An investigation of the effectiveness of a Swedish glide + vowel segment for speaker discrimination

Abstract

Research based on the Australian English diphthong and the British English segment vowel+/r/+vowel has shown that the dynamics of F1, F2, and F3 frequencies can be used for speaker discrimination. This study investigates whether this method can be applied to Swedish. Five male speakers of Swedish were recorded reading a newspaper article that contained seven glide+vowel target tokens. F1, F2, F3, and F4 frequencies were examined at equidistant time-normalized intervals through these tokens. Discriminant analysis based on six predictors from all four formants yielded a classification rate of 88%. This is comparable to earlier research based on a larger predictor set and with recordings that were more controlled by the researcher than the ones used in this study. The present study confirms the power of this approach for speaker discrimination.