Apparatus, method, and program for clustering phonemic models

• US 8,112,277 B2
• Filed: 09/22/2008
• Issued: 02/07/2012
• Est. Priority Date: 10/24/2007
• Status: Expired due to Fees

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Sakai, Masaru
• Primary Examiner(s): AZAD, ABUL K
• Application Number:
  US12/234,729
• Publication Number:
  US 20090177472A1
• Time in Patent Office: 1,233 Days
• Field of Search: 704242-245, 704256-2562
• US Class Current: 704/245
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering

Abstract

A node initializing unit generates a root node including inputted phonemic models. A candidate generating unit generates candidates of a pair of child sets by partitioning a set of phonemic models included in a node having no child node into two. A candidate deleting unit deletes candidates each including only phonemic models attached with determination information indicating that at least one of the child sets has a small amount of speech data for training. A similarity calculating unit calculates a sum of similarities among the phonemic models included in the child sets. A candidate selecting unit selects one of the candidates having a largest sum. A node generating unit generates two nodes including the two child sets included in the selected candidate, respectively. A clustering unit clusters the phonemic models in units of phonemic model sets each included in a node.

Claims

1. A clustering apparatus comprising:

an input unit configured to input at least one phonemic model attached with determination information indicating a small amount of speech data for training, and at least one phonemic model not attached with the determination information;

a node initializing unit configured to generate a node including the inputted phonemic models as a root node of a tree structure;

a candidate generating unit configured to generate candidates of a pair of child sets for a node having no child node among nodes in the tree structure, by partitioning a set of phonemic models included in the node into two;

a candidate deleting unit configured to delete a candidate including the two child sets at least one of which includes only phonemic models attached with the determination information, from the generated candidates;

a similarity calculating unit configured to calculate a similarity among the phonemic models included in each of the two child sets included in each of the candidates other than the deleted candidates, and calculates a sum of the similarities calculated for the child sets;

a candidate selecting unit configured to select one of the candidates having a largest calculated sum;

a node generating unit configured to generate two nodes including two child sets included in the selected candidate, respectively, as child nodes of the node that is a generation source of the selected candidate; and

a clustering unit configured to cluster the phonemic models in units of phonemic model sets included in nodes of the tree structure.

2. The apparatus according to claim 1, wherein

the input unit inputs at least one phonemic model attached with language determination information indicating a language having a small amount of the speech data as the determination information, and at least one phonemic model not attached with the language determination information, and

the candidate deleting unit deletes a candidate including the two child sets at least one of which includes only phonemic models attached with the language determination information, from the generated candidates.

3. The apparatus according to claim 1, wherein the candidate generating units generates the candidates for nodes each having no child node and including phonemic models attached with the determination information, among the nodes of the tree structure.

4. The apparatus according to claim 1, wherein each of the phonemic models corresponds to each state of a Hidden Markov Model (HMM) that is obtained by modeling a transition relation among acoustical features of phonemes.

5. A clustering method comprising:

inputting at least one phonemic model attached with determination information indicating a small amount of speech data for training, and at least one phonemic model not attached with the determination information;

generating a node including the inputted phonemic models as a root node of a tree structure;

generating candidates of a pair of child sets for a node having no child node among nodes in the tree structure, by partitioning a set of phonemic models included in the node into two;

deleting a candidate including the two child sets at least one of which includes only phonemic models attached with the determination information, from the generated candidates;

calculating a similarity among the phonemic models included in each of the two child sets included in each of the candidates other than the deleted candidates, and calculating a sum of the similarities calculated for the child sets;

selecting one of the candidates having a largest calculated sum;

generating two nodes including two child sets included in the selected candidate, respectively, as child nodes of the node that is a generation source of the selected candidate; and

clustering the phonemic models in units of phonemic model sets included in nodes of the tree structure.

6. A program stored on a nontransitory computer readable medium including programmed instructions for clustering phonemic models, wherein the instructions, when executed by a computer, cause the computer to perform:

inputting at least one phonemic model attached with determination information indicating a small amount of speech data for training, and at least one phonemic model not attached with the determination information;

generating a node including the inputted phonemic models as a root node of a tree structure;

generating candidates of a pair of child sets for a node having no child node among nodes in the tree structure, by partitioning a set of phonemic models included in the node into two;

deleting a candidate including the two child sets at least one of which includes only phonemic models attached with the determination information, from the generated candidates;

calculating a similarity among the phonemic models included in each of the two child sets included in each of the candidates other than the deleted candidates, and

calculating a sum of the similarities calculated for the child sets;

selecting one of the candidates having a largest calculated sum;

generating two nodes including two child sets included in the selected candidate, respectively, as child nodes of the node that is a generation source of the selected candidate; and

clustering the phonemic models in units of phonemic model sets included in nodes of the tree structure.