System and Method for Generating Challenge Items for CAPTCHAs

• US 20090319271A1
• Filed: 06/15/2009
• Published: 12/24/2009
• Est. Priority Date: 06/23/2008
• Status: Active Grant

• Current Assignee: Knapp Investment Company Limited
• Original Assignee: John Nicholas and Kristin Gross Trust U/A/D April 13 2010
• Inventors: GROSS, JOHN NICHOLAS
• Granted Patent: US 8,380,503 B2
• Time in Patent Office: Days
• US Class Current: 704/246
• CPC Class Codes: G06F 21/32 using biometric data, e.g. fingerprints, iris scans or voiceprints
• CPC Class Codes: G10L 13/027 Concept to speech synthesisers; Generation of natural phrases from machine-based concepts generation of parameters for speech synthesis out of text G10L13/08
• CPC Class Codes: G10L 13/08 Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• CPC Class Codes: G10L 17/02 Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction
• CPC Class Codes: G10L 17/04 Training, enrolment or model building
• CPC Class Codes: G10L 17/06 Decision making techniques; Pattern matching strategies
• CPC Class Codes: G10L 17/22 Interactive procedures; Man-machine interfaces
• CPC Class Codes: G10L 17/26 Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices
• CPC Class Codes: H04L 63/102 Entity profiles
• CPC Class Codes: H04L 63/123 received data contents, e.g. message integrity
• CPC Class Codes: H04M 2203/2027 Live party detection

Abstract

Challenge items for an audible based electronic challenge system are generated using a variety of techniques to identify optimal candidates. The challenge items are intended for use in a computing system that discriminates between humans and text to speech (TTS) system.

Claims

1. A method embodied in a computer readable medium for generating challenge data to be used for accessing data and/or resources of an electronic computing system comprising:

(a) automatically generating a candidate challenge sentence from a first set of words and phrases using the computing system;

(b) automatically generating at least one first utterance from a first machine text to speech system for said candidate challenge sentence using the computing system, said at least one first utterance including first acoustical characteristics;

(c) automatically generating at least one second utterance from a human speaker for said candidate challenge sentence using the computing system, said at least one second utterance including second acoustical characteristics;

(d) automatically determining a difference in said first and second acoustical characteristics using the computing system to determine a challenge sentence acoustic score for said candidate challenge sentence;

(e) automatically storing said at least one first utterance and said at least one second utterance and said candidate challenge sentence in a challenge item database using the computing system for use by an utterance based challenge system when said candidate challenge sentence acoustic score exceeds a target threshold.

2. The method of claim 1, wherein acoustical features of said first utterance and said second utterance are measured, and regions of greatest difference are identified to be used as discriminators.

3. The method of claim 1 wherein differences in articulation are measured and scored on a diphone basis.

4. The method of claim 3 wherein said diphones are sorted according to their difference between human and machine articulation.

5. The method of claim 3 further including a step:

selecting words and phrases from a corpus of text in accordance with a diphone score.

6. The method of claim 1 further including a step:

compiling a machine articulation statistical language model based on a plurality of first utterances.

7. The method of claim 1 further including a step:

compiling a dialog of multiple challenge sentences based on questions and expected answers to be provided by an entity.

8. The method of claim 1 further including a step:

generating a challenge scenario comprised of said candidate challenge sentence and one or more visual and/or text cues, which challenge scenario is stored in said database.

9. The method of claim 1 further including a step:

processing said words and phrases with a natural language engine to generate said candidate challenge sentence.

10. The method of claim 1 further including a step:

annotating said candidate challenge sentence with prosodic elements to generate an expected prosodic pronunciation of said words and phrases.

11. The method of claim 8 further including a step:

annotating said candidate challenge sentence with first prosodic elements found in said at least one first utterance and with second prosodic elements found in said least one second utterance.

12. The method of claim 11 further including a step:

determining a difference in said first and second prosodic elements to determine a challenge sentence prosodic score for said candidate challenge sentence.

13. The method of claim 8 further including a step:

annotating said candidate challenge sentence with first content elements for said visual and/or text cues found in said at least one first utterance and with second content elements for said visual and/or text cues found in said least one second utterance.

14. The method of claim 12 further including a step:

determining a difference in said first and second content elements to determine a challenge sentence content score for said candidate challenge sentence.

15. The method of claim 1 wherein a time required by the human speaker to generate said first utterance is measured.

16. The method of claim 1 wherein a time required by the machine speaker to generate said second utterance is measured.

17. The method of claim 1 where said first set of words and phrases are automatically extracted from an electronic corpus.

18. The method of claim 1 wherein said first set of words and phrases include homographs.

19. The method of claim 1 further including a step:

processing input speech by an entity using said challenge item database to distinguish between a human and a machine synthesized voice.

20. A method embodied in a computer readable medium of selecting challenge data to be used for accessing data and/or resources of a computing system comprising:

(a) providing a first set of diphones using the computing system;

(b) generating an articulation score using the computing system based on a machine text to speech (TTS) system articulation of each of said first set of diphones;

(c) selecting challenge text using the computing system to be used in an utterance based challenge system based on said articulation scores.

21. The method of claim 20 further including a step:

processing input speech by an entity using said challenge item database to distinguish between a human and a machine synthesized voice.

22. A method embodied in a computer readable medium of selecting challenge data to be used for accessing data and/or resources of a computing system comprising:

a) selecting a candidate challenge item which can include text words and/or visual images;

b) measuring first acoustical characteristics of a computer synthesized utterance when articulating challenge content associated with said candidate challenge item;

c) measuring second acoustical characteristics of a human utterance when articulating said challenge content;

d) generating a challenge item score based on measuring a difference in said first and second acoustical characteristics;

e) designating said candidate challenge item as a reference challenge item when said challenge item score exceeds a target threshold.

23. The method of claim 22 further including a step:

processing input speech by an entity using said reference challenge item to distinguish between a human and a machine synthesized voice.

24. The method of claim 22, wherein said challenge item score is also based on one or more topics associated with said text words and/or visual images and which are identified and measured in said computer synthesized utterance and human utterance respectively.

25. The method of claim 22, wherein said challenge item score is also based on prosodic elements associated with said text words and/or visual images and which are identified and measured in said computer synthesized utterance and human utterance respectively.

26. The method of claim 22, wherein said challenge item score is also based on a collaborative filtering score generated by measuring responses to a sequence of two or more of said candidate challenge items identified in said in said computer synthesized utterances and human utterances respectively.

27. The method of claim 26 wherein said collaborative filtering score is derived by identifying at least a first reference correlation between at least a first reference text descriptor for a first challenge item and a second reference text descriptor for a second challenge item, including a probability that a human reviewer identifying said first reference text descriptor when presented with said first challenge item also provides said second reference text descriptor when presented with said second challenge item, or vice-versa.

28. The method of claim 26 wherein said collaborative filtering score is derived by identifying at least a first reference correlation between a first challenge item presented in incomplete form, and a predicted response for completing said challenge item.

29. A method embodied in a computer readable medium of selecting challenge data to be used for accessing data and/or resources of a computing system comprising:

a) defining a plurality of demographic groups, said demographic groups being based on age, sex and/or domicile;

b) providing a plurality of CAPTCHA (Completely Automatic Public Turing Test To Tell Humans And Computers Apart) challenge items consisting of a combination of images and solicited utterances with the computing system;

c) for each of said challenge items using the computing system to compare a first reference response of a machine entity and a second reference response provided by a representative of said demographic group;

d) for each demographic group selecting an optimal set of CAPTCHA challenge items determined by the computing system to yield the greatest response difference over said machine entity.

30. The method of claim 29 further including a step:

identifying a demographic group for an entity; and

processing input speech by said entity using at least one of said optimal set of CAPTCHA challenge items for said demographic group to determine if it is a human or a machine synthesized voice.

31. A system for identifying challenge data to be used for accessing data and/or resources of a computing system comprising:

one or more software routines embodied in a computer readable medium adapted to cause the computing system to;

(a) automatically generate a candidate challenge sentence from a first set of words and phrases;

(b) automatically generate at least one first utterance from a first machine text to speech system for said candidate challenge sentence, said at least one first utterance including first acoustical characteristics;

(c) automatically generate at least one second utterance from a human speaker for said candidate challenge sentence, said at least one second utterance including second acoustical characteristics;

(d) automatically determine a difference in said first and second acoustical characteristics to determine a challenge sentence acoustic score for said candidate challenge sentence;

(e) automatically store said at least one first utterance and said at least one second utterance and said candidate challenge sentence in a challenge item database for use by an utterance based challenge system when said candidate challenge sentence acoustic score exceeds a target threshold.

32. A system for identifying challenge data to be used for accessing data and/or resources of a computing system comprising:

one or more software routines embodied in a computer readable medium adapted to cause the computing system to;

(a) provide a first set of diphones;

(b) generate an articulation score based on a machine text to speech (TTS) system articulation of each of said first set of diphones;

(c) select challenge text to be used in an utterance based challenge system based on said articulation scores.

33. A system for identifying challenge data to be used for accessing data and/or resources of a computing system comprising:

one or more software routines embodied in a computer readable medium adapted to cause the computing system to;

a) select a candidate challenge item which can include text words and/or visual images;

b) measure first acoustical characteristics of a computer synthesized utterance when articulating challenge content associated with said candidate challenge item;

c) measure second acoustical characteristics of a human utterance when articulating said challenge content;

d) generate a challenge item score based on measuring a difference in said first and second acoustical characteristics;

e) designate said candidate challenge item as a reference challenge item when said challenge item score exceeds a target threshold.

34. A system for identifying challenge data to be used for accessing data and/or resources of a computing system comprising:

one or more software routines embodied in a computer readable medium adapted to cause the computing system to;

a) define a plurality of demographic groups, said demographic groups being based on age, sex and/or domicile;

b) provide a plurality of CAPTCHA (Completely Automatic Public Turing Test To Tell Humans And Computers Apart) challenge items consisting of a combination of images and solicited utterances;

c) for each of said challenge items compare a first reference response of a machine entity and a second reference response provided by a representative of said demographic group;

d) for each demographic group select an optimal set of CAPTCHA challenge items determined by the computing system to yield the greatest response difference over said machine entity.