A source extraction method is proposed for dual-unmanned aerial vehicle (UAV) audition, where each UAV is equipped with a microphone array. While deploying multiple UAVs inevitably introduces extra noise, it enables the capture of richer spatial information over wider areas, leading to improved extraction performance and broader spatial coverage compared to single-UAV setups. This paper extends the blind source extraction method---rank-constrained spatial covariance matrix estimation---into a decentralised framework for dual-UAV audition setups. Specifically, while the time-varying noise variances and source spatial covariance matrices are estimated for each array separately, the target variance is jointly optimised across both arrays, with a scaling factor introduced to account for inter-array level differences. A multichannel Wiener filter is then computed using this refined target variance for source extraction. Experimental results demonstrate that the proposed method achieves performance comparable to or better than the single-UAV baseline in a designated dual-UAV setup.