SPEAKER VERIFICATION USING THE DYNAMIC TIME WARPING

Most of the nowadays commercial authentication schemes are based on two factors: for instance a password and a security ID, or the PIN code and personal data. This is meant to improve the security in comparison with the systems based on one only factor. We present a way to adapt the Dynamic Time Warping approach in order to apply it suitably to a two factors scheme. For feature extraction in speaker verification experiments we used two perceptual approaches: the PLP versus the Mel-scale methodology, with a second purpose of assessing their performance. The participants in the experiment uttered some compulsory sentences in Romanian or their names. Several combinations of vocabulary were tested.