Verbal utterance rejection using a labeller with grammatical constraints

• US 7,031,923 B1
• Filed: 03/06/2000
• Issued: 04/18/2006
• Est. Priority Date: 03/06/2000
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Navratil, Jiri, Maes, Stephane H., Chaudhari, Upendra V.
• Primary Examiner(s): ARMSTRONG, ANGELA A
• Application Number:
  US09/519,327
• Time in Patent Office: 2,234 Days
• Field of Search: 704/273, 704/270, 704/275, 704/272, 704/235, 704/256, 704/251, 704/255
• US Class Current: 704/273
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

Methods and apparatus for verifying spoken passwords and sentences with small decoder complexity requirements. A proposed method permits the verbal verification of a spoken password sentence without computationally extensive large-vocabulary decoding. A decoder preferably uses target baseforms (representing the original content to be verified) together with a special set of competing simplified baseforms that may be easily constructed using finite-state grammars (FSG).

Claims

1. An apparatus for verifying verbal utterances, said apparatus comprising:

a target password sentence generator which generates at least one target password sentence;

an acceptance arrangement which compares text based on a verbal utterance to at least one target password sentence and which accepts or rejects the verbal utterance based on its comparison to the at least one target password sentence;

a decoder which transforms a verbal utterance into decoded text for being compared to the at least one target password sentence in said acceptance arrangement, wherein said transformation comprises transcribing the verbal utterance when the verbal utterance comprises the target password; and

a finite state grammar generator which generates a finite state grammar to be employed by said decoder;

said decoder being adapted to transform a verbal utterance into decoded text via employing the finite state grammar to modify the verbal utterance.

2. The apparatus according to claim 1, wherein said decoder is adapted to transform a verbal utterance into decoded text via employing the finite state grammar to undertake a forced alignment of the verbal utterance.

3. The apparatus according to claim 1, wherein said finite state grammar generator includes a first generator for generating a first part of a finite state grammar and a second generator for generating a second part of a finite state grammar.

4. The apparatus according to claim 3, wherein:

said first generator is adapted to generate a first part of a finite state grammar that includes at least a target password sentence; and

said second generator is adapted to generate a second part of a finite state grammar that includes competing decoding paths of the target password sentence.

5. The apparatus according to claim 1, wherein said target password sentence generator is adapted to accept prompted text corresponding to at least one password.

6. The apparatus according to claim 1, wherein said target password sentence generator is adapted to generate automatically at least one password as a baseform that is derived from an acoustic enrollment.

7. The apparatus according to claim 1, wherein said acceptance arrangement is adapted to derive a match score based on comparing text based on a verbal utterance to the at least one target password sentence and to base acceptance or rejection of the verbal utterance on the match score.

8. The apparatus according to claim 7, wherein said acceptance arrangement is adapted to derive a match score via dynamic pattern matching.

9. The apparatus according to claim 7 wherein said acceptance arrangement is adapted to derive a match score via a combination of dynamic pattern matching and acoustic scoring.

10. A method of verifying verbal utterances, said method comprising:

generating at least one target password sentence;

comparing text based on a verbal utterance to at least one target password sentence;

accepting or rejecting the verbal utterance based on its comparison to the at least one target password sentence;

transforming the verbal utterance into decoded text;

said comparing step comprising comparing the decoded text to the at least one target password sentence; and

generating a finite state grammar;

said transforming step comprising transforming the verbal utterance into decoded text via employing the finite state grammar to modify the verbal utterance, wherein said modification comprises transcribing the verbal utterance when the verbal utterance comprises the target password.

11. The method according to claim 10, wherein said step of transforming the verbal utterance into decoded text comprises employing the finite state grammar to undertake a forced alignment of the verbal utterance.

12. The method according to claim 10, wherein said step of generating a finite state grammar comprises generating a first part of the finite state grammar and a second part of the finite state grammar.

13. The method according to claim 12, wherein:

said step of generating the first part of the finite state grammar comprises the inclusion of the at least one target password sentence; and

said step of generating the second part of the finite state grammar comprises the inclusion of competing decoding paths of the at least one target password sentence.

14. The method according to claim 10, wherein said step of generating at least one target password sentence comprises accepting prompted text.

15. The method according to claim 10, wherein said step of generating at least one target password sentence comprises automatically generating a baseform derived from an acoustic enrollment.

16. The method according to claim 10, wherein said step of accepting or rejecting comprises deriving a match score based on comparing text based on the verbal utterance to the at least one target password sentence and to base acceptance or rejection of the verbal utterance on the match score.

17. The method according to claim 16, wherein said step of deriving a match score comprises deriving a match score via dynamic pattern matching.

18. The method according to claim 16, wherein said step of deriving a match score comprises deriving a match score via a combination of dynamic pattern matching and acoustic scoring.

19. A program storage device readable by machine, tangibly embodying a program of instructions executable by the machine to perform method steps for verifying verbal utterances, said method comprising:

generating at least one target password sentence;

comparing text based on a verbal utterance to at least one target password sentence;

accepting or rejecting the verbal utterance based on its comparison to the at least one target password sentence;

transforming the verbal utterance into decoded text,

said comparing step comprising comparing the decoded text to the at least one target password sentence; and

generating a finite state grammar;

said transforming step comprising transforming the verbal utterance into decoded text via employing the finite state grammar to modify the verbal utterance, wherein said modification comprises transcribing the verbal utterance when the verbal utterance comprises the target password.