Comparative Analysis of Speech Parameters for the Design of Speaker Verification Systems.

摘要

Speaker verification systems are basically composed of three stages: feature extraction, feature processing and comparison of the modified features from speaker voice and from the voice that should be verified. Many features have been used in the first stage, although the current literature has not already shown the best of them. Based on the biometrics hypothesis, which states that each individual has a physical characteristic that distinguishes itself from the others, this paper realized a comparison between 12 classical widely used parameters, in order to investigate the biometrics hypothesis. The obtained results point out those parameters directly correlated to speaker's anatomy which are among the best ones that can be used in the development of speaker verification systems.