On the possibility of speaker discrimination using a glottal pulse phase-related feature

Normalized Relative Delay (NRD) is a phase-related feature that can be extracted from the harmonic structure of a periodic sound, using accurate frequency and phase estimation. We present research results showing that NRD coefficients reflect the phase structure of glottal pulses and possess a speaker discrimination capability. We use both synthetic and natural voiced vowels uttered by children, adult males and females, to illustrate both the shift-invariance property of NRDs, as well as their speaker discrimination potential, using a Fisher-related criterion of data scattering.