APPARATUS, METHOD, AND RECORDING MEDIUM FOR CLUSTERING PHONEME MODELS

• US 20090222266A1
• Filed: 02/26/2009
• Published: 09/03/2009
• Est. Priority Date: 02/29/2008
• Status: Abandoned Application

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Sakai, Masaru
• Application Number:
  US12/393,748
• Publication Number:
  US 20090222266A1
• Time in Patent Office: Days
• US Class Current: 704/254
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering

Abstract

A phoneme model clustering apparatus stores a classification condition of a phoneme context, generates a cluster by performing a clustering of context-dependent phoneme models having different acoustic characteristics of central phoneme for each model having a common central phoneme according to the classification condition, sets a conditional response for each cluster according to acoustic characteristics of context-dependent phoneme models included in the cluster, generates a set of clusters by performing a clustering on clusters according to the conditional response, and outputs the context-dependent phoneme models included in the set of clusters.

Claims

1. An apparatus for clustering phoneme models, comprising:

an input unit configured to input a plurality of context-dependent phoneme models each including a phoneme context indicating a class of an adjacent phoneme and indicating a phoneme model having different acoustic characteristic of a central phoneme according to the phoneme context;

a first storage unit configured to store therein a classification condition of the phoneme context set according to the acoustic characteristic;

a first clustering unit configured to generate a cluster including the context-dependent phoneme models having a common central phoneme and common acoustic characteristic by performing a clustering for each of the context-dependent phoneme models having a common central phoneme according to the classification condition;

a first setting unit configured to set a conditional response indicating a response to each classification condition according to the acoustic characteristic with respect to each cluster according to the acoustic characteristic of the context-dependent phoneme model included in the cluster;

a second clustering unit configured to generate a set of clusters by performing a clustering with respect to a plurality of clusters according to the conditional response corresponding to the classification condition; and

an output unit configured to output the context-dependent phoneme models included in the set of clusters.

2. The apparatus according to claim 1, wherein

the first setting unit includes

a defining unit that defines a virtual context-dependent phoneme model having a virtual phoneme context that represents a set of phoneme contexts of the context-dependent phoneme models included in the cluster and representing a set of context-dependent phoneme models included in the cluster for each cluster, and

a second setting unit that sets a conditional response indicating a response corresponding to each classification condition according to the acoustic characteristic of the set of the phoneme contexts represented by the virtual phoneme context with respect to each of the virtual phoneme contexts, and

the second clustering unit generates a set of virtual context-dependent phoneme models by performing a clustering of the virtual context-dependent phoneme models according to the conditional response corresponding to the classification condition, and

the output unit outputs the set of context-dependent phoneme models defined by the virtual context-dependent phoneme models in units of set of the virtual context-dependent phoneme models.

3. The apparatus according to claim 2, further comprising:

a second storage unit configured to store therein a central phoneme classification condition indicating a classification condition relating to a class of the central phoneme of the virtual context-dependent phoneme models, wherein

the second clustering unit further performs a clustering of a plurality of virtual context-dependent phoneme models according to not only the conditional response corresponding to the classification condition but also the central phoneme classification condition.

4. The apparatus according to claim 3, further comprising:

a third storage unit configured to store therein speech data corresponding to the context-dependent phoneme model; and

a training unit configured to train the acoustic characteristic of the virtual context-dependent phoneme model based on the speech data corresponding to each set of context-dependent phoneme models defined as the virtual context-dependent phoneme model, wherein

the second clustering unit performs a clustering of the set of the virtual context-dependent phoneme models trained by the training unit.

5. The apparatus according to claim 2, wherein

the second setting unit sets a response corresponding to each of positives and negatives with respect to the classification condition for each classification condition as the conditional response according to the acoustic characteristic of each set of the phoneme contexts represented by the virtual phoneme contexts with respect to each of the virtual phoneme contexts.

6. The apparatus according to claim 2, wherein

the second setting unit sets a response corresponding to each of positives, negatives, and indefiniteness with respect to the classification condition for each classification condition as the conditional response according to the acoustic characteristic of each set of the phoneme contexts represented by the virtual phoneme contexts with respect to each of the virtual phoneme contexts.

7. The apparatus according to claim 3, wherein

the second setting unit sets the conditional response corresponding to each classification condition with respect to the virtual phoneme context based on a result of clustering the context-dependent phoneme models obtained by the first clustering unit.

8. A method of clustering phoneme models for a phoneme model clustering apparatus including a first storage unit that stores therein a classification condition of a phoneme context set according to acoustic characteristic, the method comprising:

inputting a plurality of context-dependent phoneme models each including the phoneme context and indicating a phoneme model having different acoustic characteristic of a central phoneme according to the phoneme context;

first clustering including

performing a clustering for each of the context-dependent phoneme models having a common central phoneme according to the classification condition, and

generating a cluster including the context-dependent phoneme models having a common central phoneme and common acoustic characteristic;

first setting including setting a conditional response indicating a response to each classification condition according to the acoustic characteristic with respect to each cluster according to the acoustic characteristic of the context-dependent phoneme model included in the cluster;

second clustering including

performing a clustering with respect to a plurality of clusters according to the conditional response corresponding to the classification condition, and

generating a set of clusters; and

outputting the context-dependent phoneme models included in the set of clusters.

9. The method according to claim 8, wherein

the first setting further includes

defining a virtual context-dependent phoneme model having a virtual phoneme context that represents a set of phoneme contexts of the context-dependent phoneme models included in the cluster and representing a set of context-dependent phoneme models included in the cluster for each cluster, and

second setting including setting a conditional response indicating a response corresponding to each classification condition according to the acoustic characteristic of the set of the phoneme contexts represented by the virtual phoneme context with respect to each of the virtual phoneme contexts, and

the second clustering further includes

performing a clustering of the virtual context-dependent phoneme models according to the conditional response corresponding to the classification condition, and

generating a set of virtual context-dependent phoneme models, and

the outputting includes outputting the set of context-dependent phoneme models defined by the virtual context-dependent phoneme models in units of set of the virtual context-dependent phoneme models.

10. The method according to claim 9, wherein

the phoneme model clustering apparatus further includes a second storage unit that stores therein a central phoneme classification condition indicating a classification condition relating to a class of the central phoneme of the virtual context-dependent phoneme models, and

the second clustering further includes performing a clustering of a plurality of virtual context-dependent phoneme models according to not only the conditional response corresponding to the classification condition but also the central phoneme classification condition.

11. The method according to claim 10, wherein

the phoneme model clustering apparatus further includes

a third storage unit that stores therein speech data corresponding to the context-dependent phoneme model, and

a training unit that trains the acoustic characteristic of the virtual context-dependent phoneme model based on the speech data corresponding to each set of context-dependent phoneme models defined as the virtual context-dependent phoneme model, and

the second clustering further includes performing a clustering of the set of the virtual context-dependent phoneme models trained by the training unit.

12. The method according to claim 9, wherein

the second setting further includes setting a response corresponding to each of positives and negatives with respect to the classification condition for each classification condition as the conditional response according to the acoustic characteristic of each set of the phoneme contexts represented by the virtual phoneme contexts with respect to each of the virtual phoneme contexts.

13. The method according to claim 9, wherein

the second setting further includes setting a response corresponding to each of positives, negatives, and indefiniteness with respect to the classification condition for each classification condition as the conditional response according to the acoustic characteristic of each set of the phoneme contexts represented by the virtual phoneme contexts with respect to each of the virtual phoneme contexts.

14. The method according to claim 10, wherein

the second setting further includes setting the conditional response corresponding to each classification condition with respect to the virtual phoneme context based on a result of clustering the context-dependent phoneme models obtained by the first clustering unit.

15. A computer-readable recording medium that stores therein a computer program for clustering phoneme models for a phoneme model clustering apparatus including a first storage unit that stores therein a classification condition of a phoneme context set according to acoustic characteristic, the computer program when executed causing a computer to execute:

inputting a plurality of context-dependent phoneme models each including the phoneme context and indicating a phoneme model having different acoustic characteristic of a central phoneme according to the phoneme context;

first clustering including

performing a clustering for each of the context-dependent phoneme models having a common central phoneme according to the classification condition, and

generating a cluster including the context-dependent phoneme models having a common central phoneme and common acoustic characteristic;

first setting including setting a conditional response indicating a response to each classification condition according to the acoustic characteristic with respect to each cluster according to the acoustic characteristic of the context-dependent phoneme model included in the cluster;

second clustering including

performing a clustering with respect to a plurality of clusters according to the conditional response corresponding to the classification condition, and

generating a set of clusters; and

outputting the context-dependent phoneme models included in the set of clusters.

16. The computer-readable recording medium according to claim 15, wherein

the first setting further includes

defining a virtual context-dependent phoneme model having a virtual phoneme context that represents a set of phoneme contexts of the context-dependent phoneme models included in the cluster and representing a set of context-dependent phoneme models included in the cluster for each cluster, and

second setting including setting a conditional response indicating a response corresponding to each classification condition according to the acoustic characteristic of the set of the phoneme contexts represented by the virtual phoneme context with respect to each of the virtual phoneme contexts, and

the second clustering further includes

performing a clustering of the virtual context-dependent phoneme models according to the conditional response corresponding to the classification condition, and

generating a set of virtual context-dependent phoneme models, and

the outputting includes outputting the set of context-dependent phoneme models defined by the virtual context-dependent phoneme models in units of set of the virtual context-dependent phoneme models.

17. The computer-readable recording medium according to claim 16, wherein

the phoneme model clustering apparatus further includes a second storage unit that stores therein a central phoneme classification condition indicating a classification condition relating to a class of the central phoneme of the virtual context-dependent phoneme models, and

the second clustering further includes performing a clustering of a plurality of virtual context-dependent phoneme models according to not only the conditional response corresponding to the classification condition but also the central phoneme classification condition.

18. The computer-readable recording medium according to claim 17, wherein

the phoneme model clustering apparatus further includes

a third storage unit that stores therein speech data corresponding to the context-dependent phoneme model, and

a training unit that trains the acoustic characteristic of the virtual context-dependent phoneme model based on the speech data corresponding to each set of context-dependent phoneme models defined as the virtual context-dependent phoneme model, and

the second clustering further includes performing a clustering of the set of the virtual context-dependent phoneme models trained by the training unit.

19. The computer-readable recording medium according to claim 16, wherein

the second setting further includes setting a response corresponding to each of positives and negatives with respect to the classification condition for each classification condition as the conditional response according to the acoustic characteristic of each set of the phoneme contexts represented by the virtual phoneme contexts with respect to each of the virtual phoneme contexts.

20. The computer-readable recording medium according to claim 16, wherein

the second setting further includes setting a response corresponding to each of positives, negatives, and indefiniteness with respect to the classification condition for each classification condition as the conditional response according to the acoustic characteristic of each set of the phoneme contexts represented by the virtual phoneme contexts with respect to each of the virtual phoneme contexts.