Authentification device, authentification method and authentification program

Abstract

To provide an authentication device that improves accuracy of user identification through voice recognition.SOLUTION: An authentication device 1 includes: a registration unit 11 that determines a greeting phrase for each user, accepts the user's voice and text input representing an expected response to the greeting phrase, and registers them in association with the user; a learning unit 12 that generates a voice-recognizable model voice corresponding to the text, and performs machine learning for each user to create a filter for converting the user's voice into the model voice; a response reception unit 13 that outputs the greeting phrase and accepts voice input for the response; a voice recognition unit 14 that converts the response voice using the filter specific for each user, and obtains reading results by predetermined voice recognition software for the converted voice, and an expected value for the correctness of the reading results; and an authentication unit 15 that identifies and authenticates the user who has input the response voice based on the degree of match between the reading results and the expected response, and the expected value for the correctness of the reading result.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a user authentication technique using voice recognition. [Background technology]

[0002] BACKGROUND ART Conventionally, research has been conducted into speech recognition technology for converting speech into text, and in recent years, services based on speech recognition models using neural networks have been provided (for example, see Non-Patent Document 1). [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Wisper, Internet, [Retrieved February 19, 2024]<https: / / openai.com / research / whisper> Summary of the Invention [Problem to be solved by the invention]

[0004] However, even with the latest AI technology, Japanese speech recognition is still not sufficient, and even with a high-performance GPU, recognition accuracy can drop significantly depending on the speaker. For this reason, when using voice recognition for authentication, for example when key data such as a password is input by voice, if a wide range of inputs is accepted by allowing a low matching rate between the voice and the password, security strength will decrease, so the password needs to be long, but there is a problem that long passwords are easily forgotten. On the other hand, when performing authentication using a short password, it is necessary to require a high matching rate between the voice and the password, and to use a narrow input range. In this case, the problem arises that the correct password will not be recognized no matter how many times the person speaks it.

[0005] An object of the present invention is to provide an authentication device, an authentication method, and an authentication program that improve the accuracy of identifying a user through voice recognition. [Means for solving the problem]

[0006] The authentication device of the present invention comprises: a registration unit that determines a greeting for each user and accepts input of the user's own voice and text indicating an expected response to the greeting, and registers it in association with the user; a learning unit that generates a voice-recognizable model voice corresponding to the text and performs machine learning on a user-by-user basis a filter for converting the user's own voice to the model voice; a response acceptance unit that outputs the greeting and accepts input of a voice response to the greeting; a voice recognition unit that converts the voice of the response using the filter for each user and obtains a reading result of the converted voice using specified voice recognition software and an expected value for the accuracy of the reading result; and an authentication unit that identifies and authenticates the user who input the voice of the response based on the degree of match between the reading result and the expected response and the expected value for the accuracy of the reading result.

[0007] The learning unit may generate a synthetic voice using AI as the model voice.

[0008] The learning unit may receive an input of speech for speech recognition corresponding to the text from a user who inputs the expected answer, and use the input as the model speech.

[0009] The authentication unit may identify and authenticate the user corresponding to the filter that has the highest expected value for the accuracy of the read result.

[0010] The authentication unit may identify and authenticate the user corresponding to the filter that has the highest expected value of correctness of the read result among the read results that match the expected response.

[0011] The authentication device may include a service providing unit that fixes the filter corresponding to the user identified by the authentication unit and provides a predetermined service using the result of reading by the voice recognition unit.

[0012] The authentication method of the present invention is performed by a computer, and includes the following steps: a registration step of determining a greeting for each user, accepting input of the user's own voice and text indicating an expected response to the greeting, and registering the greeting in association with the user; a learning step of generating a voice-recognizable model voice corresponding to the text, and machine-learning a filter for converting the user's own voice to the model voice, for each user; a response acceptance step of outputting the greeting and accepting input of a voice response to the greeting; a voice recognition step of converting the voice of the response using the filter for each user, obtaining a reading result of the converted voice using specified voice recognition software and an expected value for the accuracy of the reading result; and an authentication step of identifying and authenticating the user who input the voice of the response based on the degree of match between the reading result and the expected response and the expected value for the accuracy of the reading result.

[0013] An authentication program according to the present invention is for causing a computer to function as the authentication device. [Effects of the Invention]

[0014] According to the present invention, it is possible to improve the accuracy of identifying a user by voice recognition in a user authentication sequence. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 2 is a diagram illustrating a functional configuration of an authentication device according to an embodiment. [Figure 2] 10 is a flowchart showing a processing flow in a learning phase in the embodiment. [Figure 3] 10 is a flowchart showing a processing flow in a utilization phase in an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0016] An example of an embodiment of the present invention will be described below. The authentication device of this embodiment improves the accuracy of voice recognition by converting the voice into another voice that is easier to recognize, and by requiring a specific greeting when used, identifying and authenticating the individual while improving the accuracy of voice recognition.

[0017] FIG. 1 is a diagram showing the functional configuration of an authentication device 1 in this embodiment. The authentication device 1 is an information processing device that includes a control unit 10, a memory unit 20, and various input / output interfaces including a microphone and a speaker, or is communicatively connected to a user terminal including a microphone and a speaker, and provides a predetermined service using the results of reading a voice input from a user using a predetermined voice recognition.

[0018] The control unit 10 is a part that controls the entire authentication device 1, and realizes each function in this embodiment by appropriately reading and executing various programs stored in the storage unit 20. The control unit 10 may be a CPU. The storage unit 20 is a storage area for various programs for causing the hardware group to function as the authentication device 1, various data, etc., and may be a ROM, RAM, flash memory, hard disk drive (HDD), or the like. Specifically, the control unit 10 functions as a registration unit 11, a learning unit 12, an answer acceptance unit 13, a voice recognition unit 14, an authentication unit 15, and a service provision unit 16 by executing software (authentication program) stored in the memory unit 20.

[0019] The registration unit 11 determines a greeting message for each user, and also receives input of the user's own voice and text indicating an expected response to the greeting message, and registers the same in association with the user. In this case, even if a common greeting message is determined for a plurality of users, it is preferable that the expected response to this greeting message be different for each user.

[0020] The learning unit 12 generates a model voice that can be recognized based on the input text, and performs machine learning on a filter for each user to convert the user's own voice into the model voice. Here, the model voice may be a voice that can be recognized with high accuracy by voice recognition software, for example, a synthetic voice generated by AI, or it may be a voice generated by combining speech data from a person with an easy-to-listen speaking style, such as an announcer.

[0021] In particular, if a user has a distinctive voice, they may know how to speak in an easy-to-listen-to voice and may be able to speak in an easy-to-listen-to voice temporarily. In such a case, the learning unit 12 may receive an input of a voice for speech recognition corresponding to the text from the user who input the expected answer, and use this as a model voice.

[0022] The answer receiving unit 13 outputs a registered greeting message and receives a voice input of an answer to the greeting message.

[0023] The voice recognition unit 14 converts the received response voice using a filter for each user, and obtains the string of characters read by specified voice recognition software for each converted voice, and the expected value of the accuracy of the read result for the voice.

[0024] The authentication unit 15 identifies and authenticates the user who inputs the voice response based on the degree of match between the read result by the voice recognition software and the expected response, and the expected value of the correctness of the read result. For example, if there is converted speech in which a character string matching the expected answer can be read by speech recognition, the authentication unit 15 selects the filter with the highest expected value of correctness from among these, and identifies and authenticates the user who answered as a registered user corresponding to the selected filter. Also, if a character string matching the expected answer cannot be read, the authentication unit 15 may simply identify and authenticate the user as a registered user corresponding to the filter with the highest expected value of correctness of the read character string that is equal to or greater than a threshold, or may fail authentication. In addition, conditions for identification and authentication may be set appropriately according to the required authentication strength, such as by setting a threshold value for the degree of match.

[0025] The service providing unit 16 fixes the filter corresponding to the user identified by the authentication unit 15 from among the multiple filters learned for each user, and then provides a predetermined service using the reading result by the voice recognition unit 14.

[0026] FIG. 2 is a flowchart showing the flow of processing in the learning phase of the authentication device 1 in this embodiment.

[0027] In step S1, the registration unit 11 determines a greeting message for each user by prompting the user to select, input, or presenting a greeting message to the user at the time of user registration. In step S2, the registration unit 11 asks the user to input an expected response to the determined greeting by voice in normal speaking style A, and also asks the user to input the voice input in text. In step S3, the registration unit 11 stores the greeting and the voice and text of the expected response in association with the user. If there are multiple expected responses, the registration unit 11 registers multiple patterns.

[0028] In step S4, the learning unit 12 prepares a model voice B suitable for voice recognition for the expected response registered by the registration unit 11. In step S5, the learning unit 12 learns a filter for converting speech A into speech B for each user through machine learning.

[0029] FIG. 3 is a flowchart showing the flow of processing in the usage phase of the authentication device 1 in this embodiment. This process is executed when user authentication becomes necessary, such as when starting to use a specific service. As a means of speech recognition, software is provided that reads speech, converts it into a string of characters, and outputs an expected value for the accuracy of the string.

[0030] In step S11, the answer receiving unit 13 displays or outputs a voice message to the user, requesting an expected answer to the greeting. For example, the answer receiving unit 13 may ask a user who works in a voice-related job, "Hello, how is your voice today?", and request an expected answer registered as a response related to the voice condition, such as "Good" or "Not so good."

[0031] In step S12, the voice recognition unit 14 converts the user's response to the greeting in step S11 using the filters Y1 to Yn that have been trained for the registered users U1 to Un, respectively, and performs predetermined voice recognition on each result.

[0032] In step S13, the authentication unit 15 identifies the user whose answer was accepted in step S11 from among the registered users based on the degree of match between the voice recognition result and the expected answer and the expected value of the correctness of the result string for the voice, and immediately upon identification, the authentication unit 15 considers this user to have been authenticated.

[0033] Thereafter, the service providing unit 16 performs voice recognition when providing a predetermined service after converting the input voice using a filter corresponding to the authenticated user. For example, when using an API involving voice recognition, a user identification value is assigned, and a filter corresponding to this user is used.

[0034] According to this embodiment, the authentication device 1 converts the speech of a response to a greeting into speech with high recognition accuracy through a filter trained for each user. Then, if a character string of an expected response can be read from the converted speech by speech recognition, the user corresponding to this filter is authenticated based on the degree of match. Therefore, the authentication device 1 realizes user authentication by being capable of voice recognition using both the tone of voice and the response character string of the response to the greeting.

[0035] In actual use cases, speech recognition, especially in Japanese, is difficult to accurately convert into a string of characters without context, even with technology such as AI. This is thought to be because the amount of information required for speech recognition is insufficient with simple sound data alone due to the wide variety of voice tones. In this embodiment, the fact that speech recognition will not be successful without context is utilized for authentication. The authentication device 1 determines the voice tones for which speech recognition will be successful using a filter trained for each user, and this determination can be used for authentication.

[0036] Furthermore, by using a filter corresponding to an authenticated user, the authentication device 1 can continuously identify speakers who are using the service while simultaneously improving the accuracy of speech recognition. Furthermore, the authentication method of this embodiment does not require the speaking of secret key data such as a password, and it is difficult for someone who happens to listen to a conversation with a voice recognition device to determine whether the current conversation is part of the authentication procedure, making it difficult for unauthorized intervention in the authentication sequence. In this way, the authentication device 1 can improve both the accuracy of voice recognition and the strength of authentication.

[0037] The authentication method of this embodiment can further increase the strength of authentication by increasing the uniqueness of the expected response to the greeting or tone of voice. Furthermore, if the greeting is given at the start of use, the timing of the greeting itself will be revealed, but by making the greeting itself and the method of requesting a response (greeting phrase, etc.) distinctive, it is possible to conceal how authentication will begin, and this combination can ensure sufficient strength.

[0038] Furthermore, while typical voice recognition technology aims to be able to respond to all human voices, when the present embodiment is applied to a voice input terminal (microphone device) such as a smartphone or smart speaker, the number of users is limited, so by limiting the expected responses to greetings to a few, it is possible to identify and authenticate users efficiently and accurately with only a small number of filters.

[0039] This will improve the accuracy of user identification through voice recognition, for example, and contribute to Goal 9 of the United Nations-led Sustainable Development Goals (SDGs), which is to "Build resilient infrastructure, promote sustainable industrialization and foster innovation."

[0040] Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments. Furthermore, the effects described in the above-described embodiments are merely a list of the most preferable effects resulting from the present invention, and the effects of the present invention are not limited to those described in the embodiments.

[0041] The authentication method by the authentication device 1 is realized by software. When realized by software, the programs that make up this software are installed in an information processing device (computer). These programs may be distributed by being recorded on removable media such as CD-ROMs, or may be distributed by being downloaded to computers via a network. [Explanation of symbols]

[0042] 1 Authentication device 10 Control Unit 11 Registration Department 12 Learning Department 13. Response Reception Department 14 Voice Recognition Unit 15 Authentication Section 16 Service Provision Department 20 Memory section

Claims

1. a registration unit that determines a greeting message for each user, receives input of the user's own voice and text indicating an expected response to the greeting message, and registers the input in association with the user; a learning unit that generates a model voice that can be recognized by the text and performs machine learning on a user-by-user basis to create a filter for converting the user's own voice into the model voice; a response receiving unit that outputs the greeting message and receives a voice response to the greeting message; a voice recognition unit that converts the voice of the answer using the filter for each user and obtains a reading result of the converted voice using predetermined voice recognition software and an expected value of the accuracy of the reading result; an authentication unit that identifies and authenticates the user who inputs the voice response based on the degree of match between the read result and the expected response and the expected value of the correctness of the read result.

2. The authentication device according to claim 1 , wherein the learning unit generates a synthesized voice using AI as the model voice.

3. The authentication device according to claim 1 , wherein the learning unit receives an input of a voice for speech recognition corresponding to the text from the user who input the expected answer, and uses the voice as the model voice.

4. The authentication device according to claim 1 , wherein the authentication unit identifies and authenticates the user corresponding to the filter that has the highest expected value of correctness of the read result.

5. The authentication device according to claim 1 , wherein the authentication unit identifies and authenticates the user corresponding to the filter that has the highest expected value of correctness of the read result among the read results that match the expected response.

6. 6. The authentication device according to claim 1, further comprising a service providing unit that fixes the filter corresponding to the user identified by the authentication unit and provides a predetermined service using the result of reading by the voice recognition unit.

7. a registration step of determining a greeting message for each user, receiving input of the user's own voice and text indicating an expected response to the greeting message, and registering the voice and text in association with the user; a learning step of generating a model voice capable of speech recognition corresponding to the text and machine learning a filter for converting the user's own voice into the model voice for each user; a response receiving step of outputting the greeting message and receiving a voice response to the greeting message; a speech recognition step of converting the answer voice using the filter for each user, and obtaining a reading result of the converted voice using predetermined speech recognition software and an expected value of the accuracy of the reading result; and an authentication step of identifying and authenticating the user who input the voice response based on the degree of match between the read result and the expected response and the expected value of the correctness of the read result.

8. An authentication program for causing a computer to function as the authentication device according to any one of claims 1 to 5.

Images