Abstract
This study examines the efficiency of different ways of capturing vowel formants for speaker discrimination. It compares the speaker-discriminating potential of static (F1-F4) and dynamic formant values (F1-F3), and assesses the usefulness of long-term formant distribution (LTF of F1-F4) for discriminating 16 female speakers of Czech.
The results show that dynamic parameters overall perform slightly better than static ones; the most useful parameter of all is static F4. The study found no systematic differences in discriminability of speakers with regards to the position of word stress, i.e. speaker-specific information can be present in stressed as well as unstressed syllables.
LTF seems to be a promising complement to the segment-based methods as it provides an overall picture of the behaviour of each formant. The distribution of all four formants (especially F4) has been shown to have some speaker-discriminating potential, which has been assessed both visually and statistically.