Lists of 24 isolated /CVCa/-words were read out by 15 male native speakers of Dutch with ten replications. The middle frames of the steady-state parts of the vowels (/a,i,u/) were examined. These were coded as auditory band spectra containing 19 elements (filter bands) per vector. Discriminant analyses were carried out for every vowel so as to ascertain which filter bands were most effective in discriminating the speakers. It was then studied how these filter bands were related to formant positions. Three hypotheses were evaluated. The hypothesis that speaker-identity is primarily coded in formant band-widths showed the closest match to the results obtained.

Keywords: Acoustic theory of speech; Speaker identification and verification