In a sound source detection apparatus that detects a sound source of a detection subject on the basis of collected sounds, sounds are collected by at least one sound collector, an autocorrelation between sounds collected in time series by the sound collector is calculated, and a determination as to whether or not the sound source of the detection subject exists is made on the basis of the autocorrelation. More particularly, sounds are preferably collected respectively by two or more sound collectors such that the existence of the sound source of the detection subject is determined by determining whether or not the autocorrelations of the sounds collected by the two or more sound collectors satisfy a predetermined condition. By using the autocorrelation to detect the sound source of the detection subject in this manner, high robustness against an S/N ratio is exhibited, leading to an improved detection performance when detecting the sound source of the detection subject.