"Who's Calling, Please?" Speaker-Specific Information in Vowels

Abstract

The present study investigates the effectiveness of long-term formant distributions (LTFDs) and by-vowel long-term formants (LTFs) as measures in speaker comparison, by means of a small-scale perception experiment and acoustic analysis of speech samples of four comparable speakers. Overall, LTFD3 is suggested to be a more effective speaker discriminant parameter than LTFD2 in both telephone recorded and directly recorded samples. Both LTFD2 and LTFD3 measurements effectively discriminated between all pairs of speakers, although the occurrence of within-speakers differences between distributions emphasizes the need for further research into the use of LTFD[2,3] as a measure in speaker comparison. Furthermore, the results suggest that analyses of individual vowel spaces in directly recorded samples could be a useful addition to the arsenal of speaker discriminant measures in speaker comparison studies.