The comparison of “Idiot´s Bayes” and multivariate kernel-density in forensic speaker identification using Chinese vowel /a/

The discriminant performance of likelihood ratios based on “Idiot´s Bayes” approach and multivariate kernel density were examined on the speech of 21 male Chinese speakers using telephone recordings. The parameters used in this paper are formant central-frequencies, extracted from the Chinese vowel /a/. Experimental results show that the “Idiot´s Bayes” approach provides a little stronger support for the same-speaker hypothesis and multivariate kernel density approach provides much stronger support for the different-speaker hypothesis; they both have good performance in Chinese vowel /a/.