Data processing: speech signal processing – linguistics – language – Speech signal processing – Application
Reexamination Certificate
1999-11-22
2002-06-25
Banks-Harold, Marsha D. (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Application
C704S270000
Reexamination Certificate
active
06411933
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates generally to biometric attribute validation and, more particularly, to methods and apparatus for correlating a biometric attribute with one or more biometric attribute production features to validate the production of the biometric attribute.
BACKGROUND OF THE INVENTION
The use of biometric attributes to validate, i.e., identify and/or verify, a person for access to secure applications, systems and/or facilities has increased greatly in the past several years. Some examples of personal biometric attributes that have been used in the validation process include acoustic or speech patterns, fingerprints, retinal scans, to name only a few. Unfortunately, with the increased use of biometric user validation has come increased attempts to deceive the applications, systems and facilities which employ such validation techniques in order to gain unauthorized access. This is especially true in the case of speech biometrics. Some drawbacks of the use of conventional speech biometric techniques in speaker recognition systems for making a validation decision are described below.
When conventional speaker recognition systems are deployed, it is typically assumed that the application manages to verify that the input utterances originate from a live session with a live speaker to enroll, identify or verify. This assumption extends across modalities from text-constrained (e.g., text-dependent, text-prompted, user selected password) to text-independent and speech biometrics. See, for example, U.S. Pat. No. 5,897,616, issued on Apr. 27, 1999, and entitled “Apparatus and Methods for Speaker Verification/Identification/Classification Employing Non-Acoustic and/or Acoustic Models and Databases,” the disclosure of which is incorporated by reference herein.
However, with the evolution of digital signal processing (DSP) of digital recordings, as well as advances in text-to-speech (TTS) technology and, in particular, in voice fonts, one can no longer be certain whether a live person is generating the submitted sounds. Voice fonts are known to have the potential to provide the capability to playback or synthesize speech sounding like a given individual based on some training data obtained from the individual and/or voice transformation functions. Compare, for example, U.S. patent application identified by Ser. No. 08/821,520 (docket no. YO996-247), filed on Mar. 21, 1997, and entitled “Speech Synthesis Based on Pre-Enrolled Tokens,” the disclosure of which is incorporated by reference herein.
The transition from text-dependent speaker recognition (which is known to be especially vulnerable to recordings) to text-prompted speaker recognition provided somewhat of a solution to the problem. However, even text-prompted speaker recognition does not offer any guarantee against a sophisticated TTS or playback signal processing system. The use of user selected passwords is a proposed extension of the text-prompted speaker recognition concept. However, user selected passwords are easily stolen and used to gain unauthorized access.
Text-independent speaker recognition systems are also essentially defenseless against an efficient TTS/voice font system. Only the use of a conventional text-independent system in the background of a transaction or interaction with a human operator makes it somewhat difficult for a speaker to maintain the flow of the transaction if he uses a TTS/playback system to attempt to fool the recognition system. However, with more sophisticated DSP/TTS capabilities (especially on personal digital assistant or PDA devices), there are no more guarantees with respect to user validation.
The concept of speech biometrics adds a knowledge-based dimension to the recognition process. As is known, see U.S. Pat. No. 5,897,616 and S. Maes, “Conversational Computing,” IBM Pervasive Computing Conference, Yorktown Heights, N.Y., June 1999, speech biometric systems use simultaneous content-based recognition (e.g., answers to random questions) and acoustic-based recognition techniques. However, provided that an imposter has the knowledge, a system using speech biometric techniques is essentially defenseless against such an imposter also using sophisticated voice font capabilities. As long as the imposter is able to follow the flow of the dialog, he will likely be able to gain unauthorized access. However, in the case where the speech biometrics system changes multiple non-trivial questions from one access request to another, it is no easy task for an imposter to possess sufficient knowledge and follow the flow of the dialog in order to gain unauthorized access.
Some attempts have been made at detecting the non-linearities of DSP/coders and loudspeakers to detect usage of such devices attempting to fool the system into believing that the person is actually speaking. However, these techniques are not always reliable when dealing with high quality audio equipment or new and unknown equipment.
The use of synchronized biometrics, e.g., face recognition, local mouth geometry recognition, and lip reading synchronized with utterance recognition and speaker recognition has been proposed to guarantee that the user does not use a speaker close to his mouth and lips to generate the utterance. See, for example, U.S. patent application identified by Ser. No. 09/067,829 (docket no. YO997-251), filed on Apr. 28, 1998, and entitled “Method and Apparatus for Recognizing Identity of Individuals Employing Synchronized Biometrics,” the disclosure of which is incorporated by reference herein; as well as the above-incorporated U.S. Pat. No. 5,897,616. Although this adds an additional level of security, it may not be completely fool proof against an effective voice font system combined with good lip sync capabilities.
Accordingly, it is clear that a need exists for techniques that can better guarantee that a speaker physically produced a subject utterance. More generally, a need exists for techniques that can better guarantee that a given biometric attribute has been physically produced by the person offering the biometric attribute as his own.
SUMMARY OF THE INVENTION
The present invention provides methods and apparatus for validating the production of a biometric attribute that better guarantee that a given biometric attribute has been physically produced by the person offering the biometric attribute as his own.
In one broad aspect of the invention, a method of validating production of a biometric attribute allegedly associated with a user comprises the following steps. A first signal is generated representing data associated with the biometric attribute allegedly received in association with the user. A second signal is also generated representing data associated with at least one feature detected in association with the production of the biometric attribute allegedly received from the user. Then, the first signal and the second signal are compared to determine a temporal correlation level between the biometric attribute and the production feature, wherein the validation of the production of the biometric attribute depends on the correlation level. Accordingly, the invention serves to provide substantial assurance that the biometric attribute offered by the user has been physically generated by the user.
In one embodiment, the biometric attribute is a spoken utterance and the production feature is a physiological effect attributable to the production of the spoken utterance alleged to have been produced by the user, e.g., glottal excitation or vibration. The spoken utterance may be decoded and labeled by a speech recognition system to produce the first signal. For example, a sequence of voiced and unvoiced phones is generated from the spoken utterance. Then, a first data value (e.g., a logic value “1”) is assigned to a voiced phone and a second data value (e.g., a logic value “0”) is assigned to an unvoiced phone. Thus, the first signal represents a sequence of such logic values representing the occurrence of voiced and unvoiced phones from the spoken utterance. In an alternati
Maes Stephane Herman
Zweig Geoffrey G.
Banks-Harold Marsha D.
International Business Machines - Corporation
McFadden Susan
Otterstedt Paul J.
Ryan & Mason & Lewis, LLP
LandOfFree
Methods and apparatus for correlating biometric attributes... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Methods and apparatus for correlating biometric attributes..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Methods and apparatus for correlating biometric attributes... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2898448