Sub-band cepstral variability within and between speakers under microphone and mobile conditions: A preliminary investigation

Abstract

We describe preliminary results of a sub-band analysis of speaker variability in linear-prediction cepstra measured from 6 native, male speakers’ productions of the 5 Japanese vowels under microphone and mobile conditions. Using a bandselective form of the index-weighted cepstral distance, cepstral F-ratios (between-to-within-speaker variances) were obtained for the entire band from 0 to 3.5 kHz, and for 7 contiguous sub-bands of 500 Hz width each. The resulting profiles of cepstral F-ratios highlight the superiority of certain sub-bands over the full-band and their relative immunity to mobile transmission effects. This is particularly evident in the sub-bands corresponding to the second-formant range for high front vowels.