Speaker-independent dynamic vocabulary and grammar in speech recognition

• US 5,774,628 A
• Filed: 04/10/1995
• Issued: 06/30/1998
• Est. Priority Date: 04/10/1995
• Status: Expired due to Term

• Current Assignee: Texas Instruments, Inc.
• Original Assignee: Texas Instruments, Inc.
• Inventors: Hemphill, Charles T.
• Primary Examiner(s): Tsang, Fan S.
• Application Number:
  US08/419,226
• Time in Patent Office: 1,177 Days
• Field of Search: 395/2.64, 395/2.65, 395/2.66, 395/2.54, 395/2.52, 395/2.6
• US Class Current: 704/255
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks
• CPC Class Codes: G10L 2015/228 of application context

Abstract

This is an automated speech recognition system, the system comprising: an input device for receiving voice signals; a means for computing the voice signals into stochastic RGDAGs and individual grammars; a search engine that directly processes the stochastic RGDAGs; a means for adding the individual grammars within the RGDAG; a means for replacing the individual grammars within the RGDAG; and a means for deleting the individual grammars within the RGDAG. The system can include a means for adding modified grammars within RGDAGs; a means for linking a terminal symbol from a first grammar to a start symbol in a second grammar; a means for finding a maximum depth of each symbol in the RGDAG; a means for determining if the system requires parse information for the grammar by looking for ancestor symbols; a means for specifing a plurality of RGs within the RGDAGs as starting points for the search engine; and a means for returning a parse through each of a plurality of RGs in the RGDAGs for a recognized utterance. Other devices, systems and methods are also disclosed.

Claims

1. An automated speech recognition system, said system comprising:

a. an input device for receiving voice signals;

b. a means for computing said voice signals into at least one stochastic RGDAG (Regular Grammars Directed Acyclic Graph) and individual grammars;

c. a search engine that directly processes said stochastic RGDAG;

d. a means for adding said individual grammars within said RGDAG;

e. a means for replacing said individual grammars within said RGDAG; and

f. a means for deleting said individual grammars within said RGDAG.

2. The system of claim 1, wherein said system further includes a means for adding modified grammars within an RGDAG.

3. The system of claim 1, wherein said system further includes a means for linking a terminal symbol from a first grammar to a start symbol in a second grammar.

4. The system of claim 1, wherein said system further includes a means for finding a maximum depth of each symbol in said RGDAG.

5. The system of claim 1, wherein said system further includes a means for determining if said system requires parse information for said individual grammars by looking for ancestor symbols.

6. The system of claim 1, wherein said system further includes a means for specifying a plurality of regular Grammars within said RGDAG as starting points for said search engine.

7. The system of claim 1, wherein said system further includes a means for returning a parse through each of a plurality of regular Grammars in said RGDAG for a recognized utterance.

8. A method of speech recognition providing speaker independence and dynamic vocabularies and grammars, said method comprising:

a. inputting device for receiving voice signals;

b. computing said voice signals into at least one stochastic RGDAG (Regular Grammars Directed Acyclic Graph) and individual grammars;

c. directly searching said stochastic RGDAG;

d. adding said individual grammars within said RGDAG;

e. replacing said individual grammars within said RGDAG; and

f. deleting said individual grammars within said RGDAG.

9. The method of claim 8, wherein said method further includes adding modified grammars within RGDAG.

10. The method of claim 8, wherein said method further includes linking a terminal symbol from a first grammar to a start symbol in a second grammar.

11. The method of claim 8, wherein said method further includes finding a maximum depth of each symbol in said RGDAG.

12. The method of claim 8, wherein said method further includes determining if said method requires parse information for said individual grammar by looking for ancestor symbols.

13. The method of claim 8, wherein said method further includes specifying a plurality of regular grammars within said RGDAG as starting points for a search engine.

14. The method of claim 8, wherein said method further includes returning a parse through each of a plurality of regular grammars in said RGDAG for a recognized utterance.

15. A computer automated speech recognition system, said system comprising:

a. an input device for receiving voice signals;

b. a means for computing said voice signals into at least one stochastic RGDAG (Regular Grammars Directed Acyclic Graph) and individual grammars;

c. a search engine that directly processes said stochastic RGDAG;

d. a means for adding said individual grammars within said RGDAG;

e. a means for replacing said individual grammars within said RGDAG; and

f. a means for deleting said individual grammars within said RGDAG.

16. The system of claim 15, wherein said system further includes a means for adding modified grammars within RGDAG.

17. The system of claim 15, wherein said system further includes a means for linking a terminal symbol from a first grammar to a start symbol in a second grammar.

18. The system of claim 15, wherein said system further includes a means for finding a maximum depth of each symbol in said RGDAG.

19. The system of claim 15, wherein said system further includes a means for determining if said system requires parse information for said individual grammars by looking for ancestor symbols.

20. The system of claim 15, wherein said system further includes a means for specifying a plurality of regular grammars within said RGDAG as starting points for said search engine.