US 5,687,287 A · Speaker verification method and apparatus using mixture decomposition discrimination

Abstract

A new speaker verification method, termed Mixture Decomposition Discrimination (MDD), and a new apparatus for using MDD are presented. MDD takes mixture component score information from a speaker independent recognizer and transmits this information while it is still decomposed as a mixture of component scores that indicate the response of the states of a Hidden Markov Model (HMM) before this information is combined into a single speaker independent recognizer parameter. MDD can be very effective in improving the performance of existing verification methods based on speaker dependent HMMs with cohort normalization because the errors of the two speaker verification methods are very uncorrelated statistically. Experimental results have shown that when MDD is incorporated into a system that also uses speaker dependent HMMs, the resulting hybrid system has its average equal error rate reduced by 46% compared to cohort normalized speaker independent HMMs. MDD is used with a speaker dependent linear discriminator which has relatively low computational and storage requirements. Thus, the increased performance of a hybrid MDD-Cohort Normalized HMM system may be achieved with minimal increase in computational and data storage assets.

Claims13

Dependent claims

An apparatus for speaker verification of an input word string, comprising:

a speaker independent speech recognizer using a first hidden Markov model (HMM), said speaker independent speech recognizer segmenting and recognizing said input word string to obtain an access key to a speaker verification data file of a specific speaker from a plurality of speaker verification data files;
a linear discriminator;
means for providing mixture component score information from internal processes of said speaker independent speech recognizer to said linear discriminator before said mixture component score information is combined into a single quantity;
means for storing a plurality of speaker verification data files, each speaker verification data file containing a true speaker hypothesis or model for each respective speaker is developed by testing against a respective impostor speaker hypothesis or model;
means for accessing a speaker verification data file associated with said access key from said plurality of speaker verification data files and transmitting said accessed data file to said linear discriminator;
said linear discriminator processing said accessed speaker verification data file to generate a plurality of word verification scores;
and
means for determining whether the speech input is from the specific speaker or not according to the plurality of word verification scores.

5 more elements

Dependencies are inferred from the claim text.

Speaker verification method and apparatus using mixture decomposition discrimination

Life of this patent

Speaker verification method and apparatus using mixture decomposition discrimination

Life of this patent

Abstract

Claims13