EP1129447A1 - Procede de comparaison d'emission de paroles en vue d'un controle de securite - Google Patents

Procede de comparaison d'emission de paroles en vue d'un controle de securite

Info

Publication number
EP1129447A1
EP1129447A1 EP98957754A EP98957754A EP1129447A1 EP 1129447 A1 EP1129447 A1 EP 1129447A1 EP 98957754 A EP98957754 A EP 98957754A EP 98957754 A EP98957754 A EP 98957754A EP 1129447 A1 EP1129447 A1 EP 1129447A1
Authority
EP
European Patent Office
Prior art keywords
enrollment
matrices
frames
steps
signal
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
EP98957754A
Other languages
German (de)
English (en)
Inventor
Sherrie Ann Adcock
Gerald S. Schwartz
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Voice Security Systems Inc
Original Assignee
Individual
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Individual filed Critical Individual
Publication of EP1129447A1 publication Critical patent/EP1129447A1/fr
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G10L17/06—Decision making techniques; Pattern matching strategies
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/08—Speech classification or search
    • G10L15/10—Speech classification or search using distance or distortion measures between unknown speech and reference templates

Definitions

  • TITLE METHOD OF COMPARING UTTERANCES FOR SECURITY CONTROL
  • This invention relates generally to electronic security methods which provide for modeling or otherwise comparing human features such as fingerprints, voice patterns, and retina patterns, in order to distinguish between individuals, and, more particularly, to a security method and protocol for modeling and comparing voice utterances to control the operation of a security device.
  • Parra, U.S., 5,313,556 describes the identity of an individual (known or unknown) that is determined by a sonic profile of sounds issued through his oral-nasal passages. The sounds are converted to digital electrical signals and produce a three domain format of frequency, amplitude and time samples to produce an array of peaks and valleys constituting the sonic profile of an individual.
  • a source or library of sonic profiles in the same format of a known individual have a interrelationship including relative positions of said peaks and valleys of said sonic profile of the known individual with that of said unknown individual compared and a utilization signal is provided upon detecting or non- detecting a correlation between said sonic profiles.
  • Hair et al., U.S., 3,673,331 describes voice verification that is accomplished at a plurality of spaced apart facilities each having a plurality of terminals.
  • Multiplexing structure interconnects the terminals through a communications link to a central processing station.
  • Analog reproductions of voices transmitted from the terminals are converted into digital signals.
  • the digital signals are transformed into the frequency domain at the central processing station.
  • Predetermined features of the transformed signals are compared with stored predetermined features of each voice to be verified.
  • N verify or non-verify signal is then transmitted to the particular terminal in response to the comparison of the predetermined features.
  • a security card (which may be a credit card) according to the invention has recorded on it data identifying a personal and non- counterfeitable attribute, such as the voice characteristics, of the authorized holder of the card.
  • N card utilization system provides means for comparing the attribute as recorded by these data with the corresponding attribute of the person wishing to use the card, thereby substantially eliminating the possibility of unauthorized card utilization.
  • Muroi et al., U.S., 4,833,713 describes a voice or sound recognition system including a microphone for converting a voice into an electrical voice signal, a frequency analyzer for generating a voice pattern in the form of a time-frequency distribution, and a matching unit for matching the voice pattern with registered voice patterns.
  • Feix et al. U.S., 4,449,189 describes a method and an apparatus for identifying an individual through a combination of both speech and face recognition.
  • the voice signature of an interrogated person uttering a key word into a microphone is compared in a pattern matcher with the previously stored voice signature of a known person uttering the same key word to obtain a first similarity score.
  • a key event in the utterance of the key word by the interrogated person occurs, a momentary image of that person's mouth region onto which a grid pattern has been projected is optically recorded and compared with the previously stored corresponding momentary image of the same known person to obtain a second similarity score.
  • the prior art teaches the comparing of voice signatures in time as well as frequency domain. However, the prior art does not teach a means for filtering such voice profiles by difference techniques.
  • the present invention method fulfills these needs and provides further related advantages as described in the following summary.
  • the present invention teaches certain benefits in methods which give rise to the objectives described below.
  • the present invention is a security method which compares a present verbal utterance with a previously recorded verbal utterance by comparing frequency domain representations of the present utterance, with previously recorded multiply repeated utterances of the same material, forming a basis for comparison.
  • the present method approaches the comparison by establishing energy content in a variety of cells in the frequency domain, and instead of focusing on the ability of an individual to repeat an utterance from one trial to the next, sometimes separated by days, weeks or even longer, the present approach focuses on the variability of the difference between multiple utterances of the same words or phrases.
  • the method attempts to determine if two sounds were produced by the same human voice in an attempt to discriminate between allowed and non-allowed personnel seeking to operate a secure device. Further, the method may be used to determine what command is being given by the individual, culling out the selected command from a library of such commands all uttered by the same individual.
  • the present method invention has as an important aspect, the discrimination between, and, or matching of a presently uttered verbal word or phrase with the same utterance stored in a library of such utterancEs.
  • Another aspect of the present method is the achievement of high accuracy and fast results in discrimination and, or matching of verbal utterances by using a difference method for comparison.
  • the present method is a non-obvious and highly effective procedure for extremely high speed comparison of large data sets against a challenge so as to provide the convenience, for instance, of verbal only challenges at a secure door used by a large number of individuals with separate verbal access codes, wherein the delay time for approval has been shown to be in the range of a few seconds.
  • the method also results in an extraordinary level of discrimination between individuals while providing a high level of "forgiveness" for the normal changes in tone, timber and volume of the human voice from moment to moment and day to day.
  • the discrimination capability of the present method is strong enough for use in military as well as industrial applications, and is inexpensive and simple to use so as to find application in residential use or commercial.
  • a further aspect of the present method is the use of testing for minimum energy levels in a set of frequency ranges in the frequency domain, as well as testing for corresponding energy levels that surpass a selected energy level criterion.
  • the present invention provides a step-by-step method for comparing a verbal utterance of a speaker in the present (challenge utterance), with a previously recorded verbal utterance (enrollment utterance), to determine the validity of the speaker, i.e., if the challenge utterance is being produced by the same person as was the enrollment utterance.
  • an action is authorized, such as opening a lock, dialing a secret phone number granting access to data or services, etc.
  • the method comprises certain steps which are performed in sequence.
  • the steps are defined by, first, preparing an enrollment data from an utterance from one or more persons, next, challenging the enrollment data with a present utterance from one of those persons, and finally enabling the security action if the challenge utterance is accepted as being close enough to one of the enrollment data.
  • Preparing the enrollment data comprises the steps of, converting a verbal utterance, which we shall refer to as an enrollment utterance, into a first electrical signal as by a microphone or other transducer.
  • the electrical signal is tranformed into a digital format.
  • a fast Fourier transformation of this electrical signal is conducted to produce a frequency domain representation of the enrollment utterance.
  • the Frequency domain representqation is then divided into frames of time, e.g. 10ms. Frames which show no energy content are deleted.
  • a number of samples, represented by "M" are taken for each of N frequency channels to form an M by N sample enrollment matrix Ei.
  • M and N are selected as integers of a magnitude necessary for the level of security desired with larger numbers providing greater security and vice-versa.
  • This matrix provides cell samples Mi-Ni, where i represents an integer, which are characterized by a total energy content within each cell, i.e., a number.
  • the method provides for determining if at least X, an arbitrary number, of the M samples have a selected minimum energy content in at least Y, another arbitrary number, of the N frequency channels. If not, the enrollment utterance is repeated until the criteria X and Y are satisfied, or if not, after several tries, the process of enrollment is aborted. This usually would only happen if the speaker is not able to provide enough volume in his or her speech to enable production of the minimum energy criterion or if the input is not a human voice so that a necessary spectral content is lacking.
  • the method requires forming the difference between each pair of the enrollment matrices Ei, as (E1-E2), (E1-E3), (E2-E3), ... .
  • Algorithms are applied such that each individual frame is compared with each other frame and is allowed to slip any number of frames, forward or backward in order to obtain a best match.
  • the second part of this step is also critical to the viability of the present method in that ES assures us that a recording is not being used as an impostor, i.e., we can be sure that the differences in the matrices are at least as great as would be normally expected from a human voice.
  • ES assures us that a recording is not being used as an impostor, i.e., we can be sure that the differences in the matrices are at least as great as would be normally expected from a human voice.
  • a recording is used for each of the enrollments, we find that variability between them is less than is produced by the human voice.
  • Challenging the enrollment data comprising the steps of, first, converting a challenge verbal utterance into a second electrical signal, as by a microphone or other transducer.
  • This electrical signal is converted into a digital format.
  • the digital signal is ued to perform a second fast Fourier transformation of the represented electrical signal to produce a frequency domain representation of the challenge utterance.
  • the Frequency spectrum is then divided into frames of time, e.g. 10 ms. Frames which show no energy content are deleted. Taking M samples of the second signal for each of N frequency channels to form an M by N sample challenge matrix C and determining if at least X of the M samples have a selected minimum energy content in at least Y of the N frequency channels.
  • S [(C-Dl) + (C-D2) + (C-D3)+ ...] is formed.
  • S' is accepted as a valid challenge if S' ⁇ VB, VB being a selected first verification criterion and also if each said difference between each of the pairs C and Di is greater than VS, VS being a selected second verification criterion.
  • Both VB and VS play corresponding roles to EB and ES in the enrollment procedure and are used for the same reason. When these criteria are met the challenge verbal utterance is accepted and the requisite security step is enabled.
  • such a system may use the method for a plurality of users at the same time when an enrollment is completed for each user.
  • a challenge is made by any one of the users and the challenge method is then carried out for each of the enrollments until either a successful challenge is made or all of the enrollments have been tested without success.
  • each of the data sets may include a definition of a specific security action so that when a specific match is made between a challenge and an enrollment, the specific action may be carried out in deference to other actions corresponding to the other enrollments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Game Theory and Decision Science (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Collating Specific Patterns (AREA)

Abstract

L'invention concerne un procédé de sécurité comparant une émission verbale actuelle avec une émission verbale enregistrée par la comparaison des représentations du domaine de fréquences des émissions vocales à multiples émissions vocales répétées qui forment une base permettant de déterminer une variation dans des opérations répétées par un individu avec des différences semblables entre les émissions d'enrôlement et d'intervention. Selon une variante, un ensemble de données d'enrôlement sont recherchées par chaque intervention jusqu'à ce qu'il y ait une correspondance qui indique une action, dépendant éventuellement de la correspondance spécifique. Il se peut qu'aucune correspondance ne soit détectée ce qui indique un échec.
EP98957754A 1998-11-10 1998-11-10 Procede de comparaison d'emission de paroles en vue d'un controle de securite Pending EP1129447A1 (fr)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/US1998/023928 WO2000028524A1 (fr) 1998-11-10 1998-11-10 Procede de comparaison d'emission de paroles en vue d'un controle de securite

Publications (1)

Publication Number Publication Date
EP1129447A1 true EP1129447A1 (fr) 2001-09-05

Family

ID=22268265

Family Applications (1)

Application Number Title Priority Date Filing Date
EP98957754A Pending EP1129447A1 (fr) 1998-11-10 1998-11-10 Procede de comparaison d'emission de paroles en vue d'un controle de securite

Country Status (4)

Country Link
EP (1) EP1129447A1 (fr)
JP (1) JP2002529799A (fr)
AU (1) AU1393899A (fr)
WO (1) WO2000028524A1 (fr)

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3673331A (en) * 1970-01-19 1972-06-27 Texas Instruments Inc Identity verification by voice signals in the frequency domain
BE787377A (fr) * 1971-08-09 1973-02-09 Waterbury Nelson J Cartes de securite et systeme d'utilisation de telles cartes
US4833713A (en) * 1985-09-06 1989-05-23 Ricoh Company, Ltd. Voice recognition system
US5216720A (en) * 1989-05-09 1993-06-01 Texas Instruments Incorporated Voice verification circuit for validating the identity of telephone calling card customers
JP2002517175A (ja) * 1991-02-22 2002-06-11 シーウェイ テクノロジーズ インコーポレイテッド 人間の音源を識別するための手段および装置
US5293452A (en) * 1991-07-01 1994-03-08 Texas Instruments Incorporated Voice log-in using spoken name input
US5339385A (en) * 1992-07-22 1994-08-16 Itt Corporation Speaker verifier using nearest-neighbor distance measure
US5608784A (en) * 1994-01-24 1997-03-04 Miller; Joel F. Method of personnel verification using voice recognition
US5835894A (en) * 1995-01-19 1998-11-10 Ann Adcock Corporation Speaker and command verification method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See references of WO0028524A1 *

Also Published As

Publication number Publication date
JP2002529799A (ja) 2002-09-10
AU1393899A (en) 2000-05-29
WO2000028524A1 (fr) 2000-05-18

Similar Documents

Publication Publication Date Title
US5835894A (en) Speaker and command verification method
US6519565B1 (en) Method of comparing utterances for security control
US6480825B1 (en) System and method for detecting a recorded voice
Naik Speaker verification: A tutorial
EP0647344B1 (fr) Procede de reconnaissance de suites alphanumeriques enoncees sur un reseau telephonique
EP0953972B1 (fr) Reconnaissance de la parole indépendante du locuteur simultanée avec une vérification du locuteur sur un réseau téléphonique
KR0139949B1 (ko) 미지인 사람의 신원을 확인하기 위한 음성 검증 회로
US5127043A (en) Simultaneous speaker-independent voice recognition and verification over a telephone network
EP2120232A1 (fr) Système de certification de cryptage aléatoire d'empreinte vocale, verrou de cryptage aléatoire d'empreinte vocale, et procédé de création associé à ceux-ci
US20070294083A1 (en) Fast, language-independent method for user authentication by voice
CA2172406A1 (fr) Acces a un compte commande par la voix via un reseau telephonique
US6161094A (en) Method of comparing utterances for security control
Orken et al. Development of security systems using DNN and i & x-vector classifiers
Campbell Jr Speaker recognition
US10957318B2 (en) Dynamic voice authentication
EP1129447A1 (fr) Procede de comparaison d'emission de paroles en vue d'un controle de securite
Corsi Speaker recognition: A survey
Marinov Text dependent and text independent speaker verification system: Technology and application
Campbell et al. Audio sensors and low-complexity recognition technologies for soldier systems
Chadha et al. Text-independent speaker recognition for low SNR environments with encryption
Inthavisas et al. Attacks on speech biometric authentication
JPS63157199A (ja) 話者照合装置
Chenafa et al. Speaker recognition using decision fusion
JPS63269283A (ja) 個人認証装置
Kounoudes et al. Intelligent Speaker Verification based Biometric System for Electronic Commerce Applications

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20010611

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AT BE CH DE DK ES FI FR GB GR IT LI NL PT SE

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: VOICE SECURITY SYSTEMS, INC.

18W Application withdrawn

Effective date: 20050412

D18W Application withdrawn (deleted)
STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN

18D Application deemed to be withdrawn

Effective date: 20050531