SPEECH PROCESSING APPARATUS AND PROGRAM, AND SPEECH PROCESSING METHOD

• US 20090204390A1
• Filed: 06/18/2007
• Published: 08/13/2009
• Est. Priority Date: 06/29/2006
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Yamamoto, Hitoshi, Miki, Kiyokazu
• Granted Patent: US 8,751,226 B2
• Time in Patent Office: Days
• US Class Current: 704/8
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models

Abstract

A speech processing apparatus 101 includes a recognition feature extracting unit 12 that extracts recognition feature information which is a characteristic of a speech recognition result 15 obtained by performing a speech recognition process on an inputted speech from the speech recognition result 15; a language feature extracting unit 11 that extracts language feature information which is a characteristic of a pre-registered language resource 14 from the language resource 14; and a model learning unit 13 that obtains a verification model 16 by a learning process based on the extracted recognition feature information and language feature information.

Claims

1. A speech processing apparatus comprising:

a recognition feature information extracting unit which extracts recognition feature information having a characteristic of recognition result data obtained by performing a speech recognition process on an inputted speech, from said recognition result data;

a language feature information extracting unit which extracts language feature information having a characteristic of a pre-registered language resource from said pre-registered language resource; and

a verification model obtaining unit which obtains a verification model by a learning process based on the extracted recognition feature information and language feature information.

2. The speech processing apparatus as set forth in claim 1, further comprising a selecting unit which selects the type of said recognition feature information and language feature information to be used for said learning process.

3. The speech processing apparatus as set forth in claim 2,

wherein said selecting unit selects the type of said recognition feature information and language feature information to be used for said learning process based on appearance frequency of said recognition result data and language resource.

4. The speech processing apparatus as set forth in claim 1, further comprising a weighted value setting unit which sets a weighted value of said recognition feature information and language feature information to be used for said learning process.

5. The speech processing apparatus as set forth in claim 4,

wherein said weighted value setting unit sets said weighted value according to the type of said recognition feature information and language feature information.

6. The speech processing apparatus as set forth in claim 4,

wherein said weighted value setting unit calculates a weighting sum for each of said recognition feature information and language feature information using said weighted value set to each of said recognition feature information and language feature information, and adopts the calculated weighting sum as a feature value.

7. The speech processing apparatus as set forth in claim 1, further comprising a verification unit which verifies recognition result data newly inputted as a verification target using said verification model.

8. The speech processing apparatus as set forth in claim 1,

wherein said language feature information extracting unit extracts said language feature information based on a use history with respect to a pre-registered language resource.

9. The speech processing apparatus as set forth in claim 1,

wherein said verification model obtaining unit obtains a discriminative model as said verification model.

10. The speech processing apparatus as set forth in claim 9,

wherein said verification model obtaining unit obtains a conditional random field model as said discriminative model.

11. The speech processing apparatus as set forth in claim 1,

wherein said recognition feature information extracting unit extracts an attribute regarding transcription represented by said recognition result data, part of speech of said recognition result data, and pronunciation of said recognition result data.

12. A program enabling a computer to function as the speech processing apparatus as set forth in claim 1.

13. A speech processing method comprising:

extracting recognition feature information having a characteristic of recognition result data obtained by performing a speech recognition process on an inputted speech, from said recognition result data by a speech processing apparatus;

extracting language feature information having a characteristic of a pre-registered language resource from said pre-registered language resource by said speech processing apparatus; and

obtaining a verification model by a learning process based on the extracted recognition feature information and language feature information by said speech processing apparatus.

14. The speech processing method as set forth in claim 13, further comprising selecting the type of recognition feature information and language feature information to be used for said learning process by said speech processing apparatus.

15. The speech processing method as set forth in claim 14,

wherein said type selecting by said speech processing apparatus selects the type of said recognition feature information and language feature information based on appearance frequency of said recognition result data and language resource.

16. The speech processing method as set forth in claim 13, further comprising setting a weighted value of the recognition feature information and language feature information to be used for said learning process by said speech processing apparatus.

17. The speech processing method as set forth in claim 16,

wherein said setting the weighted value by said speech processing apparatus sets said weighted value according to the type of said recognition feature information and language feature information.

18. The speech processing method as set forth in claim 16,

wherein said setting the weighted value by said speech processing apparatus calculates a weighting sum for each of said recognition feature information and language feature information using said weighted value set to each of said recognition feature information and language feature information, and adopts the calculated weighting sum as a feature value.

19. The speech processing method as set forth in claim 13, further comprising verifying recognition result data newly inputted as a verification target using said verification model by the speech processing apparatus.

20. The speech processing method as set forth in claim 13,

wherein said extracting the language feature information by said speech processing apparatus extracts said language feature information based on a use history with respect to a pre-registered language resource.

21. The speech processing method as set forth in claim 13,

wherein said obtaining the verification model by said speech processing apparatus obtains a discriminative model as said verification model.

22. The speech processing method as set forth in claim 21,

wherein said obtaining said verification model by said speech processing apparatus obtains a conditional random field model as said discriminative model.

23. The speech processing method as set forth in claim 13,

wherein said extracting the recognition feature information by said speech processing apparatus extracts an attribute regarding transcription represented by said recognition result data, part of speech of said recognition result data, and pronunciation of said recognition feature information.

24. A speech processing apparatus comprising:

means for extracting recognition feature information having a characteristic of recognition result data obtained by performing a speech recognition process on an inputted speech, from said recognition result data;

means for extracting language feature information having a characteristic of a pre-registered language resource from said pre-registered language resource; and

means for obtaining a verification model by a learning process based on the extracted recognition feature information and language feature information.