Fundamental frequency vector for a speaker identification system

Abstract

Speaking Fundamental Frequency (SFF) is a measure of the rate of opening and closing of the vocal cords. Among other things, it reveals speaker-dependent patterns and is important in the task of speaker identification. In fact, experiments conducted at the Institute for Advanced Study of the Communication Processes (IASCP) at the University of Florida have demonstrated its role in the identification process. Many techniques are available for obtaining SFF measurements; this research involved three of them. They include: (1) Cepstrum SFF extraction (CEP), (2) the Computerized Speech Lab (CSL) and (3) a Fundamental Frequency Indicator (FFI) . The Cepstrum method is a software approach designed to obtain SFF, CSL is a multifunction speech signal processor that includes a function for SFF extraction and FFI is a specially designed device used to extract an accurate measure of SFF on a continuous basis. These techniques were compared by means of an identification procedure involving three types of transmission conditions and a large population of male subjects. The results showed that the best way to obtain the SFF vector is the FFI method which employed alone yielded an 80 per cent overall correct identification rate for high quality recordings, 76 per cent for noise-added recording conditions and 90 per cent over the telephone line.