On Speaker Identification using Coarticulation of Nasal Consonants with Vowels

Abstract : A new approach which used the statistical properties of the nasal spectra was used to quantitatively study the coarticulation of nasal consonants with the vowels in isolated utterances. The mean spectra difference of the nasal followed by front vowels and by back vowels was used as the acoustic measure of coarticulation. The coarticulation of (m) was found to be strongly speaker dependent in particular. This coarticulation, which generally reflects each individual speaker's idiosyncratic characteristics and is not likely to be consciously modified in natural speech, was proved to provide good acoustic clues for speaker identification. Speaker identification was performed using the correlation decision criterion and the results indicated that the coarticulation clues are much better than nasal clues for speaker identification. (Author)