DISTRIBUTED RECOGNITON SYSTEM HAVING MULTIPLE PROMPT-SPECIFIC AND RESPONSE-SPECIFIC SPEECH RECOGNIZERS

• US 20010016813A1
• Filed: 12/29/1998
• Published: 08/23/2001
• Est. Priority Date: 12/29/1998
• Status: Active Grant

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: BROWN, DEBORAH W., ROSINSKI, RICHARD R., GOLDBERG, RANDY G., WETZEL, WILLIAM R.
• Granted Patent: US 6,377,922 B2
• Time in Patent Office: Days
• US Class Current: 704/231
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

A speech recognition system recognizes spoken utterances received as a speech signal from a user. A prompt for requesting a spoken utterance from the user is assigned a response identifier which indicates at least one of a plurality of speech recognizers to best recognize a particular type of spoken utterance. The system includes a processor for receiving the speech signal from the user in response to the prompt. The processor also directs the speech signal to the at least one speech recognizer indicated by the response identifier. The speech recognizer generates a plurality of spoken utterance choices from the speech signal and a probability associated with each of the plurality of choices. At least one of the spoken utterance choices is selected based on the associated probabilities.

Claims

1. A method for recognizing a spoken utterance of a user comprising the steps of:

assigning at least one of a plurality of speech recognizers to a prompt for requesting a spoken utterance;

presenting a user with said assigned prompt;

receiving said spoken utterance from said user in response to said prompt and generating a speech signal from said received spoken utterance; and

directing said speech signal to the at least one assigned speech recognizer;

2. The method of

claim 1

, further comprising the steps of:

generating a plurality of spoken utterance choices from the speech signal and a probability associated with each of said plurality of choices; and

selecting the spoken utterance based on the associated probabilities.

3. The method of

claim 2

, wherein the step of selecting the spoken utterance comprises the steps of:

presenting a highest probability spoken utterance choice to the user; and

determining whether the presented highest spoken utterance choice is correct.

4. The method of

claim 3

, wherein the step of selecting the spoken utterance comprises the steps of:

presenting a next highest probability spoken utterance choice to the user;

determining whether the presented next highest spoken utterance choice is correct; and

repeating the steps of presenting a next highest probability and determining whether the presented next highest probability is correct if it is determined at the step of determining whether the presented next highest probability is correct, that the next highest spoken utterance choice is not correct.

5. The method of

claim 2

, wherein the step of generating a plurality of spoken utterance choices comprises using a hidden Markov Model to generate N-best choices.

6. The method of

claim 2

, wherein the step of generating a plurality of spoken utterance choices comprises the steps of:

recognizing a first spoken utterance; and

post-processing the first spoken utterance to generate the plurality of spoken utterance choices;

7. The method of

claim 1

, wherein the step of assigning comprises assigning at least one speech recognizer that provides a highest reliability for recognizing a particular type of spoken utterance.

8. The method of

claim 7

, wherein said particular type of spoken utterance includes numbers.

9. The method of

claim 7

, wherein said particular type of spoken utterance includes letters of the alphabet.

10. The method of

claim 7

, wherein said particular type of spoken utterance includes a limited vocabulary.

11. A speech recognition system for recognizing a spoken utterance comprising:

a processor for presenting a user with a prompt; and

a plurality of speech recognizers coupled to said processor, wherein each speech recognizer generates a plurality of spoken utterance choices from a speech signal and a probability for each spoken utterance choice;

wherein said prompt is assigned at least one speech recognizer that provides a highest reliability for recognizing a particular type of spoken utterance.

12. The speech recognition system of

claim 11

, further comprising a database coupled to said processor, said database having stored therein a plurality of prompts assigned to at least one speech recognizer that provides a highest reliability for recognizing a particular type of spoken utterance.

13. The speech recognition system of

claim 11

, further comprising an input/output interface to receive a speech signal from a user.

14. The speech recognition system of

claim 13

, wherein said input/output interface is coupled through a network and a telephone.

15. The speech recognition system of

claim 13

, wherein said input/output interface comprises a microphone.

16. The speech recognition system of

claim 14

, wherein said network is the Internet.

17. The speech recognition system of

claim 14

, wherein said network is a public switched telephone network.

18. The speech recognition system of

claim 14

, wherein each speech recognizer generates a plurality of spoken utterance choices using a hidden Markov Model to generate N-best choices.

19. The speech recognition system of

claim 11

, wherein each prompt is assigned at least one speech recognizer that provides a highest reliability for recognizing a particular type of spoken utterance.

20. The speech recognition system of

claim 19

, wherein said particular type of spoken utterance includes numbers.

21. The speech recognition system of

claim 19

, wherein said particular type of spoken utterance includes letters of the alphabet.

22. The speech recognition system of

claim 19

, wherein said particular type of spoken utterance includes a limited vocabulary.

23. A method for recognizing a plurality of utterances spoken by a user, the method comprising the steps of:

generating a first prompt requesting a first utterance;

receiving said first utterance;

generating a first speech signal based on said received first utterance;

directing said first speech signal to a first speech recognizer assigned to recognize speech received in response to said first prompt;

generating a second prompt requesting a second utterance;

receiving said second utterance;

generating a second speech signal based on said received second utterance; and

directing said second speech signal to a second speech recognizer assigned to recognize speech received in response to said second prompt.