JP2007052496A - User authentication system and user authentication method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide user authentication that can achieve rigidity with two-element authentication by combining password authentication and voiceprint authentication, can shorten user authentication time, does not require complicated work, and has high operability. <P>SOLUTION: An utterance acquiring section 16 is connected to an interactive voice responding device 3 from a user call terminal 1 with a line, and acquires utterance voice of a password input into the user call terminal 1 via a telephone network 2. Password acquiring voice is input into a voice recognition server 5, and it is determined by voice recognition whether the acquired password utterance voice matches with a previously registered password. After success in password authentication, the password utterance voice is input into a voiceprint authentication section 27, and it is determined whether voiceprint data of the password utterance voice matches with registered voiceprint data in a user profile of the user. The voiceprint authentication result is reported from an authentication result reporting section 28 to a user management server 7. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a user authentication system and a user authentication method that can be applied to a voice authentication station that receives voices captured from a user terminal via a communication network and performs user authentication in real time.

  Currently, on the Internet, user authentication, which is confirmation of the identity of a user, is performed in various situations such as commercial transactions and site access. For example, the user himself / herself decides or gives a user ID and password, and when authentication is required, the user ID and password are entered on the WWW screen or dialog displayed on the user terminal, and authentication is performed. User authentication is performed by checking with a user ID and password registered in advance in the server. However, the method of authenticating with the user ID and password has a problem that if the user ID and password are leaked, spoofing is easily allowed.

Therefore, an authentication method combining password authentication and voiceprint authentication has been proposed as a method for performing user authentication easily and inexpensively while improving security (for example, see Patent Document 1). In the authentication method disclosed in this patent document, after receiving a password via a network and authenticating the password, the user's voice is acquired using a telephone line which is a communication line different from the network, and the voice print of the user voice is obtained. The user is authenticated based on the information.
JP 2004-13274 A

  However, after the password is transmitted from the user terminal through the network and the password is authenticated, the user voice is transmitted by using a telephone line which is a different communication line and the voiceprint authentication is performed by a password transmission operation through the network. Since there are two separate operations for making a call on the telephone line, there is a problem that it takes a long time to complete user authentication, and the operation for user authentication itself is troublesome twice. There is a problem.

  The present invention has been made in view of the above circumstances, and can maintain a robust combination of password authentication and voiceprint authentication while reducing user authentication time and realizing high-speed responsiveness. It is an object of the present invention to provide a user authentication system and a user authentication method that do not require switching and have improved operability.

  The user authentication system according to the present invention includes an utterance acquisition unit that acquires an utterance voice of a password input to a user call terminal via a communication network, and whether or not the acquired password utterance voice matches a registered password registered in advance. Voice recognition means for judging by voice recognition, and voiceprint authentication means for judging whether or not the voiceprint data of the password utterance voice matches the registered voiceprint data in the user profile of the user.

  According to the user authentication system configured as described above, password authentication and voiceprint authentication are performed using the uttered voice of the password input to the user call terminal, and thus robustness combining password authentication and voiceprint authentication is achieved. It can be realized, and the user authentication time can be shortened and high-speed response can be realized. In addition, since it is not necessary to switch communication lines, operability can be improved.

  In the user authentication system according to the present invention, the user profile includes a plurality of patterns of voiceprint data extracted from password utterances having different utterance speeds, and the voiceprint authentication means includes the password utterance acquired at the time of user authentication. It is determined whether or not the voice print data matches with any one of a plurality of patterns of voice print data in the user profile.

  Thereby, since the voiceprint authentication is performed by collating with the voiceprint data of a plurality of patterns extracted from the password utterance voices having different utterance speeds, it is possible to effectively prevent an erroneous determination of the voiceprint authentication. The voice rate data fluctuates due to changes in the speech rate or vocal cord tension depending on the user's state of mind such as the user's psychological state and health status at the time of password utterance, but multiple patterns with different utterance rates should be registered It can respond flexibly.

  In the user authentication system, the present invention also provides user information for acquiring the user's registered password from the user management terminal that manages the registered password based on the identification information of the user call terminal that requested user authentication or the identification information of the user. And a keyword dictionary creating means for converting the registered password obtained by the user information obtaining means into a keyword dictionary used by the voice recognition means for voice recognition.

  As a result, when the voice recognition means can recognize the password utterance using the keyword dictionary, the password authentication succeeds, and when it cannot be recognized, the password authentication fails. Since only the password registered by the user is converted into the keyword dictionary, the time required for recognition and the recognition accuracy can be significantly increased as compared to the case where speech recognition is performed using a large number of keyword dictionaries.

  According to the present invention, the user authentication system further comprises a user profile restructuring unit that registers the voiceprint data of the password uttered voice as voiceprint data in the user profile when the voiceprint authentication unit succeeds in the voiceprint authentication. Features.

  As a result, the user profile is reconstructed using the voiceprint data that has been successfully voiceprinted, so even if the user's voiceprint data changes due to changes in the user's body shape or vocal cords, the user can use the latest voiceprint data obtained at the time of user authentication. Since the profile is flexibly updated, erroneous determination of voiceprint authentication can be prevented.

  In the user authentication system, when voiceprint data is newly registered, the present invention provides a period during which no utterance is input to the user call terminal, and the environment on the user side is acceptable based on a received signal during the period during which no utterance is input. Noise determination means for determining whether or not the noise level is acceptable, and after determining that the noise determination means has an acceptable noise level, voice print data is extracted from voice utterance voice data for voice print registration and the user profile And a user profile creation means for registration.

  As a result, after confirming that the user's environment has an acceptable noise level, a password utterance for new registration of voiceprint data is performed, so erroneous determination due to low accuracy of registered voiceprint data Can be effectively prevented.

  In the user authentication system, the present invention further includes sound crack determination means for determining whether or not the input voice exceeds an input allowable level from the utterance period of the password utterance voice data for voiceprint registration. Voiceprint data is extracted after it is determined that the input allowable level is satisfied.

  As a result, if there is an input voice that exceeds the input allowable level, it is excluded from the object to be extracted as voiceprint data, so that the accuracy of voiceprint authentication can be improved.

  Further, according to the present invention, in the user authentication system, when voiceprint data is newly registered, the noise on the user side is acceptable from the state of at least one predetermined section before and after the utterance period from the password utterance voice data for voiceprint registration. Noise determination means for determining whether or not a level, and user profile creation means for extracting voice fingerprint data from an utterance period of voice utterance voice data for voiceprint registration and registering it in the user profile To do.

  As a result, it is determined whether or not the environment on the user side has an acceptable noise level from the state of the predetermined section of the password utterance voice data for voiceprint registration, so there is no need to provide a period during which no utterance is input for noise determination. The time for registering a new voiceprint can be shortened.

  In the user authentication system, the present invention further includes sound crack determination means for determining whether or not the input voice exceeds an input allowable level from the utterance period of the password utterance voice data for voiceprint registration. Voiceprint data is extracted after it is determined that the input allowable level is satisfied.

  As a result, it is possible to prevent voiceprint data based on the input voice in which sound cracking has occurred because the input voice exceeds the input allowable level, and to prevent erroneous authentication based on the sound cracking data.

  In the user authentication system, when newly registering voiceprint data, the password utterance for voiceprint registration is performed at least three times, and the first password utterance voice is compared with the second password utterance voice. Calculating the identity accuracy, comparing the first password utterance speech with the third password utterance speech, calculating the identity accuracy, comparing the second password utterance speech with the third password utterance speech, and the identity accuracy And when all the calculated personal identities exceed a predetermined value, the password utterances are adopted.

  As a result, the person uttered twice, but even if another person uttered once, it can be determined that there is a problem only with the other person's utterance, and a second utterance can be requested. The registration act that causes confusion is eliminated, and the number of utterance requests can be suppressed to the minimum number.

  Further, the user authentication method of the present invention includes a step of acquiring the speech utterance of the password input to the user call terminal via the communication network, and whether or not the acquired password utterance speech matches a registered password registered in advance. A step of determining by voice recognition, and a step of determining whether or not the voiceprint data of the password utterance voice matches the registered voiceprint data in the user profile of the user.

  ADVANTAGE OF THE INVENTION According to this invention, the user authentication system which can implement | achieve robustness by the two-factor authentication which combined password authentication and voiceprint authentication, can shorten user authentication time, and was excellent in the operativity which does not require a complicated operation, and a user An authentication method can be provided.

Hereinafter, an embodiment of the present invention will be specifically described with reference to the drawings.
FIG. 1 is a block diagram showing the configuration of a system to which a user authentication system according to an embodiment of the present invention is applied. The user call terminal 1 has a function of inputting a user's uttered voice and a communication function of enabling communication by connecting to the telephone network 2 by wire or wireless. As the user call terminal 1, a mobile phone terminal, a fixed phone terminal, a PDA equipped with a call function, or the like can be used. In this embodiment, an example using a mobile phone terminal will be described.

  On the other hand, the voice certificate authority performs voiceprint registration and user authentication in response to a request from the user call terminal 1. The voice authentication station includes an interactive voice-compatible device 3 that provides voice guidance to the user call terminal 1 connected via the telephone network 2, a voiceprint authentication server 4 that performs voiceprint authentication based on a user profile, and voice recognition of speech content (password). And a user management server 7 having a user database 6 in which registered user IDs and passwords are registered as main components. In the present embodiment, the voice certificate authority system is constructed by connecting each component (3 to 7) via a network such as a LAN or WAN. However, the voice certificate authority is constructed on one server without being distributed. Alternatively, only specific components (for example, the interactive voice support device 3, the voiceprint authentication server 4, and the voice recognition server 5) may be constructed on the same server.

  FIG. 2 is a block diagram showing the functions of the interactive speech support apparatus 3. The voice guidance support unit 11 provides voice guidance for voiceprint registration and voice guidance for user authentication to the user call terminal 1 connected to the line. The voice guidance data is stored in a memory (not shown) and is read according to a predetermined sequence.

  The caller number acquisition unit 12 is a part that acquires the caller number of the user call terminal 1 from the caller number notification signal of the user call terminal 1 that has requested connection to the interactive voice-compatible device 3. The acquired caller number is notified to the voiceprint authentication server 4 to inquire whether the user is registered. In this example, the user management server 7 is inquired of the presence / absence of user registration via the voiceprint authentication server 4. However, the user management server 7 may be directly inquired if only the presence / absence of registration is confirmed.

  The user information creation unit 13 creates user information and notifies the voice print authentication server 4 when registration is confirmed as a result of the registration inquiry by the caller number acquisition unit 12. In the case of new voiceprint registration, notification of user information from the user information creation unit 13 to the voiceprint authentication server 4 becomes a trigger for acquiring user profile data excluding voiceprint data and creating a keyword dictionary. In the case of voiceprint authentication, notification of user information from the user information creation unit 13 to the voiceprint authentication server 4 serves as a trigger for creating a keyword dictionary. In this example, the caller number is used as the user information, but the user ID acquired at the time of user registration confirmation may be used.

  The noise determination unit 14 and the spoken voice level determination unit 15 are functional blocks that are activated in the case of new voiceprint registration. The noise determination unit 14 determines whether the surrounding environment of the user who performs voiceprint registration is a noise situation in which voiceprint registration is possible. When the noise level exceeds the allowable value, the voice guidance corresponding unit 11 causes the user call terminal 1 to output a guidance result indicating that the voiceprint registration is impossible. The utterance voice level determination unit 15 asks the user to utter predetermined contents, and determines the utterance voice level at that time. In order to guide the utterance voice level of the keyword utterance voice to be registered as a voiceprint to an appropriate level, it is performed in order for the user to recognize the appropriate utterance voice level prior to the keyword utterance. The utterance voice level determination unit 15 determines whether the loudness of the user's voice (utterance voice level) is not too low or too high. When the utterance voice level is inappropriate, voice guidance such as “loud voice” and “lower voice” is output from the voice guidance corresponding unit 11 to the user call terminal 1.

  The utterance acquisition unit 16 acquires a keyword utterance voice by the user from the voice guidance support unit 11. The acquired keyword utterance voice data is sent to the profile creation requesting unit 17 and the voice recognition server 5 at the time of new voiceprint registration, and sent to the authentication requesting unit 18 and the voice recognition server 5 at the time of user authentication.

  The profile creation request unit 17 is a functional block that is activated when a new voiceprint is registered. The keyword utterance voice data acquired by the utterance acquisition unit 16 is transmitted to the voiceprint authentication server 4. The profile creation requesting unit 17 requests the voiceprint authentication server 4 to create voiceprint data from the keyword utterance voice data and register the profile.

  The authentication request unit 18 is a functional block that is activated during voiceprint authentication. The keyword utterance voice data acquired by the utterance acquisition unit 16 is transmitted to the voiceprint authentication server 4 to request voiceprint authentication.

  FIG. 3 is a block diagram showing functions of the voiceprint authentication server 4. The user registration confirmation unit 21 notifies the user management server 7 of the caller number notified from the caller number acquisition unit 12, and makes an inquiry as to whether or not the user to which the call number is assigned is registered. Information about the presence / absence of user registration returned from the user management server 7 is notified to the caller number acquisition unit 12.

  The user information acquisition unit 22 uses the caller number received from the user information creation unit 13 to acquire the user information (user ID, password) registered with the user management server 7 in association with the caller number. To emit. If the user ID is acquired at the time of user registration confirmation, the user ID can be used instead of the caller number. In the case of new voiceprint registration, the user information included in the response from the user management server 7 is input to the user profile creation unit 25 and the password is input to the keyword dictionary creation unit 23. In the case of user authentication, user information is not input to the user profile creation unit 23, but a password is entered only into the keyword dictionary creation unit 23.

  The keyword dictionary creation unit 23 receives password data from the user information acquisition unit 22 and creates a keyword dictionary that the speech recognition server 5 uses for speech recognition from the password data. In the present embodiment, a registered password of a user who is a voiceprint registration target or a registered password of a user who is a voiceprint authentication target is extracted and expanded into a keyword dictionary format, and the keyword dictionary corresponding to the registered password Only will be prepared. Therefore, when the speech recognition engine recognizes speech by referring to the keyword dictionary, speech recognition can be performed if the spoken speech matches the registered password, but speech recognition fails if the speech speech is other than that. become. That is, if the speech recognition using the keyword dictionary corresponding to the registered password is successful, it means that the password authentication is successful and the password authentication is successful. In the present embodiment, the voice recognition server 5 performs voice recognition. If password authentication is successful, user profile creation and voiceprint authentication are performed. However, if password authentication fails, the process ends without moving to user profile creation or voiceprint authentication.

  The user profile creation unit 25 is activated in response to a profile creation request from the profile creation request unit 17. The activated user profile creation unit 25 extracts voiceprint data from the password utterance voice data passed from the profile creation request unit 17. Then, a user profile in which the extracted voiceprint data is associated with the user information passed from the user information acquisition unit 22 is created and registered in the user profile database 26. In the present embodiment, the user information includes a telephone number, a user ID, and a password. However, not all of these items need to be included.

  The voiceprint authentication unit 27 is activated in response to a voiceprint authentication request from the authentication request unit 18. The activated voiceprint authentication unit 27 extracts voiceprint data from the password utterance voice data passed from the authentication requesting unit 18 and extracts registered voiceprint data from the user profile database 26 using the telephone number (sender number) as a key. Then, the voiceprint data extracted from the password utterance voice data is compared with the registered voiceprint data extracted from the user profile database 26. If the similarity score exceeds a predetermined value, the voiceprint authentication is successful, and the score must exceed the predetermined value. If the voice print authentication fails, the authentication result is output.

  The authentication result notifying unit 28 notifies the user management server 7 of the authentication result output from the voiceprint authenticating unit 27 together with the telephone number that is the user specifying information. Note that the notification destination to which the authentication result notification unit 28 notifies the authentication result is not limited to the user management server 7, and the notification destination can be changed according to the system configuration corresponding to the purpose and purpose. For example, it can be transmitted directly to the requesting user call terminal 1 or can be notified to another application server.

  The user profile reconstruction unit 29 is a part that registers the voiceprint data extracted by the voiceprint authentication unit 27 when user authentication is performed in the user profile database 26 and reconstructs the user profile.

  The voice recognition server 5 includes a voice recognition engine. The voice recognition engine recognizes the password utterance voice data given from the utterance acquisition unit 16 with reference to the keyword dictionary created by the keyword dictionary creation unit 23. If the keyword dictionary created by the keyword dictionary creation unit 23 is only for one password, only that one password can be recognized. Therefore, the success of voice recognition means the success of password authentication, and the failure of voice recognition means the failure of password authentication. The speech recognition result is transmitted to the interactive speech support apparatus 3.

  Next, the operation of the present embodiment configured as described above will be specifically described by dividing it into a new voiceprint registration operation and a user authentication operation. It is assumed that a user ID, a password, and other user information are registered in advance in the user management server 7 in order to receive an authentication service from this system that functions as a voice certificate authority.

  FIG. 4 is a flowchart when registering a new voiceprint. The user uses the user call terminal 1 to dial the telephone number of the interactive voice response device 3 and connect the line. In the interactive voice response device 3, when there is an incoming call from the user call terminal 1, the caller number acquisition unit 12 acquires the caller number of the user call terminal 1. If the caller ID is set to non notification, the voice guidance is given so that the calling number is set and the call is repeated. The caller number acquired by the caller number acquisition unit 12 is sent to the user registration confirmation unit 21. The user registration confirmation unit 21 inquires of the user management server 7 whether or not the user having the caller number is registered. The registration presence / absence information returned from the user management server 7 to the user registration confirmation unit 21 is notified to the caller number acquisition unit 12. If the user is not registered, the voice guidance support unit 11 performs voice guidance to the effect that the user is not registered, and ends the process.

  If the user has been registered, the process proceeds to a new voiceprint registration process. The user information creation unit 13 notifies the user information acquisition unit 22 of the caller number and starts preparation for creating a user profile. Since the voice guidance corresponding unit 11 makes a noise determination, a voice guidance for instructing not to speak is issued.

  The user information acquisition unit 22 acquires user information of a registered user corresponding to the caller number from the user management server 7. The acquired user ID, password, and telephone number are input to the user profile creation unit 25, and password data is input to the keyword dictionary creation unit 23 for password authentication.

  The noise determination unit 14 determines the noise situation on the user side from the received voice data when not speaking. If the noise level exceeds a predetermined value, the voice guidance is performed so as to make a call again from a slightly quieter environment, and the process is terminated. If the noise level does not exceed the predetermined value, voice guidance is given so that the user himself / herself speaks.

  Thus, by performing voiceprint registration in a quiet environment where the noise level does not exceed a predetermined value, the accuracy of voiceprint authentication can be greatly improved. The threshold of the noise level can be set according to the use and purpose, and the accuracy of voiceprint authentication can be increased by increasing the threshold.

  Next, the user is prompted to speak by voice guidance, and the speech voice level of the user is determined from the speech content spoken by the user. For example, utterance is urged by voice guidance so that the company name, affiliation, and name are spoken. The utterance voice level determination unit 15 measures the user's utterance voice level from the received voice signal. If the speech level is determined to be lower than the specified value, a voice guidance with a content such as “Please generate a louder voice” is output. If the level is higher than the specified value, “Slightly lower” Voice guidance with the content “Please generate with voice” is output. Accurate voiceprint data cannot be collected when the voiceprint size is too small or too large at the time of voiceprint registration. Therefore, before the password to be actually registered is uttered, the utterance voice level is measured in order to make the user recognize an appropriate utterance voice level.

  Here, when the utterance of the company name, affiliation, and name is completed, it is desirable to provide voice guidance so as to notify the user that the utterance has ended by pressing a specific button (for example, # button) on the user call terminal 1. Since the voice guidance support unit 11 can move to the next processing by detecting the pressing of the # button, the waiting time in the interactive voice support device 3 can be shortened, leading to a reduction in voiceprint registration time.

  Next, the voice guidance support unit 11 performs voice guidance so that the user speaks the password registered in the user management server 7 a plurality of times. In the present embodiment, it is assumed that the mobile phone number of the mobile phone used by the user is registered in the user management server 7 as a password. For example, a voice guidance such as “Perform first voiceprint registration. Please tell us the mobile phone number of the registered mobile phone” is output. The user speaks the registered mobile phone number to the user call terminal 1.

  First, the spoken mobile phone number is voice-recognized by the voice recognition server 5, and it is determined whether or not it matches the mobile phone number registered in the user management server 7. Specifically, the utterance acquisition unit 16 acquires the utterance voice of the mobile phone number and transmits it to the voice recognition server 5. On the other hand, the keyword dictionary creation unit 23 creates a keyword dictionary from the registered mobile phone number acquired from the user management server 7 based on the caller number. For example, a registered mobile phone number (03-1234...) Represented by a number is converted into pronunciation data (zero sanity sanyon...) That is a way of reading the registered mobile phone number. This pronunciation data is used as a keyword dictionary.

  The speech recognition server 5 acoustically analyzes the speech of the mobile phone number received from the speech acquisition unit 16 and recognizes the speech using the keyword dictionary created by the keyword dictionary creation unit 23. Here, the keyword dictionary created this time by the keyword dictionary creation unit 23 is only for one mobile phone number given from the user information acquisition unit 22. That is, the speech data that can be recognized this time by the speech recognition server 5 that performs speech recognition using the keyword dictionary is only the registered mobile phone number prepared this time as a keyword dictionary. For this reason, when the voice recognition server 5 succeeds in the voice recognition of the speech of the mobile phone number, the mobile phone number received from the speech acquisition unit 16 and the registered mobile phone number received from the user information acquisition unit 22 match. become. In the present embodiment, since the mobile phone number is registered as a password, the fact that the mobile phone number of the uttered voice can be recognized by voice means that the password matches and the password authentication is successful. On the other hand, if the voice recognition server 5 cannot recognize the speech of the mobile phone number, the password authentication does not match and the password authentication has failed.

  In this way, since the speech recognition server 5 only performs matching with one keyword dictionary, speech recognition can be completed in a short time, and since there is only one candidate, a large number of candidates (a large number of candidates) The recognition accuracy can be greatly improved as compared with the method of selecting the keyword having the highest similarity from the keyword dictionary.

  The keyword dictionary creating unit 23 expands the password into the keyword dictionary every time voice recognition is performed, but stores the password expanded in the keyword dictionary format in the database of the user management server 7 in advance. A keyword dictionary corresponding to the designated password may be taken out instead of the password and can be referred to from the voice recognition server 5. By configuring in this way, it is possible to save resources required for creating a dictionary every time.

  The voice recognition server 5 returns the voice recognition result (recognition success / recognition failure) to the utterance voice of the mobile phone number to the utterance acquisition unit 16. When the utterance voice word is a registered mobile phone number, “recognition success” is returned, and when the utterance voice word is other than the registration mobile phone number, “recognition failure” is returned.

  The utterance acquisition unit 16 switches the content of the voice guidance for the user call terminal 1 according to the recognition result returned from the voice recognition server 5. When “recognition failure” is returned, a voice guidance indicating that the mobile phone number does not match the registered mobile phone number is output, and the process returns to the beginning of the first voiceprint registration process. If “recognition success” is returned, the voiceprint registration is performed for the second time, so that the registered mobile phone number is uttered and a voice guidance instructing to press a specific button (for example, # button) is output. If recognition failure by the voice recognition server 5 is repeated a predetermined number of times (for example, three times) during voiceprint registration, voice guidance indicating that voiceprint registration has not been performed is output, and the registration process is terminated.

  On the other hand, when “recognition success” is returned from the voice recognition server 5 in the first voiceprint registration, the voice guidance is output to perform the second voiceprint registration, and the utterance acquisition unit 16 sends the profile creation requesting unit 17. On the other hand, the voice data of the mobile phone number used for the voice recognition is input. The profile creation requesting unit 17 issues the user profile creation request by supplying the speech data of the mobile phone number to the user profile creating unit 25.

  The user profile creation unit 25 acoustically analyzes the speech voice data of the mobile phone number and extracts voiceprint data. The user profile creation unit 25 creates a user profile in which user information (user ID, telephone number, password) acquired by the user information acquisition unit 22 based on the caller number is registered, and registers the user profile in the user profile database 26. Further, the user profile creation unit 25 adds the voiceprint data extracted this time to the user profile of the user. As a result, the user profile including the first voiceprint data, user ID, telephone number, and password is registered. This completes the first voiceprint registration.

  In the second voiceprint registration and the third voiceprint registration, the same processing as the first voiceprint registration is repeated, and the second and third voiceprint data are sequentially registered in the same user profile. In this example, when the third voiceprint data registration is completed, the registration result is notified from the authentication result notification unit 28 to the user management server 7. In addition, voice guidance to the effect that the voiceprint has been registered is made from the voice guidance corresponding unit 11 to the user call terminal 1.

  Note that it is desirable to provide sound crack determination means for determining whether or not the input voice to be registered as a voiceprint is cracked. The utterance acquisition unit 16 determines whether or not to issue a profile creation request in response to the determination result of the sound crack determination unit. If the sound is broken, voice guidance is given to prompt the user to speak the password again. If there is an input sound that exceeds the allowable input level, the input device can only quantify it at the maximum allowable input level. As a result, when returning from numerical data to sound, the sound far from the original sound Become. In view of this, the voice input data that is determined by the sound breaking determination means as not being broken is used as a voiceprint data extraction target. Or, let the user recognize the level of the input voice that does not break the sound, and then make a password utterance for voiceprint registration, and use voice guidance etc. so that the input voice does not exceed the input allowable level in the utterance at the time of voiceprint registration Lead.

  Here, the sound cracking determination includes a digital system and an analog system. In the sound crack determination by the digital system, when the sound pressure of the input voice is digitized, it can be determined that the sound is broken if the numerical value exceeds a certain level. It is difficult to judge by level in the sound crack judgment by analog system, so the input voice is processed by FFT (Fast Fourier Transform) and developed into the sound pressure distribution on the frequency axis. If it is distributed over the range, it can be determined that the sound is broken.

  Further, in the above description, voiceprint registration is performed for each voice input, but it is configured so that the mutual reliability determination is performed on the voiceprint data from the first time to the third time and then registered in the user profile. It is desirable to do. The mutual reliability determination is performed by a mutual reliability determination means (not shown).

  As described above, voice utterance for registering voiceprint data is performed at least three times. At the time of this utterance, the person himself usually utters three times. However, if a malicious user causes the other person to speak twice and the other person speaks, confusion occurs in later voiceprint authentication. These utterances are not subject to any of the keyword check, noise check, S / N ratio check, and sound crack check, and voiceprint data is created for the three utterances as that of the person. However, since one utterance actually belongs to another person, the exclusion ratio of other persons deteriorates as it is.

  In order to prevent this, a mutual reliability check is performed by the mutual reliability determination means. The identity accuracy of the first utterance and the second utterance is calculated, and this is set as A. The identity accuracy of the second utterance and the third utterance is calculated, and this is set as B. The identity accuracy of the third utterance and the first utterance is calculated and this is set as C. If any of A, B, and C does not exceed a certain threshold, voiceprint registration cannot be performed.

  If any of A, B, and C does not exceed a certain threshold, it is determined that there is a problem with all three password utterances, and all password utterances are requested again. In response to an instruction from the mutual reliability determination means, the utterance acquisition unit 16 outputs voice guidance requesting password utterances three times again according to the procedure described above.

  If A and B are bad and C is good, it is judged that there is no problem in the first password utterance and the third password utterance, and there is a problem in the second utterance, and the second password utterance is requested again. To do. In response to the instruction from the mutual reliability determination means, the utterance acquisition unit 16 outputs voice guidance requesting the second password utterance again by the procedure described above.

  If A is bad and B and C are good, any two password utterances are strange, but since it cannot be specified, all password utterances are requested again. In response to an instruction from the mutual reliability determination means, the utterance acquisition unit 16 outputs voice guidance requesting password utterances three times again according to the procedure described above.

  In this embodiment, three voiceprint data relating to the same password are registered in the user profile of each user by new voiceprint registration. Even if the same user speaks the same password, the voice print data is not exactly the same. Therefore, in order to absorb fluctuations in voiceprint data and prevent erroneous determination, a plurality of voiceprint data is collected and registered from a plurality of utterances.

  Further, in the voice guidance for prompting the utterance of the password, the voice guidance is performed so that the utterance speed is changed at the first time, the second time, and the third time. For example, “Register your mobile phone number at normal speed” when registering the first voiceprint, “Please tell your mobile phone number as soon as possible” when registering the second voiceprint, and “Slowly when registering the third voiceprint” Please give me your mobile phone number. " By collecting and registering the voiceprint data of passwords spoken according to such voice guidance, voiceprint data when speaking at normal speed, voiceprint data when speaking quickly, and when speaking slowly Voice print data is registered in the user profile.

  In addition, in order to acquire a plurality of voiceprint data having different utterance speeds for the same password, the utterance voice data is input to a computer, and a plurality of voiceprint data is acquired from the same utterance voice by changing the utterance speed on the computer. You may comprise. In such a configuration, the user needs to speak only once for voiceprint registration, so the time required for voiceprint registration can be shortened. Or, as described above, changing the utterance speed, uttering a password multiple times (three times), inputting each utterance voice data into the computer and changing the utterance speed, thereby obtaining a large number of voiceprint data with different utterance speeds This can improve the accuracy of voiceprint authentication.

FIG. 5 is a flowchart for user authentication. Hereinafter, the user authentication operation will be described.
A user who wants to receive an authentication service in this system dials the telephone number of the interactive voice-compatible device 3 from the user call terminal 1 and connects to the line. In the interactive voice response device 3, when there is an incoming call from the user call terminal 1, the caller number acquisition unit 12 acquires the caller number of the user call terminal 1. If the caller ID is set to non notification, the voice guidance is given so that the calling number is set and the call is repeated. The caller number acquired by the caller number acquisition unit 12 is sent to the user registration confirmation unit 21. The user registration confirmation unit 21 inquires of the user management server 7 whether or not a user having the caller number is registered. The registration presence / absence information returned from the user management server 7 to the user registration confirmation unit 21 is notified to the caller number acquisition unit 12. If the user is not registered, the voice guidance support unit 11 performs voice guidance to the effect that the user is not registered, and ends the process.

  If the user is registered, the user information creation unit 13 notifies the user information acquisition unit 22 of the caller number and starts preparation for creating the keyword dictionary, while the voice guidance support unit 11 registers with the user call terminal 1. A voice guidance is output instructing the user to utter a mobile phone number and finally press a specific button (for example, # button). The user speaks the mobile phone number registered in advance in the user management server 7 according to the voice guidance.

  The utterance acquisition unit 16 acquires utterance voice data of a mobile phone number uttered by the user to the user call terminal 1. First, the speech voice data of the mobile phone number is transmitted to the voice recognition server 5 for password authentication to determine whether or not it can be recognized.

  The user information acquisition unit 22 acquires a registered mobile phone number, which is a registration password, from the user management server 7 in conjunction with user registration confirmation at the time of incoming call, and supplies it to the keyword dictionary creation unit 23. The keyword dictionary creation unit 23 converts the registered mobile phone number into pronunciation data that is a way of reading the mobile phone number and holds it as a keyword dictionary in the same way as when registering a new voiceprint.

  The voice recognition server 5 acoustically analyzes the utterance voice data of the mobile phone number supplied from the utterance acquisition unit 16 and collates the analysis result with the keyword dictionary held by the keyword dictionary creation unit 23 to execute voice recognition. If the utterance voice data of the mobile phone number supplied from the utterance acquisition unit 16 and the registered mobile phone number supplied to the keyword dictionary creation unit 23 are the same, the speech recognition succeeds, but if they are different, the voice cannot be recognized. Recognition fails. That is, the mobile phone number specified by the user is registered in the user management server 7 in advance, and if the mobile phone number spoken by the user at the time of user authentication is matched, the password is matched, so the password authentication is successful. become. If they do not match, password authentication has failed. The voice recognition server 5 notifies the utterance acquisition unit 16 of the voice recognition result that is the password authentication result.

  The utterance acquisition unit 16 issues a voice print authentication request to the authentication request unit 18 when the voice recognition result is a voice recognition success. If the voice recognition is unsuccessful, it does not match the registered mobile phone number, so the voice guidance is performed so that the mobile phone number is spoken again and the specific button is pressed. If the voice recognition failure reaches a predetermined number of times, a voice guidance indicating that the authentication has not been performed is output, and the authentication process is terminated.

  The voiceprint authentication unit 27 receives a voiceprint authentication request (including speech voice data of a password-authenticated mobile phone number) from the authentication requesting unit 18. In this example, it is assumed that the caller number acquired by the caller number acquisition unit 12 is added to the request, but the telephone number that is the caller number is acquired from the user information acquisition unit 22. Also good. The voiceprint authentication unit 27 extracts voiceprint data from the utterance voice data of the mobile phone number, and acquires the registered voiceprint data of the user from the user profile database 26 using the phone number as a key. Then, the degree of similarity between the voiceprint data extracted from the utterance voice data this time and the registered voiceprint data acquired from the user profile database 26 is calculated. If the similarity score exceeds a predetermined value, the voiceprint authentication succeeds because the voiceprints match, but if the similarity score does not reach the predetermined value, the voiceprint authentication fails.

  Here, the present embodiment registers a plurality of voiceprint data as described above from the viewpoint of preventing erroneous determination of voiceprint authentication. Guidance for prompting password utterance when registering voiceprints Registering voiceprint data with different utterance speeds by instructing to change the utterance speed, such as "Normal" for the first time, "Fast" for the second time, "Slow" for the third time, etc. It is desirable to do. Depending on the user's situation at the time of password utterance (including psychological state and health state), the utterance speed changes greatly and the vocal cord tension also changes greatly. Therefore, it is mistaken that there is only one registered voiceprint data. It is difficult to prevent judgment completely. If a plurality of voiceprint data having different utterance speeds are registered as a user profile as in this embodiment, even if there is a fluctuation in the voiceprint data of the uttered voice according to the user's situation, it can be flexibly dealt with. And erroneous determination can be effectively prevented. In other words, it is possible to cope with a certain amount of fluctuation of voiceprint data by lowering the threshold of similarity, but this lowers the security level of voiceprint authentication itself. On the other hand, if a plurality of voiceprint data with different utterance speeds are used as in this embodiment, erroneous determination can be effectively suppressed without reducing the reliability of voiceprint authentication.

  The voiceprint also changes due to changes in the user's body shape or changes in vocal cords due to aging. In order to cope with such a voiceprint change, the user profile is controlled to be flexibly updated to the latest voiceprint data.

  FIG. 6 is a diagram schematically showing a registration state of voiceprint data which is a part of the user profile. When registering a new voiceprint, register the voiceprint data obtained from the password utterance voice at the normal utterance speed as the first voiceprint pattern, and register the voiceprint data obtained from the password utterance voice in the early voice as the second voiceprint pattern, Voiceprint data obtained from the password utterance voice at a slow utterance speed is registered as the third voiceprint pattern. Also, the registration date and time of each registered voiceprint data is registered. The voiceprint authentication unit 27 collates with the voiceprint data registered as the first voiceprint pattern, and if the similarity is equal to or less than a predetermined value, then the voiceprint authentication unit 27 performs collation with the voiceprint data registered as the second voiceprint pattern. If it is equal to or less than the predetermined value, it is further collated with voiceprint data registered as the third voiceprint pattern. The user profile reconstructing unit 29 is informed of which voice print data is the first to third voice print patterns whose similarity exceeds a predetermined value.

  The user profile restructuring unit 29 additionally registers the voiceprint data that has been successfully authenticated at the end of the voiceprint pattern transmitted from the voiceprint authentication unit 27 and further writes the registration date and time. The number of registrations for each voiceprint pattern is determined in advance, and when the number exceeds the number of registrations, the oldest registration date is deleted. With this user profile reconstruction, the latest voiceprint data is registered in the user profile for each voiceprint pattern. Therefore, even if the voiceprint data of the password uttered voice changes due to the change in the user's body shape or the aging, the voiceprint authentication is performed based on the latest voiceprint data, so that erroneous determination can be effectively prevented.

  The voice print authentication result by the voice print authentication unit 27 is notified from the authentication result notification unit 28 to the user management server 7 set as a notification destination. Also, the voiceprint authentication unit 27 notifies the voice guidance corresponding unit 11 of the success / failure of the voiceprint authentication, and the voice guidance support unit 11 notifies the user call terminal 1 of the success / failure of the user authentication.

  As described above, according to the present embodiment, since the password utterance voice data input from the user call terminal 1 is used for both password authentication and voiceprint authentication, the user connects to the interactive voice-compatible device 3. The user authentication service can be received by uttering the password once, and the operation on the user side can be simplified. Further, since password authentication and voiceprint authentication can be performed without switching communication lines, it is possible to shorten the time required for user authentication.

  According to the present embodiment, voice print data (first to third voice print patterns) are extracted and registered from keyword utterance voices having different utterance speeds, and the voice print extracted from the password utterance voice data at the time of user authentication. Since the voiceprint authentication succeeds if the data matches the voiceprint data of one of the voiceprint patterns, it is possible to flexibly cope with a change in the situation on the user side and effectively prevent erroneous determination of the voiceprint authentication. it can.

  In addition, according to the present embodiment, the user profile reconstruction unit 29 registers the voice print data that has succeeded in the voice print authentication into the user profile and reconstructs the user profile with the latest voice print data. Alternatively, even if the voiceprint data of the password utterance is changed due to a change in the vocal cords due to aging, since the voiceprint authentication is performed based on the latest voiceprint data, erroneous determination can be effectively prevented.

  Further, according to the present embodiment, since the keyword dictionary is created only with the registered password of the user to be authenticated, it is possible to improve the voice recognition accuracy for the password uttered voice and shorten the recognition time.

  In the above embodiment, as shown in FIG. 4, noise determination is performed before password utterance at the time of new voiceprint registration. However, after acquiring the password utterance voice, from the silent section existing before and after the password utterance voice. Noise determination may be performed.

  As shown in FIG. 7, the password utterance voice data acquired by the utterance acquisition unit 16 includes a predetermined section before the start of the utterance and a predetermined section after the end of the utterance. The voice recognition engine of the voice recognition server 5 cuts out the utterance section and performs acoustic analysis, and the time tag of the silent section (T1 to T2) before the start of the utterance and the time of the silent section (T3 to T4) after the end of the utterance. The tag can be returned to the interactive speech support apparatus 3 together with the speech recognition result. The interactive speech support apparatus 3 inputs the time tag sent from the speech recognition server 5 to the noise determination unit 14. The noise determination unit 14 takes in the password utterance voice data from the utterance acquisition unit 16, cuts out the silent section (T1 to T2) (T3 to T4) with reference to the time tag, and performs noise determination.

  Thus, by performing noise determination using the password utterance voice data, it is possible to reduce time for noise measurement by interrupting utterances only for noise determination, and it is possible to reduce time required for voiceprint registration.

  In the above description, the user call terminal 1 dials the interactive voice response device 3 to connect the line. However, voice data can be transmitted and received between the user call terminal 1 and the interactive voice response device 3. If it is, the communication form will not be limited.

  The present invention can be applied to a voice authentication station that receives a speech voice captured from a user call terminal via a communication network and performs user authentication in real time.

System configuration diagram of an embodiment of the present invention Functional block diagram of the interactive speech support apparatus shown in FIG. Functional block diagram of the voiceprint authentication server shown in FIG. Flow chart of new voiceprint registration in this embodiment Flow chart of voiceprint authentication in this embodiment The figure which shows the registration state of the voiceprint data to the user profile in this Embodiment Speech waveform diagram of utterance voice data including utterance interval and silent interval

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 User call terminal 2 Telephone network 3 Interactive voice corresponding apparatus 4 Voiceprint authentication server 5 Voice recognition server 6 User database 7 User management server 11 Voice guidance corresponding part 12 Caller number acquisition part 13 User information creation part 14 Noise judgment part 15 Speech Voice level determination unit 16 Utterance acquisition unit 17 Profile creation request unit 18 Authentication request unit 21 User registration confirmation unit 22 User information acquisition unit 23 Keyword dictionary creation unit 25 User profile creation unit 26 User profile database 27 Voice print authentication unit 28 Authentication result notification unit 29 User Profile Restructuring Unit

Claims (9)

  Speech acquisition means for acquiring speech utterance voice input to a user call terminal via a communication network, and voice recognition means for determining by speech recognition whether the acquired password utterance voice matches a registered password registered in advance. And voice print authentication means for determining whether or not the voice print data of the password uttered voice matches the registered voice print data in the user profile of the user.   The user profile has a plurality of patterns of voiceprint data extracted from password utterances having different utterance speeds, and the voiceprint authentication means includes the voiceprint data of the password utterances acquired at the time of user authentication in the user profile. 2. The user authentication system according to claim 1, wherein it is determined whether or not it matches any one of a plurality of patterns of voiceprint data.   User information acquisition means for acquiring the registered password of the user from the user management means for managing the registered password based on the identification information of the user call terminal that requested user authentication or the identification information of the user, and the user information acquisition means 3. The user authentication system according to claim 1, further comprising: a keyword dictionary creating unit that converts the registered password into a keyword dictionary used by the voice recognition unit for voice recognition.   4. The apparatus according to claim 1, further comprising: a user profile restructuring unit for additionally registering voice print data of the password uttered voice as voice print data in the user profile when the voice print authentication unit succeeds in the voice print authentication. The user authentication system according to any one of the above.   When newly registering voiceprint data, there is a period during which no utterance is input to the user call terminal, and noise that determines whether or not the user's environment has an acceptable noise level based on a received signal during the period when the utterance is not input Determination means, and user profile creation means for extracting voiceprint data from password utterance voice data for voiceprint registration and registering it in the user profile after the noise determination means determines that the noise level is acceptable. The user authentication system according to any one of claims 1 to 4.   When newly registering voiceprint data, noise that determines whether or not the environment on the user side is at an acceptable noise level from the state of at least one predetermined section before or after the utterance period from the password utterance voice data for voiceprint registration 5. The apparatus according to claim 1, further comprising: a determination unit; and a user profile creation unit that extracts voiceprint data from a speech period of voice utterance voice data for voiceprint registration and registers the extracted voiceprint data in the user profile. User authentication system.   It is provided with a sound breaking judgment means for judging whether or not the input voice exceeds the input allowable level from the utterance period of the password utterance voice data for voiceprint registration, and it is determined that the sound breaking judgment means falls within the input allowable level. 7. The user authentication system according to claim 6, wherein the voice print data is extracted after the extraction.   When registering new voiceprint data, perform password utterances for voiceprint registration at least three times, compare the first password utterance voice with the second password utterance voice, calculate the person's accuracy, and first password utterance Comparing the voice and the third password utterance voice to calculate the personality accuracy, comparing the second password utterance voice and the third password utterance voice to calculate the personality accuracy, and calculating all the personality accuracy The user authentication system according to any one of claims 1 to 4, wherein each password uttered voice is adopted when the password exceeds a predetermined value.   Acquiring a speech utterance of a password input to a user call terminal via a communication network; determining by speech recognition whether the acquired password utterance matches a registered password registered in advance; and the password Determining whether the voiceprint data of the uttered voice matches the registered voiceprint data in the user profile of the user.

Images