`a2) Patent Application Publication 0) Pub. No.: US 2003/0046083 Al
`(43) Pub. Date: Mar.6, 2003
`
`DEVINNEY, JR.et al.
`
`US 20030046083A1
`
`(54) USER VALIDATION FOR INFORMATION
`SYSTEM ACCESS AND TRANSACTION
`PROCESSING
`
`(76)
`
`Inventors; EDWARD J. DEVINNEYJR.,
`DELANCO, NJ (US); MANISH
`SHARMA, SOMERSET,NJ (US);
`CHRIS KEYSER, ROEBLING, NJ
`(US); RAINER ROTHACKER,
`CLIFFWOOD BEACH,NJ(US);
`RICHARD J. MAMMONE,
`BRIDGEWATER,NJ (US)
`
`Correspondence Address:
`Thomas H. Young, Esq.
`Merchant & Gould, P.C.
`3200 IDS Center
`80 South Eighth Street
`Minneapolis, MN 55402-2215 (US)
`
`(*) Notice:
`
`This is a publication of a continued pros-
`ecution application (CPA) filed under 37
`CER 1.53(d).
`
`(21) Appl. No.:
`
`08/976,279
`
`(22)
`
`Filed:
`
`Nov.21, 1997
`
`Related U.S. Application Data
`
`(60) Provisional application No. 60/031,638, filed on Nov.
`22, 1996.
`
`Publication Classification
`
`(SL) Ute C07 casccccssssssssstenssnssssssttntsnsstn G10L 21/00
`(52) US. Ch.
`cecesesssssssssstsesesnstnstvasnesnstnstvesse 704/273
`
`(57)
`
`ABSTRACT
`
`The present invention applies speech recognition technology
`to remote access, verification, and identification applica-
`tions. Speech recognition is used to raise the security level
`of many types of transaction systems which previously had
`serious safety drawbacks, including: point of sale systems,
`homeauthorization systems, systems for establishing a call
`to a called party (including prison telephone systems),
`internet access systems, website access systems, systems for
`obtaining access to protected computer networks, systems
`for accessing a restricted hyperlink, desktop computer secu-
`rity systems, and systems for gaining access to a networked
`server. A general speech recognition system using commu-
`nication is also presented. Further, different types of speech
`recognition methodologies are useful with the present inven-
`tion, such as “simple” security methods and systems, multi-
`tiered security methods and systems, conditional multi-
`tiered security methods
`and systems,
`and randomly
`prompted voice token methods and systems.
`
`201
`
`aw
`
`SPEECHRECOGNITION UNIT
`
`Preprocessor Unit
`LeeLee __ 212 1
`J
`
`216
`DECISION/
`GONFIDENCE
`
`TEST SPEECH
`2038
`extraction Unit |
`~ sa Feature
`PROMPT
`“
`6 Extraction Unit
`1 se Rea
`TTT IUut Ne —-~--—3)
`CTT TTT
`EX
`m 206
`
`Comparison Unit/
`Processing Unit
`
`| 210
`
`
`
`Page 1 of 30
`
`PETITIONERS' EXHIBIT 1024
`
`Page 1 of 30
`
`PETITIONERS' EXHIBIT 1024
`
`
`
`TEST SPEECH
`203
`J
`,
`PROMPT
`|
`. “4
`Feature Extraction Unit
`Lae ee
`,e ee eae _ 214,
`a|
`|
`INDEX
`Comparison Unit/
`Processing Unit
`!
`206
`
`.
`
`L
`
`—-—--—
`
`212 |
`-——_- =
`
`216
`DECISION/
`GONFIDENCE
`
`Patent Application Publication Mar. 6,2003 Sheet 1 of 18
`
`US 2003/0046083 Al
`
`FIG. 1
`
`201
`
`Mw
`
`SPEECH RECOGNITION UNIT
`
`Preprocessor Unit
`
`208 spech Data
`
`210
`
`Database
`Of
`Stored
`
`Page 2 of 30
`
`Page 2 of 30
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 2 of 18
`
`US 2003/0046083 Al
`
`FIG. 2
`Sau 7
`|
`INPUT
`UNIT
`299 1”
`
`|
`
`_—
`
`204
`
`! I | J u
`
`->
`
`UNIT
`
`218
`
`CONFIDENCE
`
`USER
`2
`INDEX
`t-~-——
`INPUT
`|
`UNIT 218 |
`
`|
`|
`|
`
`Page 3 of 30
`
`Page 3 of 30
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 3 of 18
`
`US 2003/0046083 Al
`
`221
`
`FIG. 3
`
`ZS
`-p--~ 4
`
` Obtain
`|
`Obtain
`|
`
`Spoken
`I
`Index
`|
`Password
`
`224
`
`Process In Speech
`Recognition Unit
`
`
`
`
`Authorize/
`Identify
`
`—_ o_o
`
`— am eee oe
`
`(Confidence)
`(Decision)
`—__ ee
`
`Page 4 of 30
`
`Page 4 of 30
`
`
`
`Patent Application Publication Mar. 6, 2003 Sheet 4 of 18
`
`US 2003/0046083 Al
`
`FIG. 4A
`
`242
`
`SL 248
`246
`| | a
`-—74
`
`Obtain 2nd
`
`Obtain
`Obtain 1st
`Speech
`Speech
`index
`
`
`Password
`Password —--r- |
`
`
`
`
`|
`
`Random Prompt
`For Second Password
`
`
`
`
`
` Recognize
`1st Password
`?
`
`Recognize
`
`
`
`(Confidence)
`(Decision)
`——~__
`
`Page 5 of 30
`
`Page 5 of 30
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 5 of 18
`
`US 2003/0046083 Al
`
`261
`
`- f- ~ 4y
`;
`Obtain
`|
`index
`\
`|
`--T7--
`
`FIG. 4B
`
`For Second Password
`
`a
`
`oa
`271
`262
`;
`
`
`
`Obtain 2nd
`
`Obtain ist
`
`
`
`Speech
`Speech
`p
`
`Password
`Password
`
`
`
`
`
`
` Random Prompt
`
`
`
`
`
`
`
`
`Recognize
`
`1st Password
`
`
`?
`
`~ TI
`
`ol
`
`
`
`| 1 | I |
`
`|
`
`|\
`
`|
`
`
`Recognize
`
`
`2nd Password
`
`?
`
`
`No Authorize/
`
`No Identity
`
`Authorize/
`Identify
`
`v
`(Confidence)
`(Decision)
`—_
`216
`
`Page 6 of 30
`
`Page 6 of 30
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 6 of 18
`
`US 2003/0046083 Al
`
`FIG. 4C
`
`284
`
`281
`
`L
`
`286
`
`
` Obtain
`
`Speech
`Password
`
`
`identify
`Characteristics
`
`288
`
`Recognize
`Characteristics
`2?
`
`
`
`
`
`
`
`
`282
`
`Select
`Model
`
` 292
`
`283
`
`No Authorize/
`No Identity
`
`Authorize/
`Identify
`
`294
`
`
`
`
`(Decision)
`(Confidence)
`
`{v
`
`216
`
`Page 7 of 30
`
`Page 7 of 30
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 7 of 18
`
`US 2003/0046083 Al
`
`52
`
`VOICE RECOGNITION
`SYSTEM/SERVICE
`
`CLIENT
`TERMINAL
`
`FIG. 5A
`
`Page 8 of 30
`
`Page 8 of 30
`
`
`
`Patent Application Publication Mar. 6, 2003 Sheet 8 of 18
`
`US 2003/0046083 Al
`
`
` 402
`CLIENT TERMINAL
`
`
`
`Feature Extraction
`r
`Kz _ 412
`-~-T7T77
`
`a
`~N
`-—~x¥—-*¥——
`7 TT TY
`rT ae rT
` Enerypti
`|
`<--
`Tyption/
`DIA
`Decryption_ _|
`
`
`rT TTT
`Data Output
`|
`41 User am
`
`424
`
`
`
`
`
`RECOGNITION
`SYSTEM/
`SERVICE
`
`
`428 ~Eheryption/ 1
`NY Decryption
`
`
`Comparison/
`
`Processing Unit
`
`|
`
`57
`
`434
`
`POT TTT
`|
`Comparison/
`1
`
`Processing Unit
`\
`
`
`DATABASE
`
`
`FIG. 5B
`
`Page 9 of 30
`
`Page 9 of 30
`
`
`
`Patent Application Publication Mar. 6, 2003 Sheet 9 of 18
`
`US 2003/0046083 Al
`
`
`
`VALIDATION
`SERVICE
`
`14
`
`462
`
`‘460
`
`
`
`
`VOICE
`
`IDENTIFICATION
`DATABASE
`
`16
`
`
`454
`
`456
`
`POINT
`OF
`
`TERMINAL
`
`~
`
`
`
`
`
`
`
`
`
`
`
`452
`
`
`450
`12
`
`FIG. 6
`
`Page 10 of 30
`
`Page 10 of 30
`
`
`
`Patent Application Publication
`
`Mar. 6, 2003 Sheet 10 of 18
`
`US 2003/0046083 Al
`
`es
`
`wy
`
`~~
`
`9b
`
`ay
`
`AINAI19
`
`WNIWHSL
`
`JOIOA anragualshs
`
`NOLLINDOOSY
`
`AOIAWASWALSAS
`
`Page 11 of 30
`
`Page 11 of 30
`
`
`
`
`
`Patent Application Publication
`
`Mar. 6, 2003 Sheet 11 of 18
`
`US 2003/0046083 Al
`
`
`
`NOLLINDO93YSOIOA
`
`JOINMSSINSISAS
`
`vehce
`
`987
`
`8Old
`
`3H99=
`
` 2°3uYMOLNHnoveatv)30VINI
`
`mANOHATTBL/S19
`
`aAlby
`
`08?
`
`09
`
`Page 12 of 30
`
`Page 12 of 30
`
`
`
`
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 12 of 18
`
`US 2003/0046083 Al
`
`600
`
`608
`
`
`VOICE
`
`
`RECOGNITION
`
`
`
`
`SERVICE CALL CENTER
`
`20
`
`FIG. 9
`
`Page 13 of 30
`
`Page 13 of 30
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 13 of 18
`
`US 2003/0046083 Al
`
`VIDB
`
`304
`304
`
`303
`
`&02
`
`PG
`
` RECOGNITICN
`
`SERVER
`
`ww
`
`WEB SERVER
`
`302
`
`303
`
`303
`
`PC
`
`02
`
`PC
`
`602
`
`FIG. IOA
`
`Page 14 of 30
`
`Page 14 of 30
`
`
`
`FIG. 1OB
`
`
`52
`
`Client
`Terminal
`
`
`
`4
`11
`
`16
`
`
`
`
`Recognition
`Server
`
`
`
`Voice information
`Database (VIDB)
`
`So 628
`
`
`
`
`
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 14 of 18
`
`US 2003/0046083 Al
`
`620
`
`ssL
`Connection
`
`Firewall
`
`Protected
`Network
`
`624
`
`622
`
`Page 15 of 30
`
`Page 15 of 30
`
`
`
`Patent Application Publication Mar. 6, 2003 Sheet 15 of 18
`
`US 2003/0046083 Al
`
`FIG. 10C
`
`
`Recognition
`Management
`
`Server
`
`| {
`
`52
`
`Encrypted
`
`Client
`
`Program
`
`?
`
`1
`
`Page 16 of 30
`
`
`
`
`Recognition
`Server
`
`
`cemecten|
`VOICEINFORMATION
`Terminal
`|
`DATABASE(VIDB)
`
`.
`Authentication
`
`640
`
`\
`
`|
`
`Encrypted
`Connection
`on
`
`
`
`638
`
`Restricted Hyperlink 636
`
`Page 16 of 30
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 16 of 18
`
`US 2003/0046083 Al
`
`650
`
`FIG. 11 x
`
`File Encryption
`
`DESKTOP
`STATION
`
`Voice Secured System Log On
`
`VoiceSecuredScreenSaver
`
`Administrative Application
`
`|
`\
`
`Page 17 of 30
`
`Page 17 of 30
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 17 of 18
`
`US 2003/0046083 Al
`
`FIG. 12A
`
`660
`
`S
`
`664
`
`Authentication
`Server
`
`52
`
`Client
`Terminal
`
`
`
`
`
`Voice information
`- Networked
`Server
`Database (VIDB)
`
`
`
`
`
`© Voice Secured System Log On
`
`
`<> Administrative Application
`
`662
`
`11
`
`Page 18 of 30
`
`Page 18 of 30
`
`
`
`Patent Application Publication Mar. 6,2003 Sheet 18 of 18
`
`US 2003/0046083 Al
`
`FIG. 12B
`
`ACCESS ATTEMPT 670
`
`oA°
`
`&ve
`
`— 690
`
`First level authentication enabled?
`
`674 a
`
`
`
`
`ee
`yes
`
`Does server only allow access to
`voice pass enabled clients?
`
`Is user {D located in
`database?
`
`DENY
`
`no
`
`DENY &78
`
`676
`
`yes
`
`680
`
`Is user authorized for
`Inoen-voice authentication?
`ho
`
`yes
`
`:
`
`DENY
`699
`
`yes
`
`696
`
`ACCEPT 698
`
` DENY 694
`
`
`684
`
`ACCEPT 682
`
`DENY
`
`\
`
`Page 19 of 30
`
`Page 19 of 30
`
`
`
`US 2003/0046083 Al
`
`Mar. 6, 2003
`
`USER VALIDATION FOR INFORMATION SYSTEM
`ACCESS AND TRANSACTION PROCESSING
`
`CROSS REFERENCE TO RELATED
`APPLICATIONS
`
`[0001] This application claims priority from U.S. Provi-
`sional Application Ser. No. 60/031,638, Filed Nov. 22, 1996,
`entitled “User Validation For Information System Access
`And Transaction Processing.”
`
`BACKGROUND OF THE INVENTION
`
`[0002] The inventionis a verification system for ensuring
`that transactions are completed securely. The invention uses
`the principle of speaker recognition to allow a user to
`complete a transaction.
`
`[0003]
`
`1. Field of the Invention
`
`[0004] The invention relates to the fields of signal pro-
`cessing, communications, speaker recognition and security,
`and secure transactions.
`
`[0005]
`
`2. Description of Related Art
`
`[0006] With the increased use of credit card and computer
`related transactions security of the transactions is a reoccur-
`ring
`problem of
`increasing
`concern. Conventional
`approachesfor credit card validation have included reading
`a magnetic strip of the credit card at a point of sale.
`Information stored on the credit card, such as account
`information, is forwarded over a telephone connection to a
`credit verification service at the credit card company. For
`example, an X.25 connection to the credit verification sys-
`tem has been used. A response from the credit verification
`service indicates to the salesperson whether the customer’s
`credit card is valid and whether the customer has sufficient
`
`credit. An example of the above-described system is manu-
`factured by VeriFone® of Redwood City, Calif., U.S.A..
`Theseprior art systems, however, have the disadvantagethat
`the credit card may be verified as valid and as having
`sufficient credit even if it is used by someone whois not
`authorized to use the credit card.
`
`[0007] The identity of the consumer who presents a credit
`card is manually verified by a merchant. The back of the
`credit card contains a signature strip, which the consumer
`signs upon credit card issuance. The actual signature of the
`consumerat the time of sale is compared to the signature on
`the back of the credit card by the merchant. If in the
`merchant’s judgement, the signatures match,the transaction
`is allowed to proceed.
`
`include placing
`[0008] Other systems of the prior art
`photographs of authorized users on the credit card. At the
`time of the transaction, the merchant compares the photo-
`graph on the card with the face of the person presenting the
`card. If there appears to be a match,
`the transaction is
`allowed to proceed.
`
`[0009] While signatures and photographs are personal
`characteristics of the user, they have not been veryeffective.
`Signatures are relatively easy to forge and differences
`between signatures and photographs may go unnoticed by
`inattentive merchants. These systems are manual and con-
`sequently prone to human error. Further,
`these systems
`cannot be used with credit card transactions which do not
`
`occur in person, i.e., which occur via telephone.
`
`[0010] Computer related applications, such as accessing
`systems, local area networks, databases and computer net-
`work(such as “Internet”) systems, have conventionally used
`passwords (known as personal
`identification numbers—
`“PINs”) entered from a keyboard as a security method for
`accessing information. Computer passwords have the short-
`coming of being capable of being stolen, intercepted or
`re-created by third parties. Computer programs exist for
`guessing (“hacking”) passwords. Additionally, computer
`passwords/PINs are not personal characteristics, which
`means that they are less complex and easier to generate by
`a third party with no knowledge of the authorized individu-
`al’s personal characteristics.
`
`[0011] With the advent of electronic commerce on the
`internet, goods and services are increasingly being pur-
`chased by consumers, who submit credit card or other
`“secure” information to merchants over the internet. Trans-
`actions initiated from users connected to the internet cur-
`
`rently have limited security provisions. For example, a retail
`provider receiving a user’s credit card number from the
`internet has no idea whether the person providing the
`numberis authorized to use the credit card, or has obtained
`a credit card number from an illegal source.
`
`[0012] As computers play a greater and morecritical role
`in everyday life, security has emerged as a significant
`concern. Whetherit’s restricting children from playing with
`their parent’s tax return (local access), protecting against an
`employeestealing trade secrets (network access), or limiting
`access to a value added WEBsite (remote network access),
`the ability to determine that the claimed useris the real user
`is absolutely necessary.
`
`[0013] Additional areas in which a need for heightened
`security exists are cellular telephone systems and prison
`telephone systems. In cellular systems, fraud from unautho-
`rized calling is a recurring problem. In prison systems, the
`identity of inmates must be closely monitored, for purpose
`of authorizing certain transactions, such as telephonecalls.
`
`[0014] What is needed are local and remote secure access
`systems and methods using personal characteristics of users
`for identifying and/or verifying the users.
`
`SUMMARYOF THE INVENTION
`
`[0015] The present invention is an improved method and
`system for increasing the security of credit card transactions,
`prison inmate transactions, database access requests, inter-
`net transactions, and other transaction processing applica-
`tions in which high security is necessary. According to the
`present invention, voice print and speaker recognition tech-
`nology are used to validate a transaction or identify a user.
`
`[0016] Within speaker recognition (also referred to as
`voice recognition herein),
`there exists two main areas:
`speaker identification and speaker verification. A speaker
`identification system attempts to determine the identity of a
`person within a knowngroup of people using a sample of his
`or her voice. Speaker identification can be accomplished by
`comparing a voice sample of the user in question to a
`database of voice data, and selecting the closest match in the
`database. In contrast, a speaker verification system attempts
`to determineif a person’s claimed identity (whom the person
`claims to be) is valid using a sample of his or her voice.
`Speaker verification systems are informed of the person’s
`
`Page 20 of 30
`
`Page 20 of 30
`
`
`
`US 2003/0046083 Al
`
`Mar. 6, 2003
`
`claimed identity by index information, such as the person’s
`claimed name, credit card number, or social security num-
`ber. Therefore, speaker verification systems typically com-
`pare the voice of the user in question to one set of voice data
`stored in a database, the set of voice data identified by the
`index information.
`
`[0017] Speaker recognition provides an advantage over
`other security measures such as passwords (including per-
`sonal
`identification numbers) and personal
`information,
`because a person’s voice is
`a personal characteristic
`uniquely tied to his or her identity. Speaker verification
`therefore provides a robust method for security enhance-
`ment.
`
`consists of determining
`[0018] Speaker verification
`whether or not a speech sample provides a sufficient match
`to a claimed identity. The speech sample can be text depen-
`dent or text independent. Text dependent speaker verifica-
`tion systems identify the speaker after the utterance of a
`password phrase. The password phrase is chosen during
`enrollment and the same password is used in subsequent
`verification. Typically, the password phrase is constrained
`within a specific vocabulary (i.e. numberof digits). A text
`independent speaker verification system does not use any
`pre-defined password phrases. However, the computational
`complexity of text-independent speakerverification is much
`higher than that of text dependent speaker verification
`systems, because of the unlimited vocabulary.
`
`[0019] The present invention uses speech biometrics as a
`natural interface to authenticate users in today’s multi-media
`networked environment, rather than a password that can be
`easily compromised.
`
`In accordance with the present invention, security
`[0020]
`can be incorporated in at least three access levels: at the
`desktop, on corporate network servers (NT, NOVELL,or
`UNIX), and at a WEBserver (internets/intranets/extranet).
`The security mechanisms may control access to a work
`station, to networkfile servers, to a web site, or may secure
`a specific transaction. Nesting of these security levels can
`provide additional security; for instance, a company could
`choose to have it’s work stations secured locally by a
`desktop security mechanism, as well as protect corporate
`data on a file server with a NT, NOVELL or FTP server
`security mechanism.
`
`[0021] Use of speaker recognition, and therefore voice
`biometric data, is able to provide varying levels of security
`based upon customer requirements. A biometric confirmsthe
`actual identity of the user; other prevalent high security
`methods, such as token cards, can still be compromised if the
`token card is stolen from the owner. A system can employ
`any of these methods at any access level. In all cases of the
`inventive methods described herein, the user must know an
`additional identifying piece of information. The security
`system is not compromised whether this information is
`publicly obtainable information, such as their name, or a
`private piece of information, such as a PIN,a social security
`number, or an account number.
`
`In accordance with the present invention, “simple”
`[0022]
`security systems and methods (single spoken password),
`multi-tiered security systems (multiple tiers of spoken pass-
`words) and randomly prompted voice tokens (prompting of
`words obtained through a random look-up) are provided for
`
`improved security. These security systems and methods may
`be used to increase the security of point of sale systems,
`homeauthorization systems, systems for establishing a call
`to a called party (including prison telephone systems),
`internet access systems, website access systems, systems for
`obtaining access to protected computer networks, systems
`for accessing a restricted hyperlink, desktop computer secu-
`rity systems, and systems for gaining access to a networked
`server.
`
`BRIEF DESCRIPTION OF THE DRAWINGS
`
`[0023]
`
`FIG.1 is a diagram of a speech recognition unit.
`
`[0024] FIG. 2 is a high level representation of the unit
`shown in FIG.1.
`
`[0025] FIG. 3 shows a “simple” security method and
`system.
`
`[0026] FIG. 4A shows a diagram of a multi-tiered security
`method and system.
`
`[0027] FIG. 4B showsa diagram of a multi-tiered security
`method and system with conditional tiers.
`
`[0028] FIG. 4C showsa diagram of a randomly prompted
`voice token method and system.
`
`[0029] FIG. 5A showsa schematic diagram of the general
`configuration of a speaker verification method and system.
`
`[0030] FIG. 5B shows a more specific schematic of the
`FIG. 5A method and system.
`
`[0031] FIG. 6 is a schematic diagram of a speaker recog-
`nition method and system for a point of sale system.
`
`[0032] FIG. 7 is a schematic diagram of an embodiment
`where home authorization is obtained through a call center.
`
`[0033] FIG. 8 is a schematic diagram of an embodiment
`for establishing a call
`to a called party using speaker
`recognition.
`
`[0034] FIG. 9 is a schematic diagram of an embodiment
`for use in establishing an internet connection using speaker
`recognition.
`
`[0035] FIG. 10A is a schematic diagram of an embodi-
`mentfor use in establishing a connection to a web site using
`speaker recognition.
`
`[0036] FIG. 10B is a schematic diagram of an embodi-
`ment for use in establishing a connection to a protected
`network using speaker recognition.
`
`[0037] FIG. 10C is a schematic diagram of an embodi-
`ment for use in establishing a connection to a restricted
`hyperlink on a web server using speaker recognition.
`
`[0038] FIG. 11 shows an embodimentfor use in securing
`a desktop computer using speaker recognition.
`
`[0039] FIG.12A shows a system for use in gaining access
`to a networked server using speaker recognition.
`
`[0040] FIG. 12B shows a methodfor use in gaining access
`to a networked server using speaker recognition.
`DESCRIPTION OF THE PREFERRED
`EMBODIMENT(S)
`[0041] The present invention uses speech recognition in
`combination with various security and communications sys-
`
`Page 21 of 30
`
`Page 21 of 30
`
`
`
`US 2003/0046083 Al
`
`Mar. 6, 2003
`
`tems and methods. As a result, an inventive, remotely
`accessible and fully automatic speech verification and/or
`identification system results.
`[0042]
`1. Speech Recognition Unit.
`[0043]
`FIG.1 illustrates a speech recognition system 201.
`Test speech 202 from a user is input into a speech recogni-
`tion unit 204, which contains a database of stored speech
`data. A prompt 203 may be presented to the user to inform
`the user to speak a password or enter index information. In
`a speaker verification system, an index 206 is normally
`supplied, which informs the speech recognition unit 204 as
`to which data in the database 208 is to be matched up with
`the user. In a speaker identification system, an index 206 is
`normally not input, and the speech recognition unit 204
`cycles through all of the stored speech data in the database
`to find the best match, and identifies the user as the person
`corresponding to the match. Alternatively,if a certain thresh-
`old is not met, the speech identification system 204 may
`decide that no match exists.
`
`the speech recognition unit 204
`In either case,
`[0044]
`utilizes a comparison processing unit 210 to comparethe test
`speech 202 with stored speech data in a database 208. The
`stored speech data may be extracted features of the speech,
`a model, a recording, speech characteristics, analog or
`digital speech samples, or any information concerning
`speech or derived from speech. The speech recognition unit
`204 then outputs a decision 216, either verifying (or not) the
`user, or identifying (or not) the user. Alternatively,
`the
`“decision”216 from the speech recognition umit includes a
`confidence level, with or without the verification/identifica-
`tion decision. The confidence level may be data indicating
`how close the speech recognition match is, or other infor-
`mation relating to how successful the speech recognition
`unit was in obtaining a match. The “decision”216, which
`maybea identification,verification, and/or confidence level,
`is then used to “recognize” the user, meaning to identify or
`verify the user, or perform some other type of recognition.
`Either verification or identification may be performed with
`the system 201 shown in FIG.1. Should identification be
`preferred, the database 208 is cycled through in order to
`obtain the closest match.
`
`the
`[0045] Systems which may be used to implement
`speech recognition system of FIG. 1 are disclosed in U.S.
`Pat. No. 5,522,012, entitled “Speaker Identification and
`Verification System,” issued on May 28, 1996, patent appli-
`cation Ser. No. 08/479,012 entitled “Speaker Verification
`System,” U.S. patent application Ser. No. 08/
`5
`entitled “Model Adaption System And Method For Speaker
`Verification,” filed on Nov. 3, 1997 by Kevin Farrell and
`William Mistretta, U.S. patent
`application Ser. No.
`08/
`, filed on Nov. 21, 1997, entitled “Voice Print
`System and Method,” by Richard J. Mammone, Xiaoyu
`Zhang, and Manish Sharma, each of which is incorporated
`herein by reference in its entirety.
`[0046] Referring to FIG. 1, the speech recognition unit
`204 may contain a preprocessor unit 212 for preprocessing
`the speech prior to making any comparisons. Preprocessing
`may include analog to digital conversion of the speech
`signal. The analog to digital conversion can be performed
`with standard telephony boards such as those manufactured
`by Dialogic. A speech encoding method such as ITU G711
`standard 4 and A law can be used to encode the speech
`samples. Preferably, a sampling rate of 8000 Hzis used.
`
`[0047] The preprocessor unit may perform any numberof
`noise removal or silence removal techniques on the test
`speech, including the following techniques which are known
`in the art:
`
`[0048] Digital filtering to remove pre-emphasis. In
`this case, a digital filter H(z)=1-az"1 is used, where
`a is set between 0.9 and 1.0.
`
`[0049] Silence removal using energy and zero-cross-
`ing statistics. The success of this technique is pri-
`marily based on finding a short interval which is
`guaranteed to be background silence (generally
`found a few milliseconds at the beginning of the
`utterance, before the speaker actually starts record-
`ing).
`
`[0050] Silence removal based on an energy histo-
`gram. In this method, a histogram of frame energies
`is generated. A threshold energy value is determined
`based on the assumption that the biggest peak in the
`histogram at the lower energy region shall corre-
`spond to the backgroundsilence frame energies. This
`threshold energy value is used to perform speech
`versus silence discrimination.
`
`the speech recognition unit may
`[0051] Additionally,
`optionally contain a microprocessor-based feature extraction
`unit 214 to extract features of the voice prior to making a
`comparison. Spectral speech features may be represented by
`speech feature vectors determined within each frame of the
`processed speech signal. In the feature extraction unit 214,
`spectral feature vectors can be obtained with conventional
`methods such as linear predictive (LP) analysis to determine
`LP cepstral coefficients, Fourier Transform Analysis and
`filter bank analysis. One type of feature extraction is dis-
`closed in previously mentioned U.S. Pat. No. 5,522,012,
`entitled “Speaker Identification and Verification System,”
`issued on May 28, 1996 and incorporated herein by refer-
`ence in its entirety.
`
`[0052] The speech recognition unit 204 may be imple-
`mented using an Intel Pentium platform general purpose
`computer processing unit (CPU) ofat least 100 MHz having
`about 10 MB associated RAM memoryanda hard or fixed
`drive as storage. Alternatively, an additional embodiment
`could be the Dialogic Antares card.
`
`[0053] While the speech recognition systems previously
`incorporated by reference are preferred, other speech rec-
`ognition systems may be employed with the present inven-
`tion. The type of speech recognition system is notcritical to
`the invention, any known speech recognition system may be
`used. The present invention applies these speech recognition
`systems in the field of security to increase the level of
`security of prior, ineffective, systems.
`
`[0054]
`
`2. Security Methodology and Systems.
`
`[0055] According to the present invention, speaker recog-
`nition can provide varying levels of security based upon
`customer requirements. A biometric, such as voice verifica-
`tion, confirms the actual identity of the user. Other prevalent
`high security methods, such as token cards, can still be
`compromised if the token card is stolen from the owner.
`With speaker recognition, the user need know only a single
`piece of information, what to speak, and the voice itself
`supplies another
`identifying piece of information. The
`
`Page 22 of 30
`
`Page 22 of 30
`
`
`
`US 2003/0046083 Al
`
`Mar. 6, 2003
`
`present invention contemplates at least three levels of secu-
`rity, “simple” security, multi-tiered security, and randomly
`prompted voice tokens.
`
`[0056] A more general depiction of a speaker recognition
`system 215 is shownin FIG. 2. As shownin FIG.2, the user
`supplies a spoken password 217 to the speech recognition
`unit 204. The spoken password is preferably input into a
`microphone at the user’s location (not shown) or in the
`speech recognition unit 204 (not shown). The password may
`also be obtained from a telephone or other voice commu-
`nications device (not shown). In response to the spoken
`password, or subsequent data, the speech recognition unit
`204 outputs a decision 216, which may be or include a
`confidence level. To increase the level of security, an
`optional user index input unit 218 may be included to obtain
`index information, such as a credit card number, social
`security number, or PIN. The user index input umt 218 may
`be a keyboard, card reader, joystick, mouse, or other input
`device. The index may be confidential or public, depending
`on the level of security desired. An optional prompt input
`unit 220 may be included to prompt the user for a speech
`passwordor index information. The prompt input unit may
`be a display, speaker, or other audio/visual device.
`
`[0057] A “simple” security method 221 is shown in FIG.
`3. This method may be implemented in the system of FIG.
`1 or 2. The “simple” security system requires only the
`password and the voice biometric. This type of authentica-
`tion provides a security level typical of today’s token based
`systems. Thus, in FIG.3, a spoken password 224is obtained
`as well as optional index information 226. The password and
`index may be obtained from prompting 228 the user. This
`information is then processed in the speech recognition unit
`204. The speech recognition unit 204 attempts to recognize
`230 the speaker of the password (as belonging to the person
`identified by the index information,
`if entered). If the
`speaker is recognized, authorization is granted or the person
`is identified 232. If the speaker is not recognized, authori-
`zation is denied (i.e. not granted or a “no identity” result
`occurs 234). Optionally, the speech recognition unit’s deci-
`sion 216 is or includes a confidence level.
`
`[0058] A Multi-tiered security flow diagram is shown in
`FIG. 4A. The FIG. 4A method may be implementedin the
`systems of FIG. 1 or 2. The method 241 shown in FIG. 4A
`employs multiple tiers of spoken passwords to enhance
`security even further. For instance, a user is required to
`speak their selected password as well as additional randomly
`prompted information that is currently used for authentica-
`tion today, such as mother’s maiden name, birth date, home
`town, or SSN. A multi-tier system adds randomnessto the
`system to deter attacks through mechanisms such as digital
`recordings, as well as offers enhanced biometric validation.
`For example, if system performance typically authenticates
`with a 99.5% accuracy, a twotier system will authenticate at
`99.9975%, and a three tier system at 99.999988%. Addi-
`tionally, a multi-tier system checks both multiple pieces of
`knowledge and multiple biometric samples. Because speech
`is an easy to use, natural interface, the burden placed on the
`user for a multi-tier system will still be less then that of a
`token based system. This system can be language dependent
`or language independent.
`
`[0059] As shown in FIG.4A,a first speech passwordis
`obtained 242 from the user. Index information may also,
`
`optionally, be obtained 244 from the user. After receiving the
`first speech password and optional index information, the
`voice recognition unit 204 prompts 246 for a second (ran-
`dom) password 246. The prompt may be displayed by the
`prompt input unit 220 of FIG. 2. Next, the second speech
`password is obtained 248. The voice recognition unit 204
`then determines whether it recognizes the first password
`250. If the first password is not recognized, there is no
`authorization or identification 252. If the first password is
`recognized, the voice recognition unit determines whetherit
`recognizes the second password 251. If the second password
`is not recognized there will be no authorization or identifi-
`cation 252. If the second password is recognized, authori-
`zation and/or identification will occur 254. Optionally, a
`confidence level is output as, or included in, the decision
`216.
`
`[0060] A two-tier system may be made conditional on
`rejection of a first password. FIG. 4B shows a conditional
`two-tier system 261. As shown in FIG. 4B, a first speech
`password is obtained 262. Optionally, index information is
`also obtained 264. The speech recognition unit 204 then
`determines whether it recognizes the first password 266. If
`the first password is recognized, authorization and identifi-
`cation will occur 268.
`
`If the speech recognition unit does not recognize
`[0061]
`the first password, it generates a second (random) password
`270. The second password is randomly generated by the
`speech recognition unit 204. A promptfor this password may
`be displayed 271 on a prompt input unit 220 (FIG. 2). The
`second speech password is obtained 272, and if the second
`password 270 is recognized 274, authorization or identifi-
`cation occurs 278. If the second password is not recognized,
`no authorization or identification takes place 268. Option-
`ally, the decision 216 may comprise, or include a confidence
`level.
`
`[0062] A randoml