In this paper, we present a voice recognition based victim detection system using an Unmanned Aerial Vehicle (UAV). When a large-scale disaster such as a huge earthquake occurs, the building collapses and it is expected that many people will be buried in rubble. Under such circumstances, it is very important to detect victims effectively in order to promptly rescue victims and raise the survival rate. Therefore, in this study, we propose a system for detecting and rescuing victims whose are buried in rubble. In this study, a speaker installed on the UAV makes sounds for victims to react, and victims are detected by capturing their reaction voice. This reaction voice is mixed with drone sound and other sounds in outdoor environment. For this reason, extracting the human voice from this sound mixture is important to detect victims. We first discuss about the development of an UAV with an on-board voice recognition system using a Raspberry pi. Then present about extracting the vic surrounding sounds. We experimentally showed the performance of victim detection using the system.