A hose-shaped rescue robot is one of the robots that have been developed for disaster response in times of large-scale disasters such as a massive earthquake. This robot is suitable for entering narrow and dark places covered with rubble in a disaster site and for finding victims inside it. It can transmit ambient sound captured by its builtin microphones to its operator. However, there is a serious problem, that is, the inherent noise of this robot, such as vibration sound or fricative sound, is mixed with the transmitted voice, thereby disturbing the operator's perception of a call for help from a disaster victim. In this paper, we apply the multichannel nonnegative matrix factorization (NMF) with the rank-1 spatial constraint (determined rank-1 MNMF), which was proposed by Kitamura et al., to the reduction of the inherent noise.