animal groups, the goal is to identify individuals and to reliably detect their interactive behaviors including their vocalizations. However, to reliably extract individual vocalizations from their mixtures and other environmental sounds remains a serious challenge. Promising approaches are multi-modal systems that make use of animal-borne wireless sensors and that exploit the inherent signal redundancy. In this vein, we designed a modular recording system (BirdPark) that yields synchronized data streams and contains a custom software-defined radio receiver. We record pairs of songbirds with multiple cameras and microphones and record their body vibrations with custom low-power frequency-modulated (FM) radio transmitters. Our custom multi-antenna radio demodulation technique increases the signal-to-noise ratio of the received radio signals by 6 dB and reduces the signal loss rate by a factor of 87 to only 0.03% of the recording time compared to standard single-antenna demodulation techniques. Nevertheless, neither a single vibration channel nor a single sound channel is sufficient by itself to signal the complete vocal output of an individual, with each sensor modality missing on average about 3.7% of vocalizations. Our work emphasizes the need for high-quality recording systems and for multi-modal analysis of social behavior.