This paper describes the design of a technique used to extract harmonic spectrum characteristics of human voice. The voice characteristic can be used for a speaker identification process. The cepstral analysis is the most popular method, which uses a Mel-Frequency Cepstral Coefficient vector as unique characteristics of given speaker voice. This method provides only limited reliability. The harmonic spectrum based on fundamental frequency of speaker's voice can extend the characteristic vector by more values. The extended characteristics can provide better reliability of the speaker identification.