Spoken free-form passwords for light-weight speaker verification using standard speech recognition engines

[0001] This continuation-in-part application claims the benefit of U.S. patent application Ser. No. 11/615,900 filed Dec. 22, 2006, which is incorporated by reference herein.

[0002] 1. Field of the Invention

[0003] The present invention relates to the field of speech processing, and, more particularly, to spoken free-form passwords for light-weight speaker verification using standard speech recognition engines.

[0004] 2. Description of the Related Art

[0005] Conventional password security systems are generally based on a textual password. A proper password is stored within a security system which is compared against user input. When a successful match is made, the user is authenticated, otherwise user access to a secure resource can be denied. One problem with text based passwords is that unauthorized users (e.g., hackers) are sometimes able to access a security system's password table which grants them access to all associated secured resources. Additionally, numerous malicious programs and electronic devices, such as keyloggers, are specifically designed to capture text based passwords from authorized users.

[0006] Many security systems utilize biometric input to enhance security. Biometric input can include distinctive physical or behavioral characteristics of a user which are used to identity or verify a user. A common biometric technology is a speech processing technology referred to as speaker identification and verification (SIV). With SIV, users can be identified and verified utilizing specialized speech engines. SIV technologies require users to participate in a training session so that user-specific vocal tract characteristics can be determined. These characteristics or speech features can be stored for subsequent use. Subsequent to the SIV training, a user can be prompted by a SIV system to speak a system generated phrase. A responsively supplied user utterance can be SIV analyzed by a speech processing system to determine whether speech features extracted from the user utterance match the stored speech features. SIV utilizes specialized speech processing technologies that add a significant cost to a security system, even one already possessing speech recognition capabilities. In many situations, the benefits resulting from including SIV technologies do not justify the associated costs. What is needed is a means for enhancing security, which is resistant to malicious attacks and that does not require costly infrastructure upgrades.

[0012] FIG. 1 is a schematic diagram of a system 100 for a speech system that includes speaker-dependent recognition grammars for passwords in accordance with an embodiment of the inventive arrangements disclosed herein. In system 100, a user 110 can interact with a speech system 120, which includes an automated speech recognition (ASR) engine 122. A data store 124 can be used by system 120 to store one or more speech recognition grammars. The grammars of store 124 can include a password grammar 126. The speech system 120 can permit user 110 to provide audio from which recognition grammar entries are generated. These generated entries can include a password entry which is used to authenticate the user 110 for a secure resource. The password entry can be free-form in nature and can be any user 110 supplied utterance which can include any utter-able sound, word, phrase, etc. An acoustic baseform 128 can be generated from the password containing utterance which can also be recorded 129 and stored.

[0013] A user's interactions with the speech system 120 can occur through many different clients or interfaces. In one embodiment, user 110 can utilize a Web browser 112 to interact with Web server 114 provided content. Served Web pages can be speech-enabled content which is processed by speech system 120. In another embodiment, a voice only interface/device 116 can communicate with a voice response system 118 which uses the speech system 120 for speech processing operations. In still another embodiment, the speech system 120 can be part of an integrated device, such as a computer, kiosk, or mobile device, having an audio transducer 119 for accepting and presenting audio to and from the user 110.

[0014] The speech system 120 can be a commercial off-the-shelf speech processing system. The speech system 120 can acoustically generate baseforms using a variety of known techniques, such as those disclosed in the cross-referenced application entitled “SOLUTION THAT INTEGRATES VOICE ENROLLMENT WITH OTHER TYPES OF RECOGNITION OPERATIONS PERFORMED BY A SPEECH RECOGNITION ENGINE USING A LAYERED GRAMMAR STACK”, U.S. patent application Ser. No. 11/615,900, filed Dec. 22, 2006. The ASR engine 122 can be a standard speech recognition engine instead of a speaker recognition engine which incorporates speaker identification and verification (SIV) technologies. Thus, the system 100 leverages ASR technologies to achieve lightweight speaker verification capabilities without the overhead or infrastructure requirements of a full function SIV system. Because system 100 stores baseforms 128 as password, the system 100 is more resistant to many malicious attacks than other systems that store and use text based passwords.

[0015] FIG. 2 is a flow chart of a method 200 for creating and using spoken free-form passwords to authenticate users in accordance with an embodiment of the inventive arrangements disclosed herein. The method 200 can be performed in the context of a system 100 or any system having speech recognition capabilities and an ability to acoustically generate and use speaker dependent grammars. The method 200 includes a process 205 to establish a password and a process 225 to utilize established passwords.

[0016] The password establishment process 205 can begin in step 210, where a user can be prompted to audibly provide a password. The password can be free-form and can include any user generated utterance, such as a word, a phrase, or any other noise. In one embodiment, the utterance used for the password is used to generate an acoustic baseform and is not converted into text. Consequently, the utterance can be in any language or dialect and can include slang. The flexibility of the free-form utterance advantageously permits a user to create a highly unique password which is easy for the user to remember. Further, use of an acoustic baseform as a password is uniquely associated with a user's voice and is not readable by others (unlike textual passwords). Thus, acoustic baseform passwords are difficult for unauthorized users to steal by invading (i.e., hacking into) a security system.

[0017] In step 212, free-form audio input can be received in response to the password prompt. In step 214, a consistency check can optionally be performed against the free-form input. The consistency check can determine if the received audio input is sufficiently consistent with previously received utterances. If not, then the process can proceed from step 214 to step 216, where the user can be re-prompted for the password. The process can loop from step 216 to step 212, where audio input for the re-prompted password can be received. When consistency is achieved in step 214, the method can move to step 218, where an acoustic baseform for the password can be added to a speaker-dependent speech recognition grammar.

[0018] The password utilization process 225 can begin in step 230, where a user can be prompted for an identifier/account number. In step 232, results from the prompting can be processed and used to determine a user identity. In step 234, a recognition grammar associated with the user and a password context can be determined. In step 236, a user can be prompted for a password. In step 238, an utterance can be received from the user. In step 240, the utterance can be acoustically matched against the password recognition grammar.

[0019] In step 242, a confidence score generated from the matching can be compared against a minimum confidence threshold. When the score meets or exceeds the threshold, the process can progress from step 242 to step 244 where the user can be authorized to utilize the secure system. When the threshold is not exceeded, the process can progress from step 242 to step 246, where a determination can be made as to whether a maximum number of attempts has been made. If so, the user can be presented with an access denied message in step 248. If the maximum number of established attempts is not exceeded, the process can loop from step 246 to step 238 where another utterance can be received from the user.

[0020] FIG. 3 illustrates a speech dialog showing an instance where a spoken free-form password is created and used in accordance with an embodiment of the inventive arrangements disclosed herein. The speech dialog can be performed in the context of a system 100 or a method 200. The speech dialog includes a password establishment dialog 310 and a password usage dialog 340. Both dialogs 310 and 340 illustrate communications between an automated system 312, 342 and a user 314, 344.

[0021] In dialog 310, a voice prompt 320 can be audibly presented that prompts a user to speak a free-form password. A spoken response 322 of “jack sprat ate no fat” can be spoken in response. In one embodiment, a confusable phrase grammar can be queried to ensure that the spoken response 322 is not acoustically similar to pre-existing system commands, such as cancel, stop, quit, main menu, and the like. In dialog 310, the user supplied password is not likely to be confused with pre-existing commands.

[0022] To insure that new passwords will be accurately recognized when provided to a speech recognition engine, the automated system can prompt the user to repeat the password to insure a minimum number of consistent baseforms are generated. Thus, the system can re-prompt 324 a user, who repeats the password 326. A third prompting 328 can result in the password being uttered a third time 330.

[0023] In the dialog 310, an enrollment session can successfully complete after three successful consistent pronunciations are received. The best audio (322, 326, and/or 330) of the uttered password can be saved (i.e., a Media Resource Control Protocol (MRCP) save-best-waveform support function can be used in one configuration). The automated system can inform 332 a user that the phrase has been successfully enrolled. Additional information, such as a reminder to remember the password, a re-playing of the spoken password, recording size in bytes, recording duration in milliseconds, and the like, can optionally be presented 332 at this point in the dialog 310. The user provided acoustic baseform can be saved in an indexed fashion in a record or a file associated with the user. In one embodiment, the user utterances provided in dialog 310 can be saved and used to calibrate the automated system.

[0024] In dialog 340, a voice prompt 350 can ask a user to speak a user identifier, which the user speaks 352 in response. The automated system can then prompt 354 for a password. The user can utter 356 “jack sprat ate no fat.” This is the same password created in dialog 310. A speech recognition system can compare the utterance against a stored baseform created in dialog 310, which results in a match with a relatively high confidence score. If the confidence score falls below a previously established threshold, the system can re-prompt the user. Otherwise, the system can successfully authenticate a user which results in the system presenting 358 a welcome message.

[0025] FIG. 4 is a table 400 illustrating sample values of a test scenario of multiple different users speaking a stored password in accordance with an embodiment of the inventive arrangements disclosed herein. The table 400 shows three columns including a speaker's identity 402, an average confidence score 404, and a system's confidence in a claimant 406. The table 400 is based on real-world test input for a spoken password of “jack sprat ate no fat.” The table 400 assumes that each speaker 412 had somehow acquired an owner's 410 password and that multiple password requests are made to obtain the average values (404 and 406) for table 400.

[0026] More specifically, table 400 shows that a password owner 410 has an average confidence score of approximately 0.67 and a speaker confidence value of one hundred percent. Each other speaker 412 has lower values for the average confidence score and speaker confidence value. The table 400 shows only a password owner can be one hundred percent authenticated by a free-form password system, such as system 100. Thus, use of acoustic grammars spoken by a user into a standard speech recognition system increases system security. In one embodiment, confidence thresholds (over 0.665 in the example) can be established so that only an authorized speaker will be authenticated. The confidence threshold can be lowered to permit variance in the owner's 410 spoken password to still be accepted, which may result in some unauthorized access, such as Female Hacker #1 and/or Male Hacker #6. Thus, an established confidence score can be adjusted depending on whether a system is more concerned that some spoken utterance by an owner 410 is invalidated or is more concerned that some spoken utterance by others will be improperly validated.

[0027] The present invention may be realized in hardware, software, or a combination of hardware and software. The present invention may be realized in a centralized fashion in one computer system, or in a distributed fashion where different elements are spread across several interconnected computer systems. Any kind of computer system or other apparatus adapted for carrying out the methods described herein is suited. A typical combination of hardware and software may be a general purpose computer system with a computer program that, when being loaded and executed, controls the computer system such that it carries out the methods described herein.

[0028] The present invention also may be embedded in a computer program product, which comprises all the features enabling the implementation of the methods described herein, and which when loaded in a computer system is able to carry out these methods. Computer program in the present context means any expression, in any language, code or notation, of a set of instructions intended to cause a system having an information processing capability to perform a particular function either directly or after either or both of the following: a) conversion to another language, code or notation; b) reproduction in a different material form.

[0029] This invention may be embodied in other forms without departing from the spirit or essential attributes thereof. Accordingly, reference should be made to the following claims, rather than to the foregoing specification, as indicating the scope of the invention.