Methods and systems for considering information about an expected response when performing speech recognition

• US 8,200,495 B2
• Filed: 01/13/2006
• Issued: 06/12/2012
• Est. Priority Date: 02/04/2005
• Status: Active Grant

• Current Assignee: Vocollect, Inc. (Honeywell International Inc.)
• Original Assignee: Vocollect, Inc. (Honeywell International Inc.)
• Inventors: Braho, Keith, El-Jaroudi, Amro
• Primary Examiner(s): He, Jialong
• Application Number:
  US11/331,649
• Publication Number:
  US 20060178886A1
• Time in Patent Office: 2,342 Days
• Field of Search: 704/256, 7042561-2568, 704/275, 704/270, 704/251
• US Class Current: 704/275
• CPC Class Codes: G10L 15/065 Adaptation
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A speech recognition system receives and analyzes speech input from a user in order to recognize and accept a response from the user. Under certain conditions, information about the response expected from the user may be available. In these situations, the available information about the expected response is used to modify the behavior of the speech recognition system by taking this information into account. The modified behavior of the speech recognition system according to the invention has several embodiments including: comparing the observed speech features to the models of the expected response separately from the usual hypothesis search in order to speed up the recognition system; modifying the usual hypothesis search to emphasize the expected response; updating and adapting the models when the recognized speech matches the expected response to improve the accuracy of the recognition system.

Claims

1. A method for improving a speech recognition system comprising the steps of:

initiating a speech dialog with at least one point in the dialog where there is a grammar of possible responses and a set of at least one expected response;

wherein the set is a subset of the grammar and the set includes the most likely response or responses expected to be uttered by a user at the at least one point in the speech dialog, the set of at least one expected response for the at least one point being known in the speech recognition system before receiving input speech from the user;

progressing through the speech dialog until arriving at the at least one point;

receiving input speech from the user;

generating acoustic features of the input speech using an apparatus with at least one hardware-implemented processor;

comparing the generated input speech acoustic features to acoustic models associated with words in the grammar to generate a hypothesis;

comparing the hypothesis with at least one expected response in the set to determine if the hypothesis matches the at least one expected response in the set;

if the hypothesis matches the at least one expected response in the set, adapting at least one acoustic model corresponding to the matched expected response using the acoustic features of the input speech to use the at least one adapted model with future input speech in the speech recognition system, otherwise, not adapting the at least one acoustic model corresponding to the expected response.

2. The method of claim 1 further comprising generating a confidence factor associated with the hypothesis, wherein the adapting step occurs if the confidence factor exceeds a minimum amount or falls in a specific range.

3. The method of claim 1 wherein the speech recognition uses a match/search algorithm and processes the input speech through a plurality of models with states, and the adaptation includes modifying statistics associated with the model states.

4. The method of claim 3 wherein adaptation includes modifying the state statistics by combining original statistics with new statistics associated with the input speech.

5. The method of claim 1 wherein the set of at least one expected response consists of a single expected response.

6. The method of claim 1 wherein the adapting of at least one acoustic model uses the acoustic features of multiple responses.

7. The method of claim 1 wherein the at least one expected response is obtained by at least one of evaluating an expression or retrieving the expected response from a table or data structure.

8. The method of claim 1 wherein the at least one expected response is dynamically determined.

9. The method of claim 1 wherein the at least one acoustic model is adapted by creating a new acoustic model based upon the input speech.

10. The method of claim 9 wherein the new acoustic model is created by adapting an existing acoustic model using acoustic features of the input speech and existing features.

11. The method of claim 9 wherein the new acoustic model is created by using only information from new acoustic features of the input speech.

12. The method of claim 1 wherein the adapting step occurs only if the comparison of the hypothesis to the at least one expected response also meets additional specific criteria.

13. The method of claim 12 wherein the adapting step occurs if a minimum number of successful comparisons of hypotheses with expected responses have occurred.

14. The method of claim 1 wherein the models are adapted for an individual user by performing adaptation using speech data only from that user.

15. An apparatus for recognizing speech and implementing a speech recognition function, the apparatus comprising:

circuitry for initiating a speech dialog with at least one point in the dialog where there is a grammar of possible responses and a set of at least one expected response and wherein the set is a subset of the grammar and the set includes the most likely response or responses expected to be uttered by a user at the at least one point in the speech dialog, the set of at least one expected response for the at least one point being known in the speech recognition system before receiving input speech from the user;

circuitry operable for receiving input speech from the user for progressing through the speech dialog;

circuitry configured for generating acoustic features of the input speech received from a user;

processing circuitry including a match/search algorithm having acoustic models, the acoustic models including acoustic models that are associated with the set of at least one expected response;

the processing circuitry operable for comparing the generated input speech acoustic features to acoustic models associated with words in the grammar to generate a hypothesis and further operable for comparing the hypothesis with at least one expected response in the set to determine if the hypothesis matches the at least one expected response in the set;

the processing circuitry further operable, if the hypothesis matches the at least one expected response in the set to adapt at least one acoustic model corresponding to the matched expected response using the acoustic features of the input speech to use the at least one adapted model with future input speech in the speech recognition system, otherwise, not adapting the at least one acoustic model corresponding to the expected response.

16. The apparatus of claim 15 further comprising generating a confidence factor associated with the hypothesis, wherein adaptation occurs if the confidence factor exceeds a minimum amount or falls in a specific range.

17. The apparatus of claim 15 wherein the match/search algorithm processes the input speech through a plurality of models with states, and the acoustic models are adapted by modifying statistics associated with the model states.

18. The apparatus of claim 17 wherein the adaptation includes modifying state statistics by combining original statistics with new statistics associated with the input speech.

19. The apparatus of claim 15 wherein the set of at least one expected response consists of is a single expected response.

20. The apparatus of claim 15 wherein the adapting of at least one acoustic model uses the acoustic features of multiple responses.

21. The apparatus of claim 15 wherein the at least one expected response is obtained by at least one of evaluating an expression or retrieving the expected response from a table or data structure.

22. The apparatus of claim 15 wherein the at least one expected response is dynamically determined.

23. The apparatus of claim 15 wherein the at least one acoustic model is adapted by creating a new acoustic model based upon the input speech.

24. The apparatus of claim 23 wherein the new acoustic model is created by adapting an existing acoustic model using acoustic features of the input speech and existing features.

25. The apparatus of claim 23 wherein the new acoustic model is created by using only information from new acoustic features of the input speech.

26. The apparatus of claim 15 wherein the adaptation occurs only if the comparison of the hypothesis to the at least one expected response also meets additional specific criteria.

27. The apparatus of claim 26 wherein the adapting step occurs if a minimum number of successful comparisons of hypotheses with expected responses have occurred.

28. The apparatus of claim 15 wherein the models are adapted for an individual user by performing adaptation using speech data only from that user.