Data processing: speech signal processing – linguistics – language – Speech signal processing – Application
Reexamination Certificate
1999-05-28
2001-11-06
Dorvil, Richemond (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Application
C704S272000
Reexamination Certificate
active
06314401
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention generally relates to biometric security systems, and more particularly to a mobile verification system to be used by an individual, while in motion, to gain access to a secured area on the basis of that individual's unique voice pattern.
Biometrics
Biometrics involves the use of technology to identify individuals based on certain physiological or behavioral characteristics essentially unique to each individual. Examples of the most common characteristics manipulated through biometrics are finger print identification, voice verification, hand geometry, face geometry, and retinal and/or iris scans. Each of these forms of biometrics produce varying degrees of accurate identification and verification, as well as varying amounts of invasiveness to the individual. The choice of which biometric to implement is dependant primarily upon the level of security needed and the cost which may be incurred in implementation of a system.
In a security application, the biometric characteristic of individuals who will have access to the secure information or area are converted into digital signals and registered in a database stored in a computer's memory. When an individual desires to gain access to the secure information or area, he or she must interact with the particular biometric hardware so as to form a digital pattern of the characteristic and then transmit the characteristic from the hardware to the computer containing the registered database. An algorithm then conducts a comparison of the transmitted signal to those registered patterns stored in memory. If a match is found in the database, the individual is granted access.
Specific examples of existing biometric systems are as follows:
Voice Verification
A voice verification biometric system verifies an individual's identify by comparing an individual's live voice to stored voice samples. This type of system is different from voice identification technology, which is designed to identify the person who is speaking out of a database of known individuals. Furthermore, this system differs from speech recognition technology which is designed to understand the words spoken, regardless of the speaker. The biometric features used in comparing different voiceprints can include the base tones, nasal tones, larynx vibrations, cadence, pitch, tone, frequency, and duration of the voice. Because of the variety of features in the human voice, recordings and reproductions cannot deceive sophisticated voice verification systems.
The first step in using a voice verification biometric is to register the individual to be identified into the system. Most voice verification systems are currently text-dependent (i.e., the individual being identified is required to speak the same word or phrase every time), although there is a long-range trend in moving towards text-independent systems which would verify the individual regardless of what words are spoken. For text-dependent voice verification, the individual must repeat a specified password a number of times so that the system can be trained on different variations of the voice. The person's voiceprint is converted into digital signals and the features are extracted. The features are then fed to a pattern recognition classifier, such as neural network, which trains on the registered individual's features.
During verification, the individual's voice is again converted into a digital signal and the same voice features used in training are extracted. These features are then fed to the trained neural network, which “compares” them with the original features, and either accepts or rejects the individual.
The hardware needed for voice verification require only a computer to run the voice verification software, a microphone, voice filters, and an analog to digital converter to convert the inputted voice into a digital signal. Voice verification is also very simple to use, with the individual only needing to speak a password for verification. Also, it is non-intrusive and does not require the individual to touch or push anything.
It is possible to set up the voice verification system so that the neural network is trained on each new voice sample. This would enable the network to be consistent over time by passively updating the system as an individual's voice pattern changes.
The initial registration can be a time consuming process. It requires the user to repeat a password a number of times so that the neural network can be trained. The training of the neural network is a slow process, but it does not require any user input. While the training of the neural network is slow, the verification of a sample by the neural network can be done very quickly.
Voice verification is a fairly accurate system having a low false acceptance rate (“FAR”) (although a somewhat higher false rejection rate (“FRR”)). The higher FRR is the result of changes in a person's voice. A problem with voice verification is that a sore throat or laryngitis can result in an individual failing to be properly identified.
Fingerprint Identification
Fingerprint identification is commonly used by law enforcement and government agencies. It is one of the best understood and technologically advanced biometric verification methods. Fingerprints are very unique, with only a one in a billion chance of any two persons having the same fingerprint patterns.
Fingerprint identification works by first photographing the “swirls and whorls of each fingertip,” and then converting this photograph into a digital image. A computer then takes this digital image and records the specific ridges, indentations and patterns (known as the characteristic points or minutiae) which are unique to each person. The international standard in fingerprint analysis requires the comparison of eight to twelve identical minutiae. By comparing the minutiae of the known fingerprint with the generated one, a computer can either verify or identify the person.
Fingerprint scanning equipment has an effective FAR of less than one percent and an FRR rating of three percent.
Hand Geometry
Hand geometry identification uses top plan views and side elevational views of a person's hand to identify the individual. A three-dimensional image of the length, width, and height of the hand and fingers are created using an optical scanner. Software then analyzes this image and compares information about the geometry of the hand to the database. Because only nine or ten bytes of information is stored, only a small database is needed to store information on persons in the system.
Hand geometry has the advantage of not having the negative connotation that fingerprinting does, and has been installed in over 7,000 places in America and Europe including airports, day care centers, nuclear research establishments and high-security government facilities. A major disadvantage of hand geometry is that is does not supply the same degree of security as other biometrics. Also, because the entire hand must be analyzed, hand geometry biometric readers are necessarily large devices. Hand injuries can also result in a failure to identify the individual.
Face Verification
Face verification biometric systems use an individual's face to verify his or her identity. Features on the face are used as landmarks for comparison to verify an individual. A neural network is used to compare the features of the individual to be identified with the features recorded in the database. A neural network is used because the data collected from subsequent images may be slightly different, and a neural network's robustness is able to compensate for changes in lighting, angle, size, or alterations in facial hair of glasses.
There are variety of different face verification algorithms. One of the most popular algorithms uses feature vectors from the face image as the biometric tokens. An eye localizer then determines the positions of the eyes in order to find meaningful feature vectors. Feature vectors are then created which contain images of
Abbe Stephen T.
Keiser William
Leonard Scott
Schlereth Barry
Schlereth Fritz
Dorvil Richemond
Hancock & Estabrook, LLP
McGuire George R.
New York State Technology Enterprise Corporation
LandOfFree
Mobile voice verification system does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Mobile voice verification system, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Mobile voice verification system will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2597199