In recent years, bank transfer fraud has become a serious problem in Japan. In typical cases, the victim is tricked out of money by someone impersonating a family member over the phone. Japan's declining birth rate and aging population have produced a growing number of solitary elderly people who are particularly at risk. To prevent such impersonation, this paper proposes a method of identifying a speaker from normal conversational speech without relying on particular words or phrase. Specifically, power spectra were calculated from waveform data taken from normal conversational speech, using rectangular windows at 1-second intervals, and were analyzed by the RT method. This method was verified by analyzing speech waveforms differing from the unit space. As a result, the feasibility of using this method to identify a speaker from normal conversational speech was confirmed.