Speech recognition dictionary compilation assisting system, speech recognition dictionary compilation assisting method and speech recognition dictionary compilation assisting program

• US 8,719,021 B2
• Filed: 02/02/2007
• Issued: 05/06/2014
• Est. Priority Date: 02/23/2006
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Koshinaka, Takafumi
• Primary Examiner(s): He, Jialong
• Application Number:
  US12/280,594
• Publication Number:
  US 20090024392A1
• Time in Patent Office: 2,650 Days
• Field of Search: 704/231, 704/251, 704/256, 704/256.1, 704/256.2
• US Class Current: 704/251
• CPC Class Codes: G10L 15/06 Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice G10L15/14 takes precedence
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

A speech recognition dictionary compilation assisting system can create and update speech recognition dictionary and language models efficiently so as to reduce speech recognition errors by utilizing text data available at a low cost. The system includes speech recognition dictionary storage section 105, language model storage section 106 and acoustic model storage section 107. A virtual speech recognition processing section 102 processes analyzed text data generated by the text analyzing section 101 by making reference to the recognition dictionary, language models and acoustic models so as to generate virtual text data resulted from speech recognition, and compares the virtual text data resulted from speech recognition with the analyzed text data. The update processing section 103 updates the recognition dictionary and language models so as to reduce different point(s) between both sets of text data.

Claims

1. A speech recognition dictionary compilation assisting system, comprising:

a computer processing apparatus; and

a computer-readable storage medium having data stored thereon that includes a dictionary, a language model, an acoustic model, and a speech recognition dictionary compilation assisting program that is executable by the computer processing apparatus to cause the computer processing apparatus to operate as;

a text analysis section that applies morphological analysis to input text data to produce analyzed text data comprising words of the input text data and pronunciation information for each word;

a virtual speech recognition processing section that performs a speech recognition process on said analyzed text data received from the text analysis section by applying the dictionary and the language model to said analyzed text data thereby to generate virtual text data, and that compares a pronunciation information of the virtual text data with the pronunciation information of the analyzed text data to extract and output different points of the analyzed text data and the virtual text data, each different point comprising an element of the analyzed text data and a corresponding element of the virtual text data; and

an update processing section that corrects at least one of the dictionary and the language model in accordance with the different points identified by the virtual speech recognition processing section,

wherein for each different point, the pronunciation information corresponding to a word of the analyzed text data differs from a corresponding pronunciation information of the virtual text data.

2. The speech recognition dictionary compilation assisting system according to claim 1, wherein said virtual speech recognition processing section generates a sequence of feature vectors from the analyzed text data, the sequence of feature vectors comprising acoustic parameters as elements, and performs a virtual speech recognition process on the sequence of feature vectors to generate the virtual text data.

3. The speech recognition dictionary compilation assisting system according to claim 1,

wherein said storage medium stores a table of distances or degrees of resemblance between recognition units, and

wherein said virtual speech recognition processing section generates a sequence of the recognition units from the analyzed text data, and searches in the dictionary and the language model for a string of words that has the least sum of distances or largest sum of the degrees of resemblance to generate the virtual text data.

4. The speech recognition dictionary compilation assisting system according to claim 1,

wherein said storage medium stores a table of distances or degrees of resemblance between elements that constitute a recognition unit, and

wherein said virtual speech recognition processing section generates a sequence of the elements from the analyzed text data, and searches in the dictionary and the language model for a string of words that has the least sum of distances or largest sum of the degrees of resemblance to generate the virtual text data.

5. The speech recognition dictionary compilation assisting system according to claim 1, wherein said update processing section adds a word that has appeared in the analyzed text data to the dictionary in accordance to the different points of the analyzed text data and the virtual text data.

6. The speech recognition dictionary compilation assisting system according to claim 1, wherein said update processing section corrects the language model such that a priority of a word or word string that has appeared in the analyzed text data becomes higher in accordance to the different points of the analyzed text data and the virtual text data.

7. The speech recognition dictionary compilation assisting system according to claim 6, wherein the update processing section controls an amount of changing of the priority in accordance to a frequency of appearance of the word or word string in the analyzed text data and the virtual text data.

8. The speech recognition dictionary compilation assisting system according to claim 1, wherein said update processing section corrects the language model such that a priority of a word or word string that has appeared in the virtual text data resulted from speech recognition becomes lower in accordance to the different points between the analyzed text data and the virtual text data.

9. A speech recognition dictionary compilation assisting method that uses a computer, comprising:

a text analysis step of, by the computer, applying morphological analysis to input text data to produce analyzed text data comprising words of the input text data and pronunciation information for each word;

a step of, by the computer, generating virtual text data from speech recognition from the analyzed text data output from the text analysis step by using a dictionary, a language model, and acoustic models stored in storage devices connected to the computer;

a step of, by the computer, comparing the pronunciation information of the analyzed text data with a pronunciation information of the virtual text data so as to extract and output different points therebetween, each different point comprising an element of the analyzed text data and a corresponding element of the virtual text data; and

an updating process of, by the computer, correcting at least one of the dictionary and the language model in accordance with the different points,

wherein for each different point, the pronunciation information corresponding to a word of the analyzed text data differs from a corresponding pronunciation information of the virtual text data.

10. The speech recognition dictionary compilation assisting method according to claim 9, wherein the computer generates a sequence of feature vectors from the analyzed text data, the sequence of feature vectors comprising acoustic parameters as elements, and virtually performs speech recognition so as to generate the virtual text data.

11. The speech recognition dictionary compilation assisting method according to claim 9, wherein the computer generates a sequence of the recognition units from the analyzed text data in accordance to a table of distances or degrees of resemblance between recognition units, and searches in the dictionary and the language model for a string of words that has the least sum of distances or largest sum of the degrees of resemblance to generate the virtual text data.

12. The speech recognition dictionary compilation assisting method according to claim 9, wherein the computer generates a sequence of the elements from the analyzed text data in accordance to a table of distances or degrees of resemblance between elements that constitute the recognition unit, and searches in the dictionary and the language model for a string of words that has the least sum of distances or largest sum of the degrees of resemblance to generate the virtual text data.

13. The speech recognition dictionary compilation assisting method according to claim 9, wherein the computer adds a word that has appeared in the analyzed text data to the dictionary in accordance to the different points of the analyzed text data and the virtual text data.

14. The speech recognition dictionary compilation assisting method according to claim 9, wherein the computer corrects the language model such that a priority of a word or word string that has appeared in the analyzed text data becomes higher in accordance to the different points between the analyzed text data and the virtual text data.

15. The speech recognition dictionary compilation assisting method according to claim 14, wherein the computer controls an amount of changing of the priority in accordance to a frequency of occurrence of the word or word string in the analyzed text data and the virtual text data.

16. The speech recognition dictionary compilation assisting method according to claim 9, wherein the computer corrects the language model such that a priority of a word or word string that has appeared in the virtual text data resulted from speech recognition becomes lower in accordance to the different points between the analyzed text data and the virtual text data.

17. A program stored on a non-transitory computer-readable storage medium and executable on a computer to cause the computer to operate as a speech recognition dictionary compilation assisting system that performs the following:

a text analysis process that applies morphological analysis to input text data to produce analyzed text data comprising words of the input text data and pronunciation information for each word;

a process that generates virtual text data from speech recognition from the analyzed text data output from the text analysis process by using a dictionary, a language model and acoustic models stored in non-transitory computer-readable storage devices;

a virtual speech recognition process that compares the pronunciation information of the analyzed text data with a pronunciation information of the virtual text data so as to extract and output different points therebetween, each different point comprising an element of the analyzed text data and a corresponding element of the virtual text data; and

an updating process that corrects at least one of the dictionary and the language model in accordance with the different points,

wherein for each different point, the pronunciation information corresponding to a word of the analyzed text data differs from a corresponding pronunciation information of the virtual text data.

18. The program according to claim 17 that causes the computer to generate a sequence of feature vectors from the analyzed text data, the sequence of feature vectors comprising acoustic parameters as elements, and to virtually perform speech recognition so as to generate the virtual text data.

19. The program according to claim 17 that causes the computer to generate a sequence of the recognition units from the analyzed text data in accordance to a table of distances or degrees of resemblance between recognition units, and to search in the dictionary and the language model for a string of words that has the least sum of distances or largest sum of the degrees of resemblance to generate the virtual text data.

20. The program according to claim 17 that causes the computer to generate a sequence of the elements from the analyzed text data in accordance to a table of distances or degrees of resemblance between elements that constitute a recognition unit, and to search in the dictionary and the language model for a string of words that has the least sum of distances or largest sum of the degrees of resemblance to generate the virtual text data.

21. The program according to claim 17 that causes the computer to add a word that has appeared in the analyzed text data to the dictionary in accordance to the different points of the analyzed text data and the virtual text data resulted from speech.

22. The program according to claim 17 that causes the computer to correct the language model such that a priority of a word or word string that has appeared in the analyzed text data becomes higher in accordance to the different points between the analyzed text data and the virtual text data in the updating process.

23. The program according to claim 22 that causes the computer to control an amount of changing of the priority in accordance to a frequency of occurrence of the word or word string in the analyzed text data and the virtual text data in the updating process.

24. The program according to claim 17 that causes the computer to correct the language model such that a priority of a word or word string that has appeared in the virtual text data resulted from speech recognition becomes lower in accordance to the different points between the analyzed text data and the virtual text data in the updating process.