Methodology for generating enhanced demiphone acoustic models for speech recognition

• US 7,467,086 B2
• Filed: 12/16/2004
• Issued: 12/16/2008
• Est. Priority Date: 12/16/2004
• Status: Expired due to Fees

• Current Assignee: Sony Corporation (Sony Group Corp.), Sony Electronics Inc. (Sony Group Corp.)
• Original Assignee: Sony Corporation (Sony Group Corp.), Sony Electronics Inc. (Sony Group Corp.)
• Inventors: Menendez-Pidal, Xavier, Olorenshaw, Lex S., Abrego, Gustavo Hernandez
• Primary Examiner(s): Edouard; Patrick N
• Assistant Examiner(s): GODBOLD, DOUGLAS
• Application Number:
  US11/013,888
• Publication Number:
  US 20060136209A1
• Time in Patent Office: 1,461 Days
• Field of Search: 704/254, 704/255, 704/256, 7042561-2568
• US Class Current: 704/255
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 2015/022 Demisyllables, biphones or triphones being the recognition units

Abstract

A system and method for effectively performing speech recognition procedures includes enhanced demiphone acoustic models that a speech recognition engine utilizes to perform the speech recognition procedures. The enhanced demiphone acoustic models each have three states that are collectively arranged to form a preceding demiphone and a succeeding demiphone. An acoustic model generator may utilize a decision tree for analyzing speech context information from a training database. The acoustic model generator then effectively configures each of the enhanced demiphone acoustic models as either a succeeding-dominant enhanced demiphone acoustic model or a preceding-dominant enhanced demiphone acoustic model to accurately model speech characteristics.

Claims

1. A system for implementing a speech recognition engine, comprising:

demiphone acoustic models that said speech recognition engine utilizes to perform speech recognition procedures, said demiphone acoustic models each having three states that collectively form a preceding demiphone and a succeeding demiphone; and

an acoustic model generator that analyzes speech context information to configure each of said demiphone acoustic models as either a succeeding-dominant demiphone acoustic model or a preceding-dominant dominant demiphone acoustic model, a contextual dominance for each demiphone state from a given one of said demiphone acoustic models being determined by analyzing predominant contextual information in a triphone decision tree corresponding to said each demiphone state.

2. A system for implementing a speech recognition engine, comprising:

demiphone acoustic models that said speech recognition engine utilizes to perform speech recognition procedures, said demiphone acoustic models each having three states that collectively form a preceding demiphone and a succeeding demiphone; and

an acoustic model generator that analyzes speech context information to configure each of said demiphone acoustic models as either a succeeding-dominant demiphone acoustic model or a preceding-dominant demiphone acoustic model, said speech context information being identified by decision trees that each include a series of questions, said questions each corresponding to a different acoustic speech characteristic, said questions each also being used to identify a contextual dominance characteristic corresponding to said different acoustic speech characteristic.

3. The system of claim 2 wherein said acoustic model generator analyzes all of said questions for a given demiphone model to determine a predominant contextual dominance characteristic for said given demiphone model.

4. The system of claim 2 wherein each of said three states is associated with a different one of said decision trees, each of said three states having a separate contextual dominance characteristic.

5. The system of claim 4 wherein a dominance characteristic of a middle state from said three states determines whether said demiphone acoustic models are configured as either said succeeding-dominant demiphone acoustic model or said preceding-dominant demiphone acoustic model.

6. The system of claim 2 wherein said decision trees are implemented as triphone decision trees that are based upon triphone acoustic models corresponding to said demiphone acoustic models.

7. The system of claim 6 wherein said triphone acoustic models are implemented with three triphone states that each incorporate acoustic contexts from both a preceding phone and a succeeding phone.

8. A method for implementing a speech recognition engine, comprising:

utilizing demiphone acoustic models to perform speech recognition procedures, each of said demiphone acoustic models having three states that collectively form a preceding demiphone and a succeeding demiphone; and

analyzing speech context information with an acoustic model generator to configure each of said demiphone acoustic models as either a succeeding-dominant demiphone acoustic model or a preceding-dominant demiphone acoustic model, a contextual dominance for each demiphone state from a given one of said demiphone acoustic models being determined by analyzing predominant contextual information in a triphone decision tree corresponding to said each demiphone state.

9. A method for implementing a speech recognition engine, comprising:

utilizing demiphone acoustic models to perform speech recognition procedures, each of said demiphone acoustic models having three states that collectively form a preceding demiphone and a succeeding demiphone; and

analyzing speech context information with an acoustic model generator to configure each of said demiphone acoustic models as either a succeeding-dominant demiphone acoustic model or a preceding-dominant demiphone acoustic model, said speech context information being identified by decision trees that each include a series of questions, said questions each corresponding to a different acoustic speech characteristic, said questions each also being used to identify a contextual dominance characteristic corresponding to said different acoustic speech characteristic.

10. The method of claim 9 wherein said acoustic model generator analyzes all of said questions for a given demiphone model to determine a predominant contextual dominance characteristic for said given demiphone model.

11. The method of claim 9 wherein each of said three states is associated with a different one of said decision trees, each of said three states having a separate contextual dominance characteristic.

12. The method of claim 11 wherein a dominance characteristic of a middle state from said three states determines whether said demiphone acoustic models are configured as either said succeeding-dominant demiphone acoustic model or said preceding-dominant demiphone acoustic model.

13. The method of claim 9 wherein said decision trees are implemented as triphone decision trees that are based upon triphone acoustic models corresponding to said demiphone acoustic models.

14. The method of claim 13 wherein said triphone acoustic models are implemented with three triphone states that each incorporate acoustic contexts from both a preceding phone and a succeeding phone.