Investigating speaker features from very short speech records

A procedure is presented that is capable of accurately extracting various speaker features, and is of particular value for analyzing records containing single words and shorter durations of speech. By taking advantage of the fast convergence of adaptive filtering, the approach is capable of modeling the nonstationarities due to both the vocal tract and vocal cord dynamics. This procedure is quite simple, requires no manual intervention, and is particularly unique because it derives both the vocal tract and glottal signal estimates directly from the time-varying filter coefficients rather than the prediction error signal. Several glottal signals are derived using this procedure, and are plotted to demonstrate the kind of glottal characteristics obtained therein. Finally, in order to provide a more quantitative performance measure, the procedure is used in a simple automatic speaker identity verification application