Methods of creating a corpus of spoken CAPTCHA challenges

• US 9,558,337 B2
• Filed: 07/06/2015
• Issued: 01/31/2017
• Est. Priority Date: 06/23/2008
• Status: Active Grant

• Current Assignee: Knapp Investment Company Limited
• Original Assignee: John Nicholas and Kristin Gross Trust
• Inventors: Gross, John Nicholas
• Primary Examiner(s): Colucci, Michael
• Application Number:
  US14/792,370
• Publication Number:
  US 20150310198A1
• Time in Patent Office: 575 Days
• Field of Search: 704/246, 704/243, 704/244, 704/256.4, 704/273, 705/18, 705/14.69, 379/168, 455/410, 709/225, 726/2, 726/23, 726/7
• US Class Current: 1/1
• 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

Methods of creating a corpus of spoken CAPTCHA challenges are disclosed. The methods involve training a computing system with samples of human voices and computer synthesized voices articulating a set of reference challenge items, scoring the reference challenge items based on differences in articulation between human and machine articulation of each challenge item, and selecting suitable challenge items from among the reference challenge items based on the score.

Claims

1. An automated method for generating Completely Automated Public Turing Test to Tell Computers and Humans Apart (CAPTCHA) challenge items with a CAPTCHA computing system, comprising:

generating a set of candidate CAPTCHA challenge items with a CAPTCHA synthesis engine operating on the CAPTCHA evaluation computing system, such that each candidate CAPTCHA challenge item includes a CAPTCHA question and a target CAPTCHA answer;

presenting a first computer system with said set of candidate CAPTCHA challenge items;

determining, with the first computer system, machine answers to the candidate CAPTCHA challenge items;

with the CAPTCHA computing system and based on said machine answers, measuring an ability of said first computer system to correctly determine said target CAPTCHA answers;

identifying with the CAPTCHA computing system, based on said measuring, at least a first candidate CAPTCHA challenge item of the set of candidate CAPTCHA challenge items for which said machine answers do not meet a threshold of acceptability and selecting the first candidate CAPTCHA challenge item as a selected CAPTCHA challenge item; and

storing said selected CAPTCHA challenge item.

2. The method of claim 1, wherein the CAPTCHA computing system further determines whether the answers to the candidate CAPTCHA challenge items originated from a human or a machine by measuring acoustic characteristics of spoken answers.

3. The method of claim 2, wherein the acoustic characteristics comprise prosodic characteristics.

4. The method of claim 2, wherein the CAPTCHA computing system further determines whether the answers to the candidate CAPTCHA challenge items originated from a human or a machine by processing the answers with a natural language engine and comparing the answers with statistical profiles of answers provided by humans to the same candidate CAPTCHA challenge items.

5. The method of claim 1, wherein said generating further comprises:

generating a prosodic articulation score for each of the candidate CAPTCHA challenge items during said measuring

generating the at least one selected CAPTCHA challenge item from one of the candidate CAPTCHA challenge items if the prosodic articulation score for that CAPTCHA challenge item exceeds a defined threshold.

6. The method of claim 5, further comprising storing the at least one selected CAPTCHA challenge item in a database with the score for such item.

7. The method of claim 1, further comprising:

presenting the at least one selected CAPTCHA challenge item to an entity seeking access to a computer resource;

collecting and processing a response from the entity using the CAPTCHA computing system; and

allowing the entity access to the computer resource based on an identification of the spoken response as having been generated by a human or a machine.

8. The method of claim 1, wherein the set of candidate CAPTCHA challenge items include visual cues intended to elicit a spoken prosodic response.

9. The method of claim 8, wherein the set of candidate CAPTCHA challenge items also include textual cues.

10. The method of claim 1, wherein the set of candidate CAPTCHA challenge items are textual questions.

11. The method of claim 10, wherein the textual questions are open ended.

12. An automated method for generating spoken Completely Automated Public Turing Tests to Tell Computers and Humans Apart (CAPTCHA), comprising:

presenting a computer system with candidate cues for spoken CAPTCHAs;

determining, with the computer system, answers to the candidate cues and generating one or more spoken answers using a text-to-speech (TTS) system;

measuring an ability of a spoken CAPTCHA system to correctly determine whether the answers to the candidate cues are spoken by a human or a machine by measuring acoustical characteristics of the spoken answers;

generating a score for each the candidate cues based on an ability of the spoken CAPTCHA system to determine correctly whether the answer to the candidate cue was spoken by a human or a machine;

choosing at least one spoken CAPTCHA cue from one of the candidate cues if the score for that candidate cue exceeds a defined threshold; and

storing the at least one spoken CAPTCHA cue in a database with the score for the spoken CAPTCHA.

13. The method of claim 12, wherein the acoustic characteristics comprise prosodic characteristics.

14. The method of claim 12, wherein the spoken CAPTCHA system further determines whether the answers to the candidate cues are articulated by a human or a machine by processing the answers with a natural language engine and comparing the answers with statistical profiles of answers provided by humans to the same candidate cues.

15. The method of claim 12, wherein the candidate cues are visual cues intended to elicit a prosodic response.

16. The method of claim 15, wherein the candidate cues also include textual cues.

17. The method of claim 12, wherein the candidate cues are textual questions.

18. The method of claim 17, wherein the textual questions are open ended.