JP2021064110A - Voice authentication device, voice authentication system and voice authentication method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice authentication device, a voice authentication system, and a voice authentication method capable of ensuring high security at the time of authentication and reducing the burden on the user at the time of registration. SOLUTION: An authentication character string generation unit 23 for generating a second utterance character string including characters included in a first utterance character string, and a first utterance character string generated by a user speaking. Using the voiceprint data generation unit 25 that extracts the voiceprint data corresponding to the second speech character string from the first voice data, and the second voice data and voiceprint data generated by the user speaking the second speech string. It is provided with an authentication unit 26 that authenticates the user. [Selection diagram] Fig. 3

Description

The present invention relates to a voice authentication device, a voice authentication system, and a voice authentication method for performing voice authentication.

Voice authentication technology that authenticates using voice spoken by a user is widespread. The voice authentication technology is a technology for authenticating by collating a user's voice registered in advance with a user's voice newly acquired at the time of authentication.

Typical verification methods in voice authentication technology include password methods and free word methods. In the password method, the voice data obtained by the user speaking a predetermined password is registered in advance, and the voice data is collated with the voice data obtained by the user speaking the password again at the time of authentication. Authentication is performed by doing. Further, in the free word type voice authentication technology, whether or not the characteristics of the user's voice match is determined by using the voice data registered in advance and the voice data obtained by the user uttering free contents. Authentication is performed by making a judgment.

In such a voice authentication technology, for example, if the voice data at the time of registration is stolen, the authentication may be performed illegally by spoofing. In order to avoid this, it is conceivable that a plurality of voice data obtained by the user uttering a plurality of passwords are registered in advance, and authentication is performed using one or several passwords among the plurality of passwords. ..

However, it takes a lot of time for the user to speak a plurality of passwords and register the voice data, which imposes a heavy burden on the user. Therefore, there is a demand for an authentication technology that reduces the burden on the user at the time of registration and has high security.

In Patent Document 1, instead of registering the voice data obtained by the user speaking the password, the voice print data for each element of each sound element constituting the password is registered, and the voice data obtained at the time of authentication is registered. A technique for performing authentication by comparing voice data and voice print data for each element is disclosed.

Japanese Unexamined Patent Publication No. 2005-128307

In the technique disclosed in Patent Document 1, it is necessary to register voiceprint data in advance for each sound element that can appear in the password set at the time of authentication. Therefore, in the technique disclosed in Patent Document 1, the user is made to speak a different password each time the authentication is performed. For this reason, even if the voice data at the time of registration is stolen, unauthorized authentication can be prevented, but since the voiceprint data is registered for each sound element, it has been improved that a heavy burden is imposed on the user at the time of registration. Absent.

An object of the present invention is to provide a voice authentication device, a voice authentication system, and a voice authentication method that can ensure high security at the time of authentication and reduce the burden on the user at the time of registration. ..

The voice authentication device of the present invention includes an authentication character string generator that generates an authentication character string including at least one of the characters included in the predetermined character string, and a first generated by the user speaking the predetermined character string. From the 1 voice data, the first voice print data of the portion corresponding to the authentication character string is generated, and the second voice print data is based on the second voice data generated by the user speaking the authentication character string. The user is provided with a voiceprint data generation unit that generates the data, and an authentication unit that collates the first voiceprint data with the second voiceprint data to authenticate the user.

The voice authentication system of the present invention includes a voice data generation unit that generates voice data based on a user's speech, a presentation section that presents a character string that requests the user to speak, and a character included in a predetermined character string. An authentication character string generation unit that generates an authentication character string including at least one of them, and a portion corresponding to the authentication character string from the first voice data generated by the user speaking the predetermined character string. A voiceprint data generation unit that generates first voiceprint data and generates second voiceprint data based on the second voice data generated by the user speaking the authentication character string, and the first voiceprint data. It includes an authentication unit that collates with the second voiceprint data and authenticates the user.

The voice authentication method of the present invention generates an authentication character string including at least one of the characters included in the predetermined character string, and the authentication is performed from the first voice data generated by the user speaking the predetermined character string. The first voiceprint data of the part corresponding to the character string is generated, the second voiceprint data is generated based on the second voice data generated by the user speaking the authentication character string, and the first voiceprint is generated. The user is authenticated using the data and the second voice pattern data.

It is possible to ensure high security at the time of authentication and reduce the burden on the user at the time of registration.

Diagram for explaining the configuration of the voice authentication system Diagram for explaining the configuration of the terminal device Diagram for explaining the configuration of the voice authentication device The figure for demonstrating the structure of the voice recognition part Flow chart for explaining the voice registration process for registering the user's voice in the voice authentication system Diagram showing the relationship between the service that the user wants to receive and the security level Flow chart for explaining voice authentication processing that performs voice authentication using pre-registered voice

Hereinafter, each embodiment of the present invention will be described in detail with reference to the drawings. However, more detailed explanations than necessary, such as detailed explanations of already well-known matters and duplicate explanations for substantially the same configuration, may be omitted.

The following description and referenced drawings are provided for those skilled in the art to understand the present invention, and are not intended to limit the scope of the claims of the present invention.

<Configuration of voice authentication system 1>
First, the configuration of the voice authentication system 1 according to the embodiment of the present invention will be described with reference to FIG. As shown in FIG. 1, the voice authentication system 1 includes a terminal device 10, a voice authentication device 20, and a network 30.

The terminal device 10 is a device that receives voice input from a user of the voice authentication system 1 (hereinafter, simply referred to as a user) and generates voice data. An example of the terminal device 10 is a mobile terminal device such as a tablet terminal or a smartphone. Further, as an example of the terminal device 10, a computer such as a PC (Personal Computer) installed at a specific place can be mentioned.

In the present embodiment, when the user wants to receive a desired service, the terminal device 10 registers the place where the authentication for receiving the service is performed and the voice data necessary for the authentication. It is installed in a place where it is called. The user registers the voice data necessary for voice authentication through the terminal device 10 before receiving the desired service. In addition, the user can receive the desired service when the voice authentication using the registered voice data and the voice data newly generated at the time of authentication is successful by the voice authentication device 20.

In the present embodiment, voice authentication means that authentication is performed using voice data. Further, in the following description, registering voice data used for voice authentication in advance is referred to as voice registration. In voice authentication, the success or failure of authentication is determined by collating the voice data registered by voice with the voice data newly generated at the time of authentication. The details of the voice registration process for performing voice registration and the voice authentication process for performing voice authentication in the voice authentication system 1 will be described later.

The desired service is not particularly limited in the present invention. Examples of desired services include admission to a specific facility (membership store, etc.), payment of fees, remittance to other users, and the like.

For example, when a user wishes to enter a membership-based sports club, the user performs voice registration using a terminal device 10 installed at the entrance of the sports club. After that, the user causes the voice authentication system 1 to perform voice authentication via the terminal device 10. If the authentication is successful, the user is allowed to enter the sports club.

Further, for example, when the user wishes to settle the fee at the time of purchasing the product, the user performs voice registration using the terminal device 10 installed at the cash register of the store where the product is to be purchased. After that, the user causes the voice authentication system 1 to perform voice authentication via the terminal device 10. If the authentication is successful, the fee will be settled using the user's settlement account registered in advance.

Returning to the description of FIG. The voice authentication device 20 is a device that is connected to the terminal device 10 via a network 30, receives voice data from the terminal device 10, and performs voice registration and voice authentication using the voice data. An example of the voice authentication device 20 is a cloud server.

The network 30 is a public network such as the Internet.

<Configuration of terminal device 10>
Next, the configuration of the terminal device 10 will be described with reference to FIG. As shown in FIG. 2, the terminal device 10 includes a communication unit 11, a voice data generation unit 12, a presentation unit 13, and an input unit 14.

The communication unit 11 transmits and receives various data to and from the voice authentication device 20 via the network 30.

The voice data generation unit 12 receives the voice input of the user who has spoken to the terminal device 10 and generates voice data based on the voice input. The voice data generation unit 12 is, for example, a device having a built-in microphone. The method of generating audio data by the audio data generation unit 12 is not particularly limited in the present invention, but for example, a method of AD-converting an audio signal input from a microphone to generate audio data can be adopted. The voice data generation unit 12 may further convert the voice data into a compression code.

In the following description, the data obtained by A-D conversion of the audio signal, the encoded data of the A-D converted data, and the data obtained through various processes (noise reduction, feature amount extraction, and phonetic conversion) described later are collectively referred to. And it is called voice data. In addition, among the voice data, data that can identify an individual is called voiceprint data. It should be noted that voiceprint data is generally used for voice authentication, and in the following description, voice authentication and voiceprint authentication are substantially synonymous.

The presentation unit 13 gives the user a character string (a first utterance character string described later or a second utterance character string described later) to be spoken by the user when the terminal device 10 performs voice registration or voice authentication of the user. I will present it to you. The presentation unit 13 is, for example, a display, and in this case, the presentation unit 13 presents the character string to the user by displaying the character string. Further, the presentation unit 13 is, for example, a speaker, and in this case, the presentation unit 13 presents a character string to be spoken by the user by voice. In the present invention, the presentation unit 13 is not limited to the display or the speaker, and may be a device that presents a character string in, for example, Braille. Further, the presentation unit 13 may be a combination of a display and a speaker, or another device.

In the present invention, the character string means a series of characters. In the present invention, the character string may be a series of characters that do not make sense, but it is more preferable that the character string is a series of characters (words, etc.) that make sense in the entire character string. In the present invention, the character means a symbol or the like used as one of the means of transmitting a language, and the symbol is not particularly limited, but one symbol represents one sound. (So-called syllabary) is more desirable. In the following, a case where a kana or an Arabic numeral is adopted as a character will be described.

The presentation unit 13 may present only one character string, or may present a plurality of character strings. When the presenting unit 13 presents a plurality of character strings, the user may be able to select which of the character strings to speak. In addition, the presentation unit 13 may present a plurality of character strings in an order according to the security level described later, or may be a random order.

The input unit 14 accepts input operations other than voice by the user. The input unit 14 is, for example, an operation device such as a touch panel, a keyboard, a mouse, and a trackball.

With such a configuration, the terminal device 10 presents a character string to be spoken by the user at the time of voice authentication of the user, and when the user speaks the character string, the voice is acquired as voice data.

The character string presented by the presentation unit 13 is generated by the first utterance character string generation unit 22 or the authentication character string generation unit 23 of the voice authentication device 20, which will be described later. The details of the operation of the first utterance character string generation unit 22 or the authentication character string generation unit 23 will be described later.

<Configuration of voice authentication device 20>
Next, the configuration of the voice authentication device 20 will be described with reference to FIG. As shown in FIG. 3, the voice authentication device 20 includes a communication unit 21, a first utterance character string generation unit 22, an authentication character string generation unit 23, a voice recognition unit 24, and a voiceprint data generation unit 25. , An authentication unit 26, a service provision unit 27, a user data acquisition unit 28, and a storage unit 29.

The communication unit 21 transmits and receives various data to and from the terminal device 10 via the network 30.

The first utterance character string generation unit 22 generates the first utterance character string, which is a character string to be spoken by the user during the voice registration process of the voice authentication system 1. The method of generating the first utterance character string by the first utterance character string generation unit 22 is not particularly limited, but it is desirable that a character string that is easy for the user to speak is generated.

The first utterance character string generation unit 22 may generate a character string by referring to, for example, a corpus or user-specific dictionary data stored in a storage unit 29, which will be described later. Alternatively, the first utterance character string generation unit 22 includes at least a part of personal information about the user obtained by the user data acquisition unit 28, which will be described later, or a word associated with the personal information. You may generate columns.

The authentication character string generation unit 23 generates a second utterance character string, which is a character string to be spoken by the user during the voice authentication process of the voice authentication system 1. The authentication character string generation unit 23 first generates an authentication character string including any of the characters included in the first speech character string generated by the first speech character string generation unit 22, and then generates an authentication character string. A character string for the second speech is generated by appropriately combining the columns.

Each character string generated by the first utterance character string generation unit 22 and the authentication character string generation unit 23 will be described in detail later with specific examples.

The voice recognition unit 24 converts the voice data received from the terminal device 10 by the communication unit 21 into a character string. At the same time, the voice recognition unit 24 cuts out voice data for each predetermined character string and outputs the voice data in association with the character string. Then, the voice recognition unit 24 associates the cut-out voice data, the character string, and the time information (start time and end time) of the character string, and stores the associated information in the storage unit 29.

The voice recognition unit 24 may cut out the voice data for each character included in the character string converted based on the voice data, but for example, the voice data is output for each word or phrase included in the converted character string. It is more desirable to cut out. The character string converted by the voice recognition unit 24 and the voice data to be cut out will be described in detail later with specific examples.

The voice recognition unit 24 uses the voice data obtained by the user actually speaking the first utterance character string during the voice registration process of the voice authentication system 1 to perform the above operation (conversion to a character string). , Cut out the voice data corresponding to the character string, and associate it with the time information of the character string). In the following description, the voice data obtained by the user actually speaking the first utterance character string is referred to as the first voice data.

In addition, the voice recognition unit 24 performs the above operation when the voice authentication process of the voice authentication system 1 is performed, using the voice data obtained by the user actually speaking the second utterance character string. In the following description, the voice data obtained by the user actually speaking the second utterance character string will be referred to as the second voice data. Further, the voice recognition unit 24 removes a part corresponding to an additional character string described later from the user's voice data at the time of voice recognition processing, and only the voice data and time information of the other parts are the result of voice recognition. Output as.

The voice recognition unit 24 will be described with reference to FIG. As shown in FIG. 4, the voice recognition unit 24 includes a noise reduction unit 241, an utterance detection unit 242, a feature amount extraction unit 243, a phoneme conversion unit 244, a character string conversion unit 245, an acoustic model 246, and the like. It has a language model 247 and an additional character string removing unit 248.

The noise reduction unit 241 reduces noise (noise in the user's surrounding environment, etc.) included in the voice data. The noise reduction method by the noise reduction unit 241 is not particularly limited in the present invention, and a known noise reduction method can be adopted.

The utterance detection unit 242 cuts out the portion spoken by the user from the input voice data.

The feature amount extraction unit 243 extracts, for example, MFCC (Mel-Frequency Cepstrum Coefficient) or the like as a feature amount of voice data. A known technique may be adopted for the feature amount extraction method, and the feature amount extraction method is not particularly limited in the present invention.

The phoneme conversion unit 244 converts the extracted features into phonemes. The phoneme means the smallest unit of speech in a certain language, and in the present embodiment, it indicates one sound corresponding to one character. To give a specific example, the phoneme is the voice data (voice waveform) of the part corresponding to "a" extracted from the voice data obtained by the user actually speaking the character string "aiueo". is there. One voice data is converted into a set of a plurality of phonemes by the phoneme conversion unit 244.

The method by which the phoneme conversion unit 244 converts the feature amount into a phoneme is not particularly limited in the present invention, but the following method can be adopted. For example, the phoneme conversion unit 244 converts the feature amount into a phoneme by a statistical method using the acoustic model 246. The acoustic model 246 models the probability distribution regarding the connection between the feature amount and the phoneme in advance, and the phoneme conversion unit 244 compares (matches) a certain feature amount with the acoustic model, and selects the phoneme with the highest accuracy. , Output as a phoneme corresponding to the feature amount. For example, the phoneme conversion unit 244 may add the most useful features one by one so as to increase the accuracy, and repeat the process until there is no change in the accuracy.

The character string conversion unit 245 converts phonemes into character strings. That is, the character string conversion unit 245 outputs the content of the input voice data as a character string.

The method by which the character string conversion unit 245 converts a phoneme into a character string is not particularly limited in the present invention, but the following method can be adopted. For example, the character string conversion unit 245 converts phonemes into character strings by a statistical method using the language model 247. The language model 247 is a model of a probability distribution related to a grammatical structure such as a connection of part of speech (word), and is a model including dictionary data. The language model 247 may be dynamically updated according to the result of voice recognition or voice authentication.

The character string conversion unit 245 extracts, for example, the sequence of phonemes output from the phoneme conversion unit 244, the corresponding word, part of speech, or a connection thereof from the language model 247, and corresponds to the most accurate character string to the phoneme. Output as a character string to be used. By such a method, voice data that makes sense as a sentence can be accurately converted into a character string. Further, the character string conversion unit 245 cuts out voice data for each converted character string (word), acquires time information (start time and end time) of the character string, and converts the converted character string and the cut out voice data. And the time information are associated and output.

The additional character string removing unit 248 refers to the storage unit 29, detects a character string corresponding to the authentication character string from the character strings output by the character string conversion unit 245, and character strings other than the detected authentication character string ( (Equivalent to an additional character string) is removed. Then, the additional character string removing unit 248 outputs only the authentication character string, the voice data associated with the authentication character string, and the time information.

Returning to the description of FIG. The voiceprint data generation unit 25 generates voiceprint data from the voice data associated with each character string output by the voice recognition unit 24. When a plurality of character strings are output from the voice recognition unit 24, the voiceprint data generation unit 25 generates each voiceprint data from the voice data associated with each character string.

The voiceprint data is data related to the voiceprint, which is a characteristic unique to the user's voice. The method for generating the voiceprint data is not particularly limited in the present invention, and a known method may be appropriately adopted. As a general voiceprint data generation method, there is a method of generating voiceprint data by taking a difference from a standard model of phonology created from the voices of a large number of speakers.

The voice pattern data generation unit 25 generates voice pattern data from the voice data associated with the character string output by the voice recognition unit 24 based on the first voice data during the voice registration process of the voice authentication system 1. .. In the following description, the voiceprint data generated by the voiceprint data generation unit 25 during the voice registration process will be referred to as the first voiceprint data. The first voiceprint data generated by the voiceprint data generation unit 25 is associated with a character string output by the voice recognition unit 24 based on the first voice data during the voice registration process, and is stored in the storage unit 29 described later. Will be done.

On the other hand, the voice pattern data generation unit 25 obtains voice pattern data from the voice data associated with the character string output by the voice recognition unit 24 based on the second voice data during the voice authentication process of the voice authentication system 1. Generate. In the following description, the voiceprint data generated by the voiceprint data generation unit 25 during the voice authentication process will be referred to as the second voiceprint data.

The authentication unit 26 collates the first voiceprint data stored in the storage unit 29 with the second voiceprint data generated by the voiceprint data generation unit 25 during the voice authentication process of the voice authentication system 1. , Authenticate the user. The authentication unit 26 determines that the authentication is successful when the matching rate between the first voiceprint data and the second voiceprint data is equal to or more than a predetermined threshold value, and determines that the authentication is unsuccessful when the matching rate is less than the predetermined threshold value.

The service providing unit 27 provides a desired service to the user when the authentication unit 26 succeeds in authenticating the user. In the present embodiment, for convenience of explanation, the voice authentication device 20 has the service providing unit 27, but the present invention is not limited to this.

In the present invention, instead of the voice authentication device 20 having the service providing unit 27, the voice authentication system 1 may further have a service providing device that provides a desired service to the user (that is, the voice authentication device). A service providing device independent of 20 may be provided). In this case, the service providing device may provide the service when the authentication by the authentication unit 26 of the voice authentication device 20 is successful.

The user data acquisition unit 28 acquires data (user data) related to the user's personal information (name, address, date of birth, gender, age, etc.) input by the user via the terminal device 10. The user data acquisition unit 28 may acquire user data input by the user via the input unit 14 of the terminal device 10, for example, via the network 30 and the communication unit 21.

The storage unit 29 is a storage device that stores various data used in the voice authentication device 20. The storage unit 29 stores, for example, at least one of the corpus and dictionary data for generating the first utterance character string. Further, the storage unit 29 stores the user's voice data, the character string, and the information in which the time information of the character string is associated with each other. Further, in the storage unit 29, the user's identification information, the first speech character string, the authentication character string, the second speech character string, the first voice data, the second voice data, and the first voice print data are stored for each user. , And the second voice pattern data are stored in association with each other.

<Operation example of voice authentication system 1>
Hereinafter, an operation example of the voice authentication system 1 having the above-described configuration will be described with reference to FIGS. 5 and 6.

[1] Voice registration process First, a voice registration process for registering a user's voice in the voice authentication system 1 will be described with reference to FIG.

In step S1, the voice authentication device 20 receives the registration processing start request from the user input via the terminal device 10. The user requests to start the voice registration process, for example, via the input unit 14 of the terminal device 10. As a result, the voice registration process of the voice authentication system 1 is started.

In step S2, the first utterance character string generation unit 22 of the voice authentication device 20 generates a first utterance character string which is a character string for the user to speak at the time of voice registration of the user.

As described above, the first utterance character string generated by the first utterance character string generation unit 22 is preferably a character string that is easy for the user to speak, but may be a random character string. For example, it may be a character string including at least a part of personal information such as a user's date of birth, age, address, name, or a word associated with the personal information.

An example of a random character string is to randomly obtain a phrase (for example, "Genkotsuyama no Tanuki-san") that is often used in a specific country or region from the corpus. The character string acquired from the corpus is stored in the storage unit 29 as dictionary data according to the user's attributes derived from the user data, and the first utterance character string generation unit 22 refers to the dictionary data to the second. A character string for one utterance may be randomly generated. The dictionary data may be updated at any time according to the result of voice recognition or voice authentication.

Examples of character strings including personal information such as the user's date of birth, age, and address include "1995 Nen 1 Gatsu 1 Nichi", "33 Sai", and "Chiyoda Kusoto Kanda". In this example, it is written in hiragana, but katakana may be used.

Further, the first utterance character string generation unit 22 may generate the first utterance character string by combining a random character string acquired from the corpus or dictionary data and a character string based on the user data (1st utterance character string generation unit 22). For example, "Genkotsu 33 Sai no Tanuki" etc.).

It is more desirable that the first utterance character string generation unit 22 generate a plurality of different first utterance character strings by such an operation.

Returning to the description of FIG. In step S3, the presentation unit 13 of the terminal device 10 presents the first utterance character string generated by the first utterance character string generation unit 22 in step S2 to the user. At the same time, the presentation unit 13 requests the user to speak (read) the first utterance character string.

For example, when the presentation unit 13 is a display, the presentation unit 13 displays a message such as "Please read the displayed characters aloud" together with the first utterance character string. When the presentation unit 13 is a speaker, the presentation unit 13 reproduces the first utterance character string as voice after issuing a message such as "Please speak the following words." When a plurality of first utterance character strings are generated, the presentation unit 13 presents a plurality of first utterance character strings and selects one of the first utterance character strings to speak. Outputs the message requested to.

As a result, when the user speaks the first utterance character string, the voice data generation unit 12 of the terminal device 10 generates the first voice data corresponding to the first utterance character string.

In step S4, the voice authentication device 20 acquires the first voice data from the terminal device 10.

In step S5, the voice recognition unit 24 of the voice authentication device 20 converts the first voice data into a character string, cuts out the voice data for each word (or character, or phrase) included in the converted character string, and sets the character. The column and the cut out audio data are associated and output.

Hereinafter, a specific example will be described. When the first voice data is the voice data obtained by the user speaking "Genkotsuyama no Tanuki-san", the voice recognition unit 24 corresponds to the voice data corresponding to "Genkotsu", "Yama". Cut out voice data, voice data corresponding to "no", voice data corresponding to "tanuki", and voice data corresponding to "san".

In this example, the voice recognition unit 24 has described the case where the first utterance character string is divided into words to cut out the voice data, but how much the voice recognition unit 24 divides the voice data (word by character, word). In the present invention, there is no particular limitation as to whether or not the data is cut out separately for each clause. For example, the voice recognition unit 24 may divide the first utterance character string into each character (“ge”, “n”, “ko”, “tsu” ...) And cut out the voice data.

Returning to the description of FIG. In step S6, the authentication character string generation unit 23 of the voice authentication device 20 generates an authentication character string, which is a character string used in the voice authentication process described later.

The authentication character string generated by the authentication character string generation unit 23 includes characters (words, phrases) included in the first utterance character string generated by the first utterance character string generation unit 22.

For example, when the character string for the first utterance is "Genkotsuyamanotanuki-san", the following is an example of the character string for authentication.
(1) Genkotsu (2) Tanukiyama (3) Tanukinoko (4) Koyama-san (5) Geyata

The character string (1) is an example in which an authentication character string is generated from a part of the first utterance character string without changing the order of characters or the meaning of a word from the first utterance character string. The character strings (2) and (3) are the first, as long as the order of the characters is partially changed from the first speech character string, but the meaning of each word included in the first speech string is not deviated. This is an example in which an authentication character string is generated from a part of a one-speech character string.

Further, the character string (4) is an example in which an authentication character string is generated from a part of the first utterance character string so that a meaning different from that of the first utterance character string is generated. The character string (5) randomly extracts characters from a part of the first utterance character string to generate an authentication character string, regardless of the order of the first utterance character string and the meaning of the word. This is an example.

In this way, the authentication character string generation unit 23 generates an authentication character string including the characters included in the first utterance character string. In the above example, the authentication character string generation unit 23 generates a plurality of authentication character strings, but the present invention is not limited to this, and only one authentication character string may be generated. ..

In step S6, the authentication character string generation unit 23 was converted by the voice recognition unit 24 based on the first voice data in step S5 as the first speech character string to be used for generating the authentication character string. It is more desirable to use a character string. However, the present invention is not limited to this, and the authentication character string generation unit 23 may directly acquire the first utterance character string from the first utterance character string generation unit 22.

In step S7, the authentication character string generation unit 23 generates a second utterance character string using the authentication character string generated in step S6. The second utterance character string is a character string for causing the user to speak in the voice authentication process described later.

When a plurality of authentication character strings are generated in step S6 described above, in this step S7, the authentication character string generation unit 23 generates a second utterance character string using at least one of them. At this time, the authentication character string generation unit 23 may generate a second utterance character string using a plurality of authentication character strings. Further, the authentication character string generation unit 23 may generate a second utterance character string by adding an additional character string which is a character string other than the authentication character string to one or more authentication character strings. Good.

The additional character string is an extra character string added to the authentication character string for the purpose of improving the security level of the voice authentication process. The method of generating the additional character string is not particularly limited, but for example, it may be generated by acquiring a character string that is easy to use together with the authentication character string from the corpus or dictionary data, and it is completely random and does not make any sense at all. The character string may be an additional character string.

A specific example of the second utterance character string when the authentication character string is the above character strings (1) to (5) will be described below.

As an example, the authentication character string generation unit 23 simply uses one of the above character strings (1) to (5) (for example, the character string (2) "Tanukiyama") as it is for the second utterance character string. May be.

Alternatively, as an example, the authentication character string generation unit 23 may generate a second utterance character string by combining some of the above character strings (1) to (5).

When the character string (1) "Genkotsu" and the character string (3) "Tanuki no Ko" are combined, the second utterance character string becomes "Genkotsu Tanuki no Ko". Further, the authentication character string generation unit 23 may use each of the two character strings (1) "Genkotsu" and the character string (3) "Tanuki no Ko" as the second utterance character string.

Alternatively, if the character string (2) "tanukiyama", the character string (4) "koyama-san", and the character string (5) "geyatasa" are combined, the second utterance character string is "geyata". It becomes "Satanuki Yamakoyama-san". Further, the authentication character string generation unit 23 generates three character strings, that is, the character string (2) "Tanukiyama", the character string (4) "Koyama-san", and the character string (5) "Geyatasa". It may be a character string for two utterances.

Further, as one example, the authentication character string generation unit 23 adds an additional character string to any one of the above character strings (1) to (5) or a combination thereof, and is used for the second utterance. You may generate a string. For example, when the additional character string "Nakayoshi" is added to the character string (3) "Tanuki no Ko", the second utterance character string becomes "Nakayoshi Nuki no Ko".

In this way, when there are a plurality of authentication character strings, the authentication character string generation unit 23 may use at least one of them as the second speech string, or may combine a plurality of not all of them to make the second speech. It may be a character string for the second speech, or it may be a character string for the second speech by combining all of them. Further, the authentication character string generation unit 23 adds an additional character string, which is a character string unrelated to or has an affinity for the authentication character string, to the authentication character string to obtain a second speech character string. It may be generated.

Here, the authentication character string generation unit 23 adds the number of authentication character strings and the additional character strings for generating the second utterance character string according to the preset security level (authentication level). Determine the presence or absence. The security level may be determined in advance for each of a plurality of factors such as the content of the service to be received by the user, the place to receive the service, the frequency of use, the environmental sound level, and the like.

Hereinafter, the security level will be described with reference to FIG. 6 by giving a specific example. For example, the relationship between the content of the service that the user intends to use and the security level is as follows, for example. That is, as shown in FIG. 6, the entrance / exit to a certain facility has a security level of “1”. Payments of 9,999 yen or less have a security level of "2". Payments of 10,000 yen or more have a security level of "3". Remittances to others have a security level of "4".

In this embodiment, the larger the security level number, the higher the security is required. In the above example of the service content, remittance to another person is performed from entering or leaving a certain facility. It has been shown that the required safety is higher.

The relationship between the place of use and the security level is as follows, for example. That is, as shown in FIG. 6, when the usage place is the user's home, the security level is "1". If the place of use is a financial institution, the security level is "2". If the place of use is a public institution (government office, government office, etc.), the security level is "3". When the place of use is a convenience store, it is assumed that there are many people around, so the required safety is higher and the security level is "4".

The relationship between the frequency of use and the security level is as follows, for example. That is, as shown in FIG. 6, if it is within one day from the previous use of the service, the security level is "1". If it is within one week from the previous use of the service, the security level is "2". If it is within one month from the previous use of the service, the security level is "3". When the service is used for the first time, the security level is "4".

Further, although not shown in FIG. 6, the security level may be appropriately set based on the environmental sound level. The environmental sound level is a value indicating the loudness of the environmental sound that the user receives in the environment in which the user intends to use the service. The security level may be set to be higher as the environmental sound level becomes smaller (= the surroundings become quieter), and the security level becomes higher as the environmental sound becomes louder (= the surroundings become noisy). May be set as.

In addition to the above examples, for example, a security level may be set in advance for each user. Specifically, for example, the security level of user A may be set to "1" and the security level of user B may be set to "3". The security level for each user may be set in advance according to, for example, the credit rating calculated based on the user's service usage history so far.

Further, the security level set for each of the plurality of elements described above may be combined to determine the final security level. Although FIG. 6 shows an example in which four levels of security are set, security levels of five or more levels or three or less levels may be set. In addition, different levels of security may be set for each element.

A specific example in which the security level is set by a combination of a plurality of elements will be described. For example, even if the place of use is a convenience store where many people are expected to be noisy (security level "4" in the example of Fig. 6), the service content is relatively small, such as payment of 5,000 yen. In the case of payment (security level "2" in the example of FIG. 6), the final security level may be set to, for example, "3".

The authentication character string generation unit 23 generates a second utterance character string by combining more authentication character strings and additional character strings as the security level set in this way is higher. To give a specific example, for example, when the security level is four levels as shown in FIG. 6, the authentication character string generation unit 23 has the above-mentioned authentication character strings (1) to (1) at the security level "4". A second speech character string is generated by combining all of 5) with additional character strings other than the authentication character strings (1) to (5). On the other hand, at the security level "1", the authentication character string generation unit 23 creates a second utterance character string by combining, for example, only two of the above-mentioned authentication character strings (1) to (5). Just generate it.

When the security level is relatively high, it is more desirable that the authentication character string generation unit 23 generate the second utterance character string using the authentication character string based on the user data.

By the process described above, the higher the required security level, the more complicated the second utterance character string based on the more authentication character strings and the additional character strings will be generated.

Returning to the description of FIG. In step S8, the voiceprint data generation unit 25 generates the first voiceprint data corresponding to the authentication character string based on the voice data cut out by the voice recognition unit 24.

Hereinafter, the generation of the first voiceprint data by the voiceprint data generation unit 25 will be described with reference to specific examples. In the following explanation, the first utterance character string is "Genkotsuyama no Tanuki-san", and the first utterance character string is divided for each character, and the authentication character string is (1) "Genkotsu", (2). It is assumed that "Tanukiyama", (3) "Tanuki no Ko", (4) "Koyama-san", and (5) "Geyatasa" are generated.

First, the voice pattern data generation unit 25 is cut out by the character string conversion unit 245 of the voice recognition unit 24 based on the first voice data obtained by the user speaking the first utterance character string "Genkotsuyama no Tanuki-san". Of the voice data, only the part corresponding to the authentication character string is acquired.

A specific example will be described. The voiceprint data generation unit 25 generates the first voiceprint data for each of the authentication character strings (1) to (5) generated by the voice recognition unit 24 based on the voice data generated for each character from the first utterance character string. To generate.

In step S9, the voice authentication device 20 provides the user identification information, the first utterance character string, the first voice data, a plurality of authentication character strings, the second utterance character string, and the authentication character string, respectively. The corresponding first voice pattern data is stored in the storage unit 29 as a speaker model associated with each other. As a result, the speaker model for each user is stored in the storage unit 29, and the voice registration process for the user to use the voice authentication system 1 is completed.

In the voice registration process shown in FIG. 5, the authentication character string generation unit 23 generates the authentication character string in step S6 after the step S5 in which the voice recognition unit 24 converts the first voice data into the character string. If there is, it may be performed at any timing. That is, step S6 may be performed immediately after, for example, the voice recognition unit recognizes the character string, or after that (for example, after the voice-registered user logs in to the voice authentication system 1 for voice authentication). It may be.

Further, in the voice registration process shown in FIG. 5, the step S8 in which the voiceprint data generation unit 25 generates the first voiceprint data is after the step S5 in which the voice recognition unit 24 cuts out the voice data corresponding to the authentication character string. For example, it may be performed at any timing. That is, step S8 may be performed immediately after the voice data associated with the recognized character string is output from, for example, the voice recognition unit 24, or after that (for example, voice registration of another user). It may be executed in the background where the processing or voice recognition processing is performed).

[2] Voice Authentication Process Next, a voice authentication process when a user intends to use a service will be described with reference to FIG. 7. The voice authentication process is performed on the user who has completed the above-mentioned voice registration process.

In step S11, the voice authentication device 20 receives the service start request from the user input via the terminal device 10. The service start request by the user is made, for example, via the input unit 14 of the terminal device 10. As a result, the user's voice authentication process is started.

In step S12, the presentation unit 13 of the terminal device 10 presents the second utterance character string generated in the voice registration process described above to the user. At the same time, the presentation unit 13 requests the user to speak (read) the second utterance character string. As for the method of presenting the second utterance character string by the presentation unit 13, for example, the same method as the presentation of the first utterance character string in step S3 of FIG. 5 may be adopted.

When the user speaks the second utterance character string in response to the request in step S12, the voice data generation unit 12 of the terminal device 10 generates the second voice data corresponding to the second utterance character string.

In step S13, the voice authentication device 20 acquires the second voice data from the terminal device 10.

In step S14, the voice recognition unit 24 of the voice authentication device 20 converts the second voice data into a character string, cuts out the voice data for each word (or character, or phrase) included in the converted character string, and sets the character. The column and the cut out audio data are associated and output. Further, when the second voice data includes a part corresponding to the additional character string, the voice recognition unit 24 discards the voice data corresponding to the additional character string.

Hereinafter, a specific example will be described. When the second voice data is the voice data obtained by the user speaking "Nakayoshi Nuki no Ko", the voice recognition unit 24 first starts with the second voice data and corresponds to the voice data corresponding to "Nakayoshi". , The voice data corresponding to "Tanuki", the voice data corresponding to "no", and the voice data corresponding to "ko" are cut out respectively.

Next, the voice recognition unit 24 refers to the speaker model stored in the storage unit 29, extracts only the voice data corresponding to the authentication character string from the cut out voice data, and extracts the authentication character string. Remove audio data that does not correspond to. The voice data that does not correspond to the authentication character string is the voice data that corresponds to the additional character string. In the above example, the voice recognition unit 24 discards the voice data corresponding to "Nakayoshi" and outputs the voice data corresponding to "raccoon dog", "no", and "ko".

Returning to the description of FIG. In step S15, the voice pattern data generation unit 25 generates the second voice pattern data by using the voice data output by the voice recognition unit 24 based on the second voice data in step S14. In the case of the above example, the voiceprint data generation unit 25 generates the second voiceprint data corresponding to "raccoon dog", "no", and "ko".

In step S16, the authentication unit 26 performs voice authentication of the user by using the first voiceprint data generated in the above-mentioned voice registration process and the second voiceprint data generated in step S15.

Hereinafter, a specific example will be described. In the following example, as in the example shown in the voice registration process described above, the authentication character string is the character string (1) "Genkotsu", (2) "Tanukiyama", (3) "Tanuki no Ko", ( 4) "Koyama-san" and (5) "Geyatasa". Then, it is assumed that the voiceprint data generation unit 25 generates the second voiceprint data corresponding to "raccoon dog", "no", and "ko".

In this case, the storage unit 29 stores the first voiceprint data corresponding to the authentication character string (3) “raccoon dog”. The authentication unit 26 reads the first voiceprint data from the storage unit 29, and acquires the second voiceprint data corresponding to "tanuki", "no", and "ko" from the voiceprint data generation unit 25. Then, the authentication unit 26 collates the first voiceprint data and the second voiceprint data for each of the character strings "tanuki", "no", and "ko", and calculates the matching rate. The authentication unit 26 determines that the authentication is successful when the matching rate is equal to or higher than the predetermined threshold value, and determines that the authentication is unsuccessful when the matching rate is lower than the predetermined threshold value.

Returning to the description of FIG. In step S17, if the authentication result in step S16 is successful (step S17: success), the process proceeds to step S18, and if the authentication result is unsuccessful (step S17: failure), the process proceeds to step S19. move on.

In step S18, the service providing unit 27 starts providing the service requested by the user to the user who has succeeded in voice authentication.

On the other hand, in step S19, the service providing unit 27 causes the user who has failed in voice authentication to be notified via the presentation unit 13 of the terminal device 10 or the like that the authentication has failed.

<Action, effect>
As described above, the voice authentication device 20 according to the embodiment of the present invention is an authentication character string generation unit 23 that generates an authentication character string including at least one of the characters included in the first speech character string. And, from the first voice data generated by the user speaking the first speech character string, the voice print data of the part corresponding to the authentication character string is generated, and the second speech character including the authentication character string is generated. The voiceprint data generation unit 25, which generates the second voiceprint data based on the second voice data generated by the user speaking the column, collates the first voiceprint data with the second voiceprint data to authenticate the user. It includes an authentication unit 26.

With such a configuration, a first utterance character string that causes the user to speak during the voice registration process for registering the user's voice in the system and a second utterance character string that causes the user to speak during the voice authentication process for voice authentication of the user. Can be different character strings from each other. As a result, for example, even if the voice data in which the user utters the first utterance character string is stolen, the stolen data during the voice authentication process does not result in successful authentication. Therefore, the security of voice authentication is improved.

In addition, the authentication character string is generated using the characters included in the first utterance character string, and the voice pattern is based on the first voice data obtained by having the user actually speak the first utterance character string. Generate data. As a result, it is not necessary to utter the authentication character string itself when the user is registered. Therefore, even when a plurality of authentication character strings are prepared, the burden on the user during the voice registration process is reduced. Therefore, it is possible to substantially generate a plurality of authentication character strings by uttering a phrase having a length that does not burden the user only once. Therefore, the time and effort required for registration can be reduced, and the convenience of the user is improved.

That is, according to the voice authentication device 20 according to the embodiment of the present invention, it is possible to both improve the security of voice authentication and reduce the time and effort of the user at the time of user registration.

Further, in the voice authentication device 20 according to the embodiment of the present invention, the authentication character string generation unit 23 generates a plurality of authentication character strings including the characters included in the first utterance character string, and a plurality of authentications are performed. Generates a second utterance string containing at least one of the strings. At this time, the authentication character string generation unit 23 generates a second utterance character string including a number of authentication character strings corresponding to a preset security level (authentication level).

With such a configuration, it is possible to flexibly generate the second utterance character string according to the required high security of voice authentication. That is, when the required security level is high, high security can be ensured by combining a large number of authentication character strings to generate a complicated second utterance character string. Further, when a very high security level is not required, a relatively simple second utterance character string can be generated by combining a small number of authentication character strings. As a result, when the user actually speaks the second utterance character string during the voice authentication process, a complicated second utterance character string can be spoken even though the required security level is low, so that the second utterance character string can be spoken at the time of authentication. A situation in which a heavy burden is placed on the user is avoided. Therefore, it is possible to perform voice authentication processing in which the required security level and the burden on the user are well-balanced.

Further, according to the voice authentication device 20 according to the embodiment of the present invention, the authentication character string generation unit 23 adds an additional character string different from the authentication character string to the authentication character string for the second speech. Generate a string. Then, the authentication character string generation unit 23 changes the number of authentication character strings constituting the second utterance character string based on a preset security level, and whether or not to add an additional character string. To determine.

In this way, since the user is made to speak by adding a character string that is not included in the generated authentication character string, the variation of the utterance can be increased, and the unauthorized use can be further reduced.

Further, according to the voice authentication device 20 according to the embodiment of the present invention, the first utterance character string generation unit 22 for generating the first utterance character string based on the personal information of the user is further provided.

With such a configuration, the first utterance character string to be spoken by the user during the voice registration process can be a character string including the user's personal information. Since the second utterance character string uttered by the user during the voice authentication process is generated based on the first utterance character string, the user is caused by including the character string related to the user in the first utterance character string. Can generate a second utterance character string that is easy to recognize and speak.

<Modification example>
Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and can be appropriately modified and implemented without departing from the spirit of the present invention. is there.

In the above-described embodiment, as an example of the voice authentication device of the present invention, the voice authentication device 20 as a cloud server that acquires the voice data of the user generated by the terminal device 10 via the network 30 has been described. The present invention is not limited to this.

Not all the functions of the voice authentication device of the present invention need to be mounted on the voice authentication device 20 as a cloud server, and any of the functions may be mounted on the terminal device 10 side. For example, any one of the first utterance character string generation unit 22, the authentication character string generation unit 23, the voice recognition unit 24, the voiceprint data generation unit 25, and the authentication unit 26 may be mounted on the terminal device 10.

Further, all the functions of the voice authentication device of the present invention may be mounted on the terminal device side. In this case, the voice authentication system does not require any configuration other than the terminal device. However, even in that case, for example, only the service providing unit that provides the service to the user may be installed outside the terminal device, and in that case, the terminal device and the service providing unit may be able to communicate via the network. ..

In the above-described embodiment, authentication is performed using only the voice of the user, but the present invention is not limited to this, and multi-factor authentication may be performed by adding authentication of other elements. Examples of multi-factor authentication include: That is, the user's biometric information (face image, iris image, fingerprint image, finger vein image, etc.) is registered in advance, and when the user requests the provision of a service, the biometric information is used together with the voice authentication process. Perform the authentication that was done. If the service is provided only when both are successful, the security of authentication can be further improved.

The elements used other than the voice data may be not only biological information but also knowledge information (what the user knows) and possession information (what the user has). That is, multi-factor authentication may be performed.

For example, when authentication is performed using both voice and password, by asking the user to enter the password before voice, the user's speaker model associated with the password can be referred to, and at the time of voice authentication, the user's speaker model can be referred to. It is possible to reduce the time required to narrow down the speaker model to be referred to. As a result, both high security and user convenience can be achieved at the same time.

Further, in the above-described embodiment, the additional character string is simply a character string different from the authentication character string, but for example, the additional character string may be changed by the user (using user data). In such a case, the number of users whose voice recognition results match can be further limited, and thereby the accuracy of voiceprint authentication and personal authentication can be improved.

The present invention is suitable for a voice authentication device or a voice authentication system that performs voice authentication of a user.

1 Voice authentication system 10 Terminal device 11 Communication unit 12 Voice data generation unit 13 Presentation unit 14 Input unit 20 Voice authentication device 21 Communication unit 22 First speech character string generation unit 23 Authentication character string generation unit 24 Voice recognition unit 241 Noise Reduction unit 242 Speech detection unit 243 Feature extraction unit 244 Sound element conversion unit 245 Character string conversion unit 246 Sound model 247 Language model 248 Additional character string removal unit 25 Voice pattern data generation unit 26 Authentication unit 27 Service provision unit 28 User data acquisition unit 29 Storage 30 network

Claims (15)

An authentication character string generator that generates an authentication character string that includes at least one of the characters included in the predetermined character string, and
From the first voice data generated by the user speaking the predetermined character string, the first voice pattern data of the portion corresponding to the authentication character string is generated, and the user speaks the authentication character string. A voice string data generation unit that generates a second voice pattern data based on the generated second voice data,
An authentication unit that authenticates the user by collating the first voiceprint data with the second voiceprint data.
A voice authentication device. A voice recognition unit that converts voice data into a character string, divides the voice data based on the converted character string, and outputs the character string and the divided voice data.
Further prepare,
The voice authentication device according to claim 1. The authentication character string generation unit generates the authentication character string by combining a plurality of characters included in the predetermined character string.
The voice authentication device according to claim 1 or 2. The authentication character string generation unit generates a plurality of the authentication character strings different from each other, and generates an utterance authentication character string including at least one of the plurality of authentication character strings.
The voiceprint data generation unit generates the second voiceprint data based on the voice data generated by the user speaking the speech authentication character string.
The voice authentication device according to any one of claims 1 to 3. The authentication character string generation unit adds an additional character string different from the authentication character string to the authentication character string to generate the speech authentication character string.
The voice authentication device according to claim 4. The authentication character string generation unit changes the number of the authentication character strings included in the utterance authentication character string based on a preset authentication level.
The voice authentication device according to claim 4 or 5. The authentication character string generation unit determines whether or not to add an additional character string to the authentication character string based on a preset authentication level.
The voice authentication device according to claim 6. The authentication level is set in relation to the loudness of the surrounding environmental sound at the time of voice authentication by the authentication unit.
The voice authentication device according to claim 6 or 7. The authentication level is set in relation to the content of the service provided to the user when the user is successfully authenticated.
The voice authentication device according to any one of claims 6 to 8. The authentication level is set in relation to the location of the service provided to the user if the user is successfully authenticated.
The voice authentication device according to any one of claims 6 to 9. The authentication level is set in relation to the execution frequency of the service provided to the user when the user is successfully authenticated.
The voice authentication device according to any one of claims 6 to 10. The voice according to any one of claims 1 to 11, further comprising a service providing unit that permits the user to provide a predetermined service when the user is successfully authenticated by the authentication unit. Authentication device. A predetermined character string generation unit that generates the predetermined character string based on the personal information of the user.
The voice authentication device according to any one of claims 1 to 12, further comprising. A voice data generator that generates voice data based on the user's utterance,
A presentation unit that presents a character string that asks the user to speak, and
An authentication character string generator that generates an authentication character string that includes at least one of the characters included in the predetermined character string, and
From the first voice data generated by the user speaking the predetermined character string, the first voice pattern data of the portion corresponding to the authentication character string is generated, and the user speaks the authentication character string. A voice string data generation unit that generates a second voice pattern data based on the second voice data generated in
An authentication unit that authenticates the user by collating the first voiceprint data with the second voiceprint data.
A voice authentication system equipped with. Generates an authentication string that contains at least one of the characters contained in the given string,
From the first voice data generated by the user speaking the predetermined character string, the first voiceprint data of the portion corresponding to the authentication character string is generated.
The second voiceprint data is generated based on the second voice data generated by the user speaking the authentication character string.
The user is authenticated using the first voiceprint data and the second voiceprint data.
Voice authentication method.

Images