The value of F0, F3, and F4 in identifying disguised voices

Abstract

This study examines the value of F0, F3, and F4 for identifying speakers from a group of ten male and female speakers when the speakers deliberately changed their F0 and/or their articulatory patterns. The speakers were recorded producing a short passage in their normal speaking voice, a lower than normal speaking voice, and using a vocal disguise of their choice. Although F1 and F2 are strongly affected by articulation, the higher formants show relatively little effect of articulation. This relative stability may make them more useful than F1 and F2 for examining speaker identity. While formants reflect vocal tract cavity dimensions, F0 reflects a different aspect of speaker identity, namely the mass and stiffness of the speaker’s vocal folds. Like F3 and F4, F0 shows relatively little effect from articulation. However, speakers can voluntarily shift F0 into a range that is different from what they normally use. Measurements of F0, F3, and F4 were taken on selected stressed vowels from each recording condition. Measurements from the normal voice condition were used to train a discriminant analysis, which then classified vowels from the low F0 and vocal disguise conditions by speaker. Accuracy of the classification results will be presented and discussed.