Speaker verification system
AssigneeNEC Corporation
InventorMasao Watari
Life of this patent
Estimating the term…
Abstract
Control reference pattern candidates corresponding to a verification reference patterns of a registered speaker are synthesized by connecting unit speech patterns of a plurality of speakers. A plurality of control reference patterns similar to the verification reference pattern are determined from among the control reference pattern candidates. First dissimilarity between an input pattern of a speaker to be verified and the verification reference pattern specified by the registered speaker and second dissimilarity between the input pattern and the control reference patterns specified by the registered speaker are calculated. The speaker to be verified is judged as the registered speaker on the basis of the first and second dissimilarities.
Claims11
A speaker verification system comprising:
- a verification reference pattern memory having addresses which are accessed by speaker identification numbers of registered speakers including a speaker to be verified and are for storing verification reference patterns representative of passwords spoken by said registered speakers;
- a speech segment pattern memory for storing speech segment patterns representative of predetermined speech segments spoken by a plurality of speakers including said speaker to be verified;
- synthesis means for synthesizing control reference pattern candidates representative of said passwords by connecting said speech segment patterns;
- control reference pattern determination means for calculating individual dissimilarities between said verification reference patterns and said control reference pattern candidates and for choosing a predetermined number of control reference patterns from among said control reference pattern candidates based upon said dissimilarities;
- input means for inputting an input identification number and an input pattern representative of one of said passwords that is spoken to represent one of said verification reference patterns by said speaker to be verified, said input identification number being a speaker identification number of said speaker to be verified;
- calculating means for calculating a first dissimilarity between said input pattern and one of said verification reference patterns that is accessed in said verification reference pattern memory by said input identification number, said calculating means also for calculating a second dissimilarity between said input pattern and each of said control reference patterns;
- and
- judgment means for judging whether or not said speaker to be verified is one of said registered speakers based on said first and second dissimilarities.
A speaker verification system comprising:
- a verification reference pattern memory having addresses which are accessed by speaker identification numbers of registered speakers including a speaker to be verified and for storing verification reference patterns representative of passwords spoken by said registered speakers;
- segmentation means for segmenting one of said verification reference patterns into vowel patterns and consonant patterns, said one of said verification reference patterns representing one of said passwords that is spoken by each of said registered speakers;
- a vowel pattern memory for storing th vowel patterns obtained by segmenting said verification reference patterns;
- synthesis means for synthesizing control reference pattern candidates representative of said passwords by connecting said vowel patterns read from said vowel pattern memory with said consonant patterns segmented from said verification reference pattern;
- control reference pattern determination means for calculating individual dissimilarities between said verification reference patterns and said control reference pattern candidates and for choosing a predetermined number of control reference patterns from among said control reference pattern candidates based upon said dissimilarities;
- input means for inputting an input identification number and an input pattern representative of one of said spoken passwords to represent one of said verification reference patterns by said speaker to be verified, said input identification number being one of said speaker identification numbers of said speaker t be verified;
- calculating means for calculating a first dissimilarity between said input pattern and one of said verification reference patterns that is accessed in said verification reference pattern memory by said input identification number, said calculating means also for calculating a second dissimilarity between said input pattern and each of said control reference patterns;
- and
- judgment means for judging whether or not said speaker to be verified is one of said registered speakers based on said first and second dissimilarities.
A speaker verification system comprising:
- a verification reference pattern memory having addresses which are accessed by speaker identification numbers of registered speakers including a speaker to be verified and for storing verification reference patterns representative of passwords spoken by said registered speakers;
- a word pattern memory for storing word patterns representative of a predetermined set of words spoken by a plurality of speakers including said speaker to be verified;
- synthesis means for synthesizing control reference pattern candidates representative of said passwords by connecting the word patterns read out from said word pattern memory;
- control reference pattern determination means for calculating individual dissimilarities between said verification reference patterns and said control reference patterns and for choosing a predetermined number of control reference patterns from among said control reference pattern candidates based upon said dissimilarities;
- input means for inputting an input identification number and an input pattern representative of one of said passwords that is spoken to represent one of said verification patterns by said speaker to be verified, said input identification number being a speaker identification number of said speaker to be verified;
- calculating means for calculating a first dissimilarity between said input pattern and one of said verification reference patterns that is accessed in said verification reference pattern memory by said input identification number, said calculating means also for calculating a second dissimilarity between said input pattern and each of said control reference patterns;
- and
- judgment means for judging whether or not said speaker to be verified is one of said registered speakers based on said first and second dissimilarities.
Dependencies are inferred from the claim text.