Methods and apparatus for flexible speech recognition

• US 7,698,136 B1
• Filed: 01/28/2003
• Issued: 04/13/2010
• Est. Priority Date: 01/28/2003
• Status: Active Grant

• Current Assignee: [24]7.ai, Inc.
• Original Assignee: Voxify, Inc. ([24]7.ai, Inc.)
• Inventors: Shana'a, Adeeb W. M., Nguyen, Patrick T. M., Desai, Amit V.
• Primary Examiner(s): Chawan; Vijay B
• Application Number:
  US10/353,571
• Time in Patent Office: 2,632 Days
• Field of Search: 704/256.5, 704/256, 704/257, 704/241, 704/231, 704/240
• US Class Current: 704/241
• CPC Class Codes: G10L 15/08 Speech classification or search

Abstract

The present invention is directed to a computer implemented method and apparatus for flexibly recognizing meaningful data items within an arbitrary user utterance. According to one example embodiment of the invention, a set of one or more key phrases and a set of one or more filler phrases are defined, probabilities are assigned to the key phrases and/or the filler phrases, and the user utterances is evaluated against the set of key phrases and the set of filler phrases using the probabilities.

Claims

1. A computer implemented method, comprising:

receiving a speech signal;

extracting a sequence of acoustic feature vectors from the speech signal;

identifying a set of one or more words from the speech signal;

identifying a set of one or more phrases from the speech signal, the set of one or more phrases including a sequence of words, wherein the set of one or more phrases comprises key phrases and filler phrases;

assigning a first probability to a phrase of the one or more phrases, the first probability representing a likelihood that the sequence of acoustic feature vectors was produced by the phrase;

assigning a second probability to the phrase of the one or more phrases, the second probability representing a likelihood that the phrase would be received;

assigning a score to the phrase of the one or more phrases, the score calculated using the first probability and the second probability;

and

evaluating the speech signal against the set of phrases using the score.

2. The computer implemented method as recited in claim 1, further comprising providing a prompt to a user, wherein the prompt is designed to elicit a semi-constrained user response.

3. The computer implemented method as recited in claim 1, wherein defining the set of key phrases comprises specifying a key phrase grammar.

4. The computer implemented method as recited in claim 1, wherein the set of filler phrases covers at least a portion of the phonetic spectrum of phrases that can be spoken in a class of applications that share phonetic similarities.

5. The computer implemented method as recited in claim 4, wherein substantially all speech signals in said class of applications can be matched by a phrase in the set of filler phrases.

6. The computer implemented method as recited in claim 4, wherein defining the set of filler phrases comprises specifying a filler grammar.

7. The computer implemented method of claim 1, wherein a weighting factor assigned to a first key phrase can be adjusted to reduce the mis-recognition of the first key phrase as a second key phrase or a filler phrase.

8. The computer implemented method of claim 1, wherein a weighting factor assigned to the filler phrase can be adjusted to reduce the mis-recognition of the filler phrase as a key phrase.

9. The computer implemented method as recited in claim 1, further comprising:

determining the presence of a trigger phrase; and

using the presence of the trigger phrase to affect the recognition of a triggered key phrase.

10. The computer implemented method of claim 9, wherein the triggered key phrase is temporally close to the trigger phrase within the speech signal.

11. The computer implemented method as recited in claim 1, further comprising:

identifying a set of one or more confusion phrases likely to be present in a speech signal; and

determining the presence of a confusion phrase in the speech signal.

12. The computer implemented method of claim 11, wherein the confusion phrase is not a key phrase.

13. The computer implemented method of claim 11, further comprising disregarding a determined confusion phrase.

14. The computer implemented method of claim 11, wherein determining the presence of a confusion phrase includes assigning a weighting factor to a confusion phrase that affects the likelihood of matching a portion of the speech signal to the confusion phrase.

15. The computer implemented method as recited in claim 1, wherein weighting factors for filler phrases and key phrases arc not all the same.

16. The computer implemented method as recited in claim 1, wherein a first weighting factor for a first filler phrase is different than a second weighting factor for a second filler phrase.

17. The computer implemented method as recited in claim 1, wherein a first weighting factor for a first key phrase is different than a second weighting factor for a second key phrase.

18. The computer implemented method as recited in claim 1, wherein evaluating the speech signal further comprises:

specifying a key phrase context-free grammar that produces the key phrases;

specifying a filler context-free grammar that produces the filler phrases;

specifying a sentence context-free grammar using at least the key phrase context-free grammar and the filler context-free grammar; and

applying a speech recognition engine to the speech signal using the sentence context-free grammar.

19. The computer implemented method of claim 18, wherein specifying a sentence context-free grammar includes assigning at least one weighting factor to at least one production rule associated with the sentence context-free grammar.

20. The computer implemented method of claim 19, wherein assigning at least one weighting factor to at least one production rule further includes assigning weighting factors to a key phrase context-free grammar production rule and a filler context-free grammar production rule.

21. The computer implemented method of claim 19, wherein the weighting factors assigned to production rules are not all the same.

22. The computer implemented method of claim 18, wherein applying a speech recognition engine further comprises determining a match for one or more filler phrases from the speech signal.

23. The computer implemented method of claim 18, wherein applying a speech recognition engine further comprises disregarding a recognized filler phrase.

24. A computer implemented method as recited in claim 18, wherein the filler context free grammar produces a sequence of one or more constituent filler words.

25. The computer implemented method of claim 24, wherein the set of constituent filler words is predetermined.

26. The computer implemented method of claim 25, wherein the set of constituent filler words covers the spectrum of phoneme combinations that can be input in an intended class of applications.

27. The computer implemented method of claim 25, wherein the set of constituent filler words approximates the spectrum of phoneme combinations that can be input in an intended class of applications.

28. The computer implemented method of claim 27, wherein the set of constituent filler words contains less than the entire set of phonemes that can be input in the intended class of applications.

29. The computer implemented method of claim 18, wherein the sentence context-free grammar comprises at least one filler context-free grammar preceding the key phrase context-free grammar and at least one filler context-free grammar following the key phrase context-free grammar.

30. The computer implemented method as recited in claim 18, further comprising:

specifying a first context-free grammar to produce pre-data trigger phrases;

specifying a second context-free grammar to produce post-data trigger phrases;

wherein specifying the sentence context-free grammar using at least the key phrase context-free grammar and the filler context-free grammar, optionally includes using the first context-free grammar and the second context-free grammar.

31. The computer implemented method of claim 30, wherein the sentence context-free grammar comprises a combination of a filler context-free grammar and the first context-free grammar preceding the key phrase context-free grammar, and a combination of a filler context-free grammar and the second context-free grammar following the key phrase context-free grammar.

32. The computer implemented method as recited in claim 18, further comprising:

specifying a confusion context-free grammar;

specifying a sentence context-free grammar using at least the key phrase context-free grammar and the filler context-free grammar, and optionally the confusion context-free grammar.

33. The computer implemented method of claim 32, wherein the sentence context-free grammar comprises a combination of a filler context-free grammar and a confusion context-free grammar preceding the key phrase context-free grammar, and a combination of a filler context-free grammar and a confusion context-free grammar following the key phrase context-free grammar.

34. A computer implemented method, comprising:

receiving a speech signal;

extracting a sequence of acoustic feature vectors from the speech signal;

identifying a set of one or more words from the speech signal;

identifying a set of one or more phrases from the speech signal, the set of one or more phrases including a sequence of words, wherein the set of one or more phrases comprises key phrases and filler phrases;

specifying a key phrase context-free grammar that produces the key phrases;

specifying a filler context-free grammar that produces the filler phrases;

assigning a first probability to a phrase of the one or more phrases, the first probability representing a likelihood that the sequence of acoustic feature vectors was produced by the phrase;

assigning a second probability to the phrase of the one or more phrases, the second probability representing a likelihood of receiving the phrase;

assigning a score to the phrase of the one or more phrases, the score calculated using the first probability and the second probability;

defining a sentence context-free grammar using the key phrase context-free grammar, the filler context-free grammars, the scores; and

evaluating the speech signal against the sentence context-free grammar to extract a key phrase from the speech signal.