Cohort model selection apparatus and method

• US 6,393,397 B1
• Filed: 06/14/1999
• Issued: 05/21/2002
• Est. Priority Date: 06/17/1998
• Status: Expired due to Fees

• Current Assignee: Motorola, Inc. (Motorola Solutions, Inc.)
• Original Assignee: Motorola, Inc. (Motorola Solutions, Inc.)
• Inventors: Choi, Ho Chuen, Zhu, Xiaoyuan, Song, Jianming
• Primary Examiner(s): Dorvil, Richemond
• Application Number:
  US09/332,927
• Time in Patent Office: 1,072 Days
• Field of Search: 704/247, 704/246, 704/239, 704/240, 704/231, 704/250, 704/252, 704/255, 704/256, 704/238
• US Class Current: 704/250
• CPC Class Codes: G10L 17/04 Training, enrolment or model building
• CPC Class Codes: G10L 17/12 Score normalisation

Abstract

An apparatus for selecting a cohort model for use in a speaker verification system includes a model generator (108) for determining a target speaker model (114) from a speech sample collected from the target speaker (106). A cohort selector (110) determines a similarity value between each of a number of predetermined existing speaker models from a model pool (112) and the target speaker model (114) and a dissimilarity value between each of the existing speaker models and any previously selected cohort models (116). An existing speaker model which is most similar to the target speaker model, but most dissimilar to previously chosen cohort models, is then chosen as another cohort model for the target speaker.

Claims

1. A method of selecting at least one cohort model for use in a speaker verification system, the method comprising the steps of:

providing a group of existing speaker models;

receiving target speaker voice utterances from a target speaker;

digitizing at least portions of the received utterances to provide at least one speech sample;

determining a target speaker model from the at least one speech sample;

determining at least one similarity value between each of a plurality of the existing speaker models and the target speaker model; and

utilising the at least one similarity value to select at least one similar existing speaker model as a cohort model for the target speaker.

2. A method of selecting a cohort model according to claim 1, further comprising the steps of:

determining at least one dissimilarity value between at least some of the plurality of the existing speaker models and each cohort model previously selected; and

selecting at least one of the existing speaker models which is similar to the target speaker model and dissimilar to the at least one cohort model previously selected as at least one cohort model for the target speaker.

3. A method of selecting a cohort model according to claim 2, wherein the step of determining at least one dissimilarity value between at least some of the plurality of the existing speaker models and each cohort speaker model previously selected includes the step of, for each existing speaker model and cohort model for which a dissimilarity value is to be determined:

comparing the value of at least one of the parameters of the existing speaker model with the value of at least one corresponding parameter of the cohort model to determine the dissimilarity value.

4. A method of selecting a cohort model according to claim 3, wherein the step of selecting at least one of the existing speaker models which is similar to the target speaker model but dissimilar to the at least one previously selected cohort model involves combining in a predetermined combination the dissimilarity values of two or more previously selected cohort models and selecting at least one of the existing speaker models which has a high similarity value and a high combined dissimilarity value.

5. A method of selecting a cohort model according to claim 4, wherein the predetermined combination is normalised to the similarity values.

6. A method of selecting a cohort model according to claim 4, wherein the predetermined combination is an average.

7. A method of selecting a cohort model according to claim 3, wherein at least one of the parameters is a vector.

8. A method of selecting a cohort model according to claim 7, wherein the vector is a representation of the at least one parameter in n-dimensional space.

9. A method of selecting a cohort model according to claim 8, wherein each parameter of the set of parameters is represented by a vector and the step of determining at least one similarity value between an existing speaker model and the target speaker model includes the steps of:

determining at least two vectors for each existing speaker model and for the target speaker model;

for each existing speaker model vector, determining the distance in the n-dimensional space between that existing speaker model vector and each target speaker model vector and, for each existing speaker model vector, storing whichever distance has a minimum value; and

summing the stored minimum distances to provide the at least one similarity value.

10. A method of selecting a cohort model according to claim 9, wherein the step of determining at least one dissimilarity value between an existing speaker model and a cohort model includes the steps of:

determining at least two vectors for each existing speaker model and for the cohort model;

for each existing speaker model vector, determining the distance in the n-dimensional space between that existing speaker model vector and each cohort model vector and, for each existing speaker model vector, storing whichever distance has a minimum value; and

summing the stored minimum distances to provide the at least one dissimilarity value.

11. A method of selecting a cohort model according to claim 7, wherein the vector represents a frequency response of a time sample of the utterance.

12. A method of selecting a cohort model according to claim 7, wherein the vector is quantised.

13. A method of selecting a cohort model according to claim 1, wherein each speaker model and cohort model comprises a set of parameters, each parameter representing a characteristic of the speech of the speaker, and the step of determining at least one similarity value between each of a plurality of the existing speaker models and the target speaker model includes the step of, for each existing speaker model for which a similarity value is to be determined:

comparing the value of at least one of the parameters of the existing speaker model with the value of at least one corresponding parameter of the target speaker model to determine the similarity value.

14. An apparatus for selecting at least one cohort model for use in a speaker verification system, the apparatus comprising:

a database of existing speaker models;

a receiver for receiving target speaker voice utterances from a target speaker;

a speech digitizer coupled to the receiver to provide at least one speech sample;

a modeller coupled to the speech digitizer for producing and storing a target speaker model from the at least one speech sample;

similarity determining means coupled to the database and the modeller for determining at least one similarity value between each of a plurality of the existing speaker models and the target speaker model;

storage means coupled to the similarity determining means for storing the similarity values;

selection means coupled to the storage means for comparing the similarity values and selecting at least one similar existing speaker model as a cohort model for the target speaker; and

a memory coupled to the selection means for storing the cohort model.

15. An apparatus for selecting a cohort model according to claim 14, further comprising:

dissimilarity determining means coupled to the database and the memory for determining at least one dissimilarity value between at least some of the plurality of the existing speaker models and each cohort model previously selected;

wherein

the selection means is coupled to the dissimilarity determining means for selecting at least one of the existing speaker models which is similar to the target speaker model and dissimilar to the at least one cohort model previously selected as at least another one cohort model for the target speaker.

16. An apparatus for selecting a cohort model according to claim 14, wherein each speaker model and cohort model comprises a set of parameters, each parameter representing a characteristic of the speech of the speaker, and the similarity determining means comprises:

a comparator circuit for comparing the value of at least one of the parameters of the existing speaker model with the value of at least one corresponding parameter of the target speaker model to determine the similarity value.

17. An apparatus for selecting a cohort model according to claim 16, wherein the dissimilarity determining means comprises:

a comparator circuit for comparing the value of at least one of the parameters of the existing speaker model with the value of at least one corresponding parameter of the cohort model to determine the dissimilarity value.

18. An apparatus for selecting a cohort model according to claim 17, wherein the selection means comprises combining means for combining in a predetermined combination the dissimilarity values of two or more previously selected cohort models and the selection means selects at least one of the existing speaker models which has a high similarity value and a high combined dissimilarity value.

19. An apparatus for selecting a cohort model according to claim 18, wherein the combining means includes a normaliser for normalising the predetermined combination to the similarity values.

20. An apparatus for selecting a cohort model according to claim 17, wherein the comparator circuit comprises:

means for storing at least two vectors representing at least two of the parameters in n-dimensional space for each existing speaker model and each previously selected cohort model;

means for determining the distance in the n-dimensional space, for each existing speaker model vector, between that existing speaker model vector and each cohort model vector;

means for storing, for each existing speaker model vector, whichever distance to a cohort model vector has a minimum value; and

means for summing the stored minimum distances to provide the at least one dissimilarity value.

21. An apparatus for selecting a cohort model according to claim 16, wherein the comparator circuit comprises:

means for storing at least two vectors representing at least two of the parameters in n-dimensional space for each existing speaker model and the target speaker model;

means for determining the distance in the n-dimensional space, for each existing speaker model vector, between that existing speaker model vector and each target speaker model vector;

means for storing, for each existing speaker model vector, whichever distance to a target speaker model vector has a minimum value; and

means for summing the stored minimum distances to provide the at least one similarity value.

22. An apparatus comprising:

a speaker model database;

a target speaker voice utterance receiver having an output;

a speech digitizer having a speech sample output and an input coupled to the output of the target speaker voice utterance receiver;

a modeller having an input coupled to the speech sample output, a first memory containing a target speaker model that corresponds to the speech sample output, and an output;

a similarity determiner being operably coupled to the speaker model database and to the output of the modeller, and having an output that provides at least one similarity value between each of a plurality of existing speaker models and the target speaker model;

a second memory operably coupled to the similarity determiner and having the at least one similarity value stored therein;

a selector operably coupled to the second memory and having an output that provides a selected speaker model;

a third memory operably coupled to the output of the selector and having the selected speaker model stored therein as a cohort model.

23. An apparatus according to claim 22, wherein the modeller comprises means for modelling the target speaker model from target speaker voice utterances received from by the target speaker voice utterance receiver and digitised by the speech digitizer.

24. An apparatus according to claim 22, wherein the selector comprises selection means for comparing the stored similarity values and selecting an existing speaker model that has a high similarity value as the cohort model.

25. An apparatus according to claim 22, further comprising a dissimilarity determiner operably coupled to the speaker model database and to the output of the modeller, and having an output that provides at least one dissimilarity value between each of the plurality of existing speaker models and each cohort model previously selected and a second memory operably coupled to the dissimilarity determiner and having the at least one dissimilarity value stored therein.

26. An apparatus according to claim 25, wherein the selector comprises selection means for comparing the stored similarity values and the stored dissimilarity values and selecting an existing speaker model that has a high similarity value and a high dissimilarity value as a cohort model.