Program, information processing device and information processing method

Abstract

To provide a program, an information processing device and an information processing method capable of preventing damage of a crime due to a deep fake that pretends to be an acquaintance by using a generated synthetic voice.SOLUTION: A speech voice of a caller 300 is acquired, the acquired speech voice is inputted to a classification model 1061 learned so as to output classification information as to whether the speech voice is a synthetic voice and the classification information is outputted when the speech voice is inputted, and the outputted classification information is outputted to a listener.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a program, an information processing device, and an information processing method. [Background technology]

[0002] In recent years, there has been a rise in crimes where people pretend to be acquaintances to deceive recipients, such as "it's me" scams and bank transfer scams. In response to this, Patent Document 1 discloses a malicious call countermeasure system that includes a call pattern analysis unit and determines the risk level according to the analysis results of whether or not the call pattern during a call corresponds to a specific pattern. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-027424 Summary of the Invention [Problem to be solved by the invention]

[0004] However, the above-mentioned malicious call prevention system cannot prevent damage caused by such fraud if the content of the call does not fall under predetermined call patterns, and it cannot deal with so-called deep fake crimes, which have recently been committed using synthetic voices generated by AI (artificial intelligence).

[0005] The present disclosure has been made in consideration of the above circumstances, and aims to provide a program, an information processing device, and an information processing method that can prevent crimes committed by deepfakes, in which generated synthetic voice is used to impersonate an acquaintance. [Means for solving the problem]

[0006] The program according to the present disclosure causes a computer to execute a process of acquiring a call voice from a speaker, inputting the acquired call voice into a learning model that has been trained to output classification information as to whether the call voice is synthetic voice or not when the call voice is input, outputting the classification information, and outputting the output classification information to the recipient.

[0007] The information processing device according to the present disclosure includes a control unit, and the control unit acquires a call voice from a speaker, and when the call voice is input, inputs the acquired call voice into a learning model trained to output classification information as to whether the call voice is a synthetic voice or not, and outputs the output classification information to the recipient.

[0008] The information processing method according to the present disclosure causes a computer to execute a process of acquiring a call voice from a speaker, inputting the acquired call voice into a learning model that has been trained to output classification information as to whether the call voice is synthetic voice or not when the call voice is input, outputting the classification information, and outputting the output classification information to a recipient. Effect of the Invention

[0009] According to the present invention, it is possible to prevent crimes committed by deepfakes, in which generated synthetic voice is used to impersonate an acquaintance. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is an explanatory diagram conceptually illustrating an embodiment according to the present disclosure. [Diagram 2] FIG. 2 is a functional block diagram showing a configuration example of a terminal device according to the present embodiment. [Diagram 3] FIG. 11 is an illustrative diagram showing an example of a warning screen displayed on a display unit in the embodiment. [Figure 4] FIG. 1 is an explanatory diagram of a classification model. [Diagram 5] FIG. 2 is an explanatory diagram showing an example of a record layout of classification training data in a training data DB. [Figure 6] 13 is a flowchart showing a procedure for generating a classification model. [Figure 7] 6 is a flowchart for explaining an example of how the terminal device of the first embodiment responds when an incoming call is received. [Figure 8] FIG. 1 is an explanatory diagram conceptually illustrating an embodiment according to the present disclosure. [Figure 9] FIG. 2 is a functional block diagram showing a configuration example of a terminal device according to the present embodiment. [Figure 10] FIG. 2 is an explanatory diagram showing an example of a record layout of response training data in a training data DB. [Figure 11] FIG. 13 is an explanatory diagram showing an example of a record layout of a learning model management DB. [Figure 12] 13 is a flowchart showing a procedure for generating a response model. [Figure 13] 10 is a flowchart for explaining a response example of a terminal device of embodiment 2 when an incoming call is received. [Figure 14] 13 is an exemplary diagram showing a modified example of a warning screen displayed on the display unit. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, the present invention will be described with reference to the drawings showing an embodiment thereof.

[0012] (Embodiment 1) FIG. 1 is an explanatory diagram conceptually illustrating an embodiment according to the present disclosure. In this embodiment, reference numeral 1 denotes a terminal device (information processing device), and reference numeral 100 denotes a server device. The terminal device 1 is connected to the server device 100 via a network N including a mobile phone communication network, a wireless LAN (Local Area Network), the Internet, and the like. In this embodiment, the terminal device 1 is, for example, a mobile phone, and when an incoming call is received, the terminal device 1 receives from the server device 100 a classification result as to whether the call voice of the sender is a synthetic voice or not, and responds appropriately to the call from the malicious sender 300.

[0013] FIG. 2 is a functional block diagram showing an example of the configuration of the terminal device 1 according to this embodiment. The terminal device 1 includes a control unit 11, a storage unit 12, a speaker 13, a microphone 14, a recording unit 15, a display unit 16, a reception unit 17, a first communication unit 18, a second communication unit 19, and a reading unit 20. Each component is connected via a bus.

[0014] The control unit 11 includes an arithmetic processing device such as a CPU, an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit), etc., and performs various information processing, control processing, etc. related to the terminal device 1 by reading and executing a control program P2 stored in the memory unit 12.

[0015] The storage unit 12 includes memory elements such as a RAM (Random Access Memory) and a ROM (Read Only Memory), and stores a control program P2 or data required for the control unit 11 to execute processing. The storage unit 12 also temporarily stores data required for the control unit 11 to execute arithmetic processing. Furthermore, as described above, a notification destination 121 for notifying the user when the call voice of the sender is a synthetic voice is stored. The notification destination 121 is, for example, a contact address for a relative, family member, or the police, and is received in advance from the user via the reception unit 17.

[0016] The speaker 13 outputs the call voice from the sender, and the microphone 14 outputs the voice of the user (listener) to the sender. In addition, the recording unit 15 records the call voice of the sender received by the first communication unit 18 in response to an instruction from the control unit 11, and stores it in the storage unit 12.

[0017] The display unit 16 is a liquid crystal display, an organic EL (electroluminescence) display, or the like, and displays various information according to instructions from the control unit 11. For example, the display unit 16 displays a warning screen, which will be described later. Furthermore, the reception unit 17 is a touch panel, and is provided on the display unit 16. The control unit 11 receives an instruction from the user via the reception unit 17 in response to a touch operation by the user.

[0018] 3 is an exemplary diagram showing an example of a warning screen displayed on the display unit 16 in this embodiment. The warning screen displays a warning message 161 saying "This is synthetic voice. Be careful", and for example, the warning message 161 flashes. In addition, the warning screen displays a confidence level 162 indicating the probability that the call voice of the sender sent from the server device 100 is synthetic voice, a "START" button 163 (recording reception unit) for receiving an instruction to record the call voice of the sender, and an "EXECUTE" button 164 (notification reception unit) for receiving an instruction to notify the notification destination 121 when the call voice of the sender is synthetic voice. For example, the "START" button 163 and the "EXECUTE" button 164 are so-called toggle buttons.

[0019] The first communication unit 18 accepts an incoming call and makes a call via the mobile phone communication network. That is, the first communication unit 18 receives a call voice from a caller and transmits a call voice from a call recipient to the caller, thereby performing audio output.

[0020] The second communication unit 19 performs data communication with the server device 100, for example, via a wireless LAN and the Internet. For example, the second communication unit 19 transmits a call voice from a talker to the server device 100, and receives from the server device 100 a classification result as to whether the call voice of the talker is a synthetic voice or not, as described later.

[0021] The reading unit 20 reads a portable storage medium 200a including a CD (Compact Disc)-ROM or a DVD (Digital Versatile Disc)-ROM. The control unit 11 may read the control program P2 from the portable storage medium 200a via the reading unit 20 and store it in the storage unit 12. The control unit 11 may also download the control program P2 from another computer via a network N or the like and store it in the storage unit 12. Furthermore, the control unit 11 may read the control program P2 from a semiconductor memory 200b.

[0022] The server device 100 includes a control unit 101, a storage unit 102, a communication unit 103, a reading unit 104, and a large-capacity storage unit 105. Each component unit is connected via a bus (see FIG. 1).

[0023] The control unit 101 includes an arithmetic processing unit such as a CPU, an MPU, a GPU, etc., and performs various information processing, control processing, etc. related to the server device 100 by reading and executing a control program P1 stored in the storage unit 102. The control program P1 can be deployed so as to be executed on a single computer, or on one site, or distributed across multiple sites and on multiple computers interconnected by a communication network. In this embodiment, the control unit 101 will be described as being a single processor, but it may also be a multiprocessor.

[0024] The storage unit 102 includes memory elements such as RAM and ROM, and stores the control program P1 or data required for the control unit 101 to execute processing. The storage unit 102 also temporarily stores data required for the control unit 101 to execute arithmetic processing.

[0025] The communication unit 103 performs data communication with the terminal device 1 via, for example, a wireless LAN and the Internet. For example, the communication unit 103 transmits to the terminal device 1 a result of discrimination as to whether the call voice from the sender is a synthetic voice or not.

[0026] The reading unit 104 reads the portable storage medium 200a including a CD-ROM or a DVD-ROM. The control unit 101 may read the control program P1 from the portable storage medium 200a via the reading unit 104 and store it in the mass storage unit 105. The control unit 101 may also download the control program P1 from another computer via a network or the like and store it in the mass storage unit 105. Furthermore, the control unit 101 may read the control program P1 from the semiconductor memory 200b.

[0027] The mass storage unit 105 includes a recording medium such as a hard disk drive (HDD), a solid state drive (SSD), etc. The mass storage unit 105 includes a learning model database (DB) 106 and a training data DB 107.

[0028] The learning model DB 106 stores a classification model 1061 (learning model). The classification model 1061 is a learned learning model generated by machine learning. The classification model 1061 classifies whether the call voice of the sender sent from the terminal device 1 is a synthetic voice or not. In other words, the classification model 1061 is a classifier that classifies whether the call voice of the sender is a synthetic voice or a human voice.

[0029] FIG. 4 is an explanatory diagram of the classification model 1061. The classification model 1061 is configured by a neural network in which a plurality of neurons (also called nodes) are connected. The classification model 1061 has an input layer to which the call voice of the sender is input, a hidden layer including a convolution layer or a pooling layer that extracts features, and an output layer that outputs a confidence level (score) indicating the probability that the call voice of the sender is a synthetic voice, and is trained to output a confidence level (score) when the call voice of the sender is input. The confidence level score is a numerical value between 0 and 1.

[0030] The type of neural network constituting the classification model 1061 is arbitrary and is appropriately selected according to the format of the input data and the format of the data to be output. For example, the classification model 1061 employs a learning model such as an RNN (Recurrent Neural Network), Transformer, or GPT (Generative Pre-Training)-2 that handles time-series information.

[0031] The training data DB 107 stores training data for constructing (creating) the classification model 1061. Specifically, the training data DB 107 stores classification training data 1071 that is training data for the classification model 1061.

[0032] 5 is an explanatory diagram showing an example of a record layout of the classification training data 1071 in the training data DB 107. The classification training data 1071 includes a training ID string, an input data string, and an output data string.

[0033] The training ID column stores a unique ID for each piece of training data to identify the training data. The input data column stores input data of telephone call voice including synthetic voice and human voice. The output data column stores the classification result, i.e., training data. For the input data, the output data column stores a label that is a human voice (e.g., "0") and a label that is a synthetic voice generated by AI (Artificial Intelligence) (e.g., "1").

[0034] Fig. 6 is a flowchart showing the procedure of the process of generating the classification model 1061. The process of generating the classification model 1061 by performing machine learning will be described with reference to Fig. 6.

[0035] The control unit 101 of the server device 100 acquires a plurality of pieces of input data and output data corresponding to the input data from the classification training data 1071 of the training data DB 107 (see FIG. 5) (step S101).

[0036] The control unit 101 uses the acquired input data and the output data corresponding to the input data to generate a classification model 1061 that outputs a confidence score that the call voice of the sender is a synthetic voice (step S102).

[0037] The control unit 101 sequentially inputs the input data to the classification model 1061 and sequentially outputs the confidence levels (scores). The control unit 101 compares the output confidence levels (scores) with the output data, which is teacher data included in the classification training data 1071, and optimizes various parameters of the intermediate layer so that the two are similar to each other, thereby generating the classification model 1061.

[0038] The control unit 101 stores the generated classification model 1061 in the learning model DB 106 of the mass storage unit 105 (step S103), and ends the series of processes.

[0039] FIG. 7 is a flowchart for explaining an example of how the terminal device 1 of the first embodiment responds when an incoming call is received.

[0040] When the terminal device 1 accepts an incoming call (step S201) and the user (receiver) of the terminal device 1 performs an operation to respond, the control unit 11 captures the call voice (voice data) of the caller received via the first communication unit 18 and transmits it to the server device 100 via the second communication unit 19 (step S202).

[0041] Voice data of the call voice of the sender from the terminal device 1 is received by the communication unit 103 of the server device 100. At this time, the control unit 101 of the server device 100 inputs the call voice of the sender received via the communication unit 103 to the input layer of the classification model 1061, and transmits a confidence factor (score) output from the output layer of the classification model 1061 to the terminal device 1 via the communication unit 103. The confidence factor (score) from the server device 100 is received by the second communication unit 19 of the terminal device 1 (step S203).

[0042] The control unit 11 of the terminal device 1 determines whether the call voice of the sender is a synthetic voice or not based on the confidence level (score) from the server device 100 (step S204). If the confidence level (score) from the server device 100 is less than a threshold, the control unit 11 determines that the voice is not a synthetic voice, that is, that the voice is a human voice (step S204: NO), and ends the process.

[0043] On the other hand, if the certainty factor (score) from the server device 100 is equal to or greater than the threshold, the control unit 11 determines that it is synthetic speech (step S204: YES), and displays the above-mentioned warning screen on the display unit 16 (see FIG. 3). That is, the display unit 16 flashes a warning message 161 saying "This is synthetic speech. Be careful", and displays the certainty factor (score) 162 from the server device 100 (step S205).

[0044] Next, when the user touches the "START" button 163 on the warning screen to record the call voice of the sender, the control unit 11 instructs the recording unit 15 to record the call voice. In response, the recording unit 15 starts recording the call voice of the sender (step S206).

[0045] In addition, in order to notify the notification destination 121 that the call voice of the sender is a synthetic voice, in response to the user touching the "Execute" button 164 on the warning screen, the control unit 11 instructs the first communication unit 18 to notify the notification destination 121 in the storage unit 12. In response to this, the first communication unit 18 makes a call to the notification destination 121 (step S207). For example, the first communication unit 18 contacts the notification destination 121, such as a relative, family member, or the police, and transmits a voice, which is stored in advance in the storage unit 12, indicating that the call voice of the sender is a synthetic voice.

[0046] After that, the control unit 11 judges whether the call has ended (step S208). If it is judged that the call has not ended (step S208: NO), the control unit 11 repeats such a judgment until the call ends.

[0047] For example, when the warning screen is displayed and the user hangs up the phone, the control unit 11 determines that the call has ended (step S208: YES) and instructs the recording unit 15 to stop recording the call voice of the sender, and in response to this instruction, the recording unit 15 stops recording the call voice of the sender (step S209). The call voice of the sender recorded by the recording unit 15 is stored in, for example, the storage unit 12 in association with the date, and the process ends.

[0048] In this embodiment, as described above, when an incoming call is received, the terminal device 1 acquires the call voice of the sender and transmits it to the server device 100. If the call voice of the sender is a synthetic voice based on the confidence factor 162 from the server device 100, the terminal device 1 issues a warning to the user (listener). Therefore, users can take preventive measures such as being more vigilant or hanging up the phone, and can prevent themselves from being deceived by deepfakes, in which synthetic voices generated by AI are used to impersonate acquaintances.

[0049] In addition, in this embodiment, as described above, the certainty factor (score) 162 is displayed on the warning screen in addition to the warning message 161, so that the recipient can classify for himself or herself whether the call voice of the sender is synthetic voice or not based on the certainty factor 162.

[0050] In addition, in this embodiment, as described above, if the call voice of the sender is a synthetic voice, the call voice of the sender can be recorded, so that evidence can be secured in case damage caused by the deep fake occurs.

[0051] Furthermore, in this embodiment, as described above, if the caller's voice is synthetic voice, the notification destination 121 can be notified, so that help can be sought from those around when dealing with the deep fake.

[0052] In the above, an example has been described in which the warning screen (warning message 161) is displayed on the display unit 16 to warn the user when the call voice of the sender is a synthetic voice, but the present invention is not limited to this. For example, the warning may be given via the speaker 13. For example, when the call voice of the caller is a synthetic voice, the speaker 13 may be configured to output a predetermined warning sound or to read out a predetermined warning phrase. In addition, the terminal device 1 may be configured to include a vibration unit and to issue a warning using the vibration unit. For example, when the call voice of the caller is a synthetic voice, the vibration unit may be configured to vibrate in a predetermined pattern.

[0053] In the above, the server device 100 is described as having the learning model DB 106 (classification model 1061), but the present invention is not limited to this. The terminal device 1 may be configured to have the learning model DB 106 (classification model 1061).

[0054] (Embodiment 2) FIG. 8 is an explanatory diagram conceptually illustrating an embodiment according to the present disclosure. In the second embodiment, a terminal device 1 is connected to a server device 100 via a network N, as in the first embodiment.

[0055] FIG. 9 is a functional block diagram showing an example of the configuration of the terminal device 1 according to the present embodiment. The terminal device 1 is a mobile phone, and includes a control unit 11, a storage unit 12, a speaker 13, a microphone 14, a recording unit 15, a display unit 16, a reception unit 17, a first communication unit 18, a second communication unit 19, and a reading unit 20. These components have already been described in the first embodiment, and detailed description thereof will be omitted.

[0056] The terminal device 1 according to the second embodiment further includes a TTS (Text to Speech) unit 21. As described later, the TTS unit 21 is a so-called voice synthesizer that converts a response text sent from the server device 100 into an audio output.

[0057] The server device 100 includes a control unit 101, a storage unit 102, a communication unit 103, a reading unit 104, and a large-capacity storage unit 105 (see FIG. 8). The control unit 101, the storage unit 102, the communication unit 103, and the reading unit 104 are the same as those in the first embodiment, and detailed description thereof will be omitted.

[0058] The mass storage unit 105 includes a recording medium such as an HDD, an SSD, etc. The mass storage unit 105 includes a learning model DB 106, a training data DB 107, and a learning model management DB 108.

[0059] The learning model DB 106 stores a classification model 1061 and a response model 1062 (language generation model). The classification model 1061 is a trained learning model generated by machine learning, and classifies the call voice of the speaker sent from the terminal device 1 as synthetic voice or not, as in the first embodiment.

[0060] In addition, the response model 1062 is a trained language generation model generated by machine learning. When the control unit 101 determines that the call voice of the speaker sent from the terminal device 1 is a synthetic voice, the response model 1062 generates a response text to prevent deception by the speaker.

[0061] The response model 1062 is configured by a neural network in which a plurality of neurons (also called nodes) are connected. The response model 1062 has an input layer to which a text (hereinafter referred to as a call content text) representing the call content of the sender is input, a hidden layer including a convolution layer or a pooling layer for extracting features, and an output layer for outputting a response text (response data) corresponding to the call content of the sender, and is trained to output a response text corresponding to the input call content text when the call content text of the sender is input. The type of neural network that constitutes the response model 1062 is arbitrary, and for example, a learning model such as GPT-2 is adopted.

[0062] The training data DB 107 stores training data for constructing (creating) the classification model 1061 and the response model 1062. Specifically, the training data DB 107 stores classification training data 1071, which is training data for the classification model 1061, and response training data 1072, which is training data for the response model 1062. The classification training data 1071 has already been described, and only the response training data 1072 will be described below.

[0063] 10 is an explanatory diagram showing an example of a record layout of the response training data 1072 in the training data DB 107. The response training data 1072 includes a training ID string, an input data string, and an output data string.

[0064] The training ID column stores a unique ID of each training data to identify the training data. The input data column stores various call content texts as input data. The output data column stores appropriate response texts that can correspond to the input call content texts.

[0065] The learning model management DB 108 stores information about the trained classification model 1061 and the response model 1062. Specifically, the learning model management DB 108 stores files of the classification model 1061 and the response model 1062, date and time information when the classification model 1061 and the response model 1062 were generated, and the like.

[0066] FIG. 11 is an explanatory diagram showing an example of a record layout of the learning model management DB 108. As shown in FIG. The learning model management DB108 includes a model ID column, a learning model column, a type column, and a generation date and time column. The model ID column stores a uniquely specified learning model ID to identify the trained classification model 1061 and the response model 1062. The learning model column stores files of the trained classification model 1061 and the response model 1062. The type column stores the type of the learning model. The type is classified, for example, by output data output by each of the classification model 1061 and the response model 1062. The generation date and time column stores date and time information on the generation of the trained classification model 1061 and the response model 1062.

[0067] Fig. 12 is a flowchart showing the procedure of the process of generating the response model 1062. The process of generating the response model 1062 by performing machine learning will be described with reference to Fig. 11.

[0068] The control unit 101 of the server device 100 acquires a plurality of pieces of input data and output data corresponding to the input data from the response training data 1072 of the training data DB 107 (see FIG. 10) (step S301).

[0069] The control unit 101 generates a response model 1062 that outputs a response text corresponding to the input call content text, using the acquired input data and the output data corresponding to the input data (step S302).

[0070] The control unit 101 sequentially inputs the input data to the response model 1062 and sequentially outputs the response texts. The control unit 101 compares the output response texts with the output data, which is teacher data included in the response training data 1072, and optimizes various parameters of the intermediate layer so that the two are similar to each other, thereby generating the response model 1062.

[0071] The control unit 101 stores the generated classification model 1061 in the learning model DB 106 of the mass storage unit 105 (step S303), and ends a series of processes. Specifically, the control unit 101 assigns a model ID to the generated response model 1062, and stores the file and generation date and time of the response model 1062 as one record in the learning model management DB 108 in association with the assigned model ID.

[0072] The server device 100 according to the second embodiment further includes an STT (Speech to Text) unit 109. The STT unit 109 converts the call voice (voice data) of the sender sent from the terminal device 1 into the call content text. The call content text of the sender converted by the STT unit 109 is input to the response model 1062.

[0073] FIG. 13 is a flowchart for explaining an example of how the terminal device 1 of the second embodiment responds when an incoming call is received.

[0074] When the terminal device 1 accepts an incoming call (step S401) and the user (receiver) of the terminal device 1 performs an operation to respond, the control unit 11 captures the call voice (voice data) of the caller received via the first communication unit 18 and transmits it to the server device 100 via the second communication unit 19 (step S402).

[0075] Voice data of the call voice of the sender from the terminal device 1 is received by the communication unit 103 of the server device 100 and temporarily stored in the storage unit 102. The control unit 101 of the server device 100 reads out the call voice of the sender stored in the storage unit 102 and inputs it to the input layer of the classification model 1061, and temporarily stores in the storage unit 102 the confidence factor (score) output from the output layer of the classification model 1061.

[0076] At this time, the control unit 101 judges whether the call voice of the sender is a synthetic voice or not based on the confidence factor (score) output from the output layer of the classification model 1061. If the confidence factor (score) is equal to or greater than a threshold, the control unit 101 judges that the call voice of the sender is a synthetic voice, reads out the call voice of the sender stored in the storage unit 102, and inputs it to the STT unit 109 to acquire the call content text. Next, the control unit 101 inputs the acquired call content text to the response model 1062, and acquires the response text output from the response model 1062. The control unit 101 transmits the acquired response text together with the confidence factor (score) to the terminal device 1 via the communication unit 103. The response text and the confidence factor (score) from the server device 100 are received by the second communication unit 19 of the terminal device 1 (step S403). The control unit 11 of the terminal device 1 temporarily stores the received confidence factor (score) and response text in the storage unit 12.

[0077] Next, the control unit 11 displays the above-mentioned warning screen on the display unit 16 (see FIG. 3). That is, the display unit 16 displays the warning message 161 in a flashing manner and displays the confidence level (score) 162 (step S404).

[0078] Next, when the user touches the "START" button 163 on the warning screen to record the call voice of the sender, the control unit 11 instructs the recording unit 15 to record the call voice. In response, the recording unit 15 starts recording the call voice of the sender (step S405).

[0079] In addition, in order to notify the notification destination 121 that the call voice of the sender is a synthetic voice, in response to the user touching the "Execute" button 164 on the warning screen, the control unit 11 instructs the first communication unit 18 to notify the notification destination 121 in the storage unit 12. In response to this, the first communication unit 18 makes a call to the notification destination 121 (step S406).

[0080] Next, the control unit 11 reads out the response text received from the server device 100 and stored in the storage unit 12, and sends it to the TTS unit 21. The TTS unit 21 converts the response text into voice data and outputs it as audio via the first communication unit 18 (step S407). That is, the voice reading out the response text (hereinafter, the response voice) is output to the speaker via the first communication unit 18.

[0081] For example, voice messages such as "I will report this to the police" and "The contents of the conversation have been recorded and will be submitted to the police" are output to the caller.

[0082] After that, the control unit 11 judges whether the call has ended (step S408). If it is judged that the call has not ended (step S408: NO), the control unit 11 repeats such a judgment until the call ends.

[0083] If the control unit 11 determines that the call has ended (step S408: YES), the recording unit 15 stops recording the call voice of the sender (step S409). The call voice of the sender recorded by the recording unit 15 is stored in the storage unit 12, for example.

[0084] In this embodiment, if the call voice of the sender is determined to be synthetic voice, a warning is issued to the user (receiver). Therefore, the user can take preventive measures such as increasing vigilance or hanging up the phone, and it is possible to prevent deception by deep fake, in which synthetic voice generated by AI is used to impersonate an acquaintance.

[0085] In addition, in this embodiment, as described above, when the call voice of the sender is a synthetic voice, a response voice is output to the sender based on the response text acquired from the server device 100. This makes it possible to discourage a malicious sender 300 who tries to deceive the recipient by masquerading as an acquaintance using synthetic voice, and to prevent damage caused by the deep fake from occurring.

[0086] In the above, an example has been described in which, when the call voice of the sender is a synthetic voice, the response text received from the server device 100 is automatically converted by the TTS unit 21 in the terminal device 1 into voice data and output to the sender, but the present invention is not limited to this. For example, the response text received from the server device 100 may be configured to be output to the sender in a synthetic voice generated by the TTS unit 21, or the recipient may select whether to speak to the sender in his or her own voice. This will be described in detail below.

[0087] 14 is an exemplary diagram showing a modified warning screen displayed on the display unit 16. As described above, the modified warning screen displays a flashing warning message 161, a confidence level 162, a "Start" button 163, and an "Execute" button 164.

[0088] Furthermore, the warning screen of the modified example displays a "display response" button 165 and an "automatic response" button 166. The "display response" button 165 is a button for accepting an instruction to display the response text received from the server device 100 on the display unit 16, and the "automatic response" button 166 is a button for accepting an instruction to convert the response text received from the server device 100 into voice data using the TTS unit 21 and output it to the speaker. That is, when the listener touches the "display response" button 165, the response text received from the server device 100 is displayed on the display unit 16, so that the listener can read out the response text, for example, into the microphone 14. Also, when the listener touches the "automatic response" button 166, the response text received from the server device 100 is converted into voice data by the TTS unit 21 and output to the speaker.

[0089] In the above, the server device 100 is described as having the learning model DB 106 (the classification model 1061 and the response model 1062), but the present invention is not limited to this. The terminal device 1 may be configured to have the learning model DB 106 (the classification model 1061 and the response model 1062).

[0090] Although an example in which the response model 1062 is trained using training data has been shown, the present invention is not limited to this. The response text may be acquired by inputting a prompt previously stored in the storage unit 102 together with the call content text to the response model 1062 without training the response model 1062. For example, a prompt such as "You are a GPT, an expert in dealing with fraudulent calls. Please create a sentence to repel fraudulent calls with reference to the call content text shown below" is stored. The control unit 11 inputs the prompt and the call content text (or the text of the call content history if a dialogue is already ongoing) to the response model 1062 and acquires the response text.

[0091] The same parts as those in the first embodiment are given the same reference numerals and detailed description thereof will be omitted.

[0092] The technical features (constituent elements) described in the first and second embodiments can be combined with each other, and by combining them, new technical features can be arrived at. The embodiments disclosed herein are illustrative in all respects and should not be considered as limiting. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the scope and meaning equivalent to the claims.

[0093] The matters described in each embodiment can be combined with each other. In addition, the independent claims and dependent claims described in the claims can be combined with each other in any and all combinations regardless of the citation format. Furthermore, the claims use a format in which a claim cites two or more other claims (multiple claim format), but this is not limited to this. The claims may also be written in a format in which a multiple claim cites at least one other multiple claim (multi-multi claim). [Explanation of symbols]

[0094] 1 Terminal equipment 11 Control section 100 Server device 101 Control section 163 Start button 164 Run button 1061 Classification model (learning model) 1062 Response model (language generation model)

Claims

1. Acquire the call voice of the caller, inputting the acquired telephone call voice into a learning model that has been trained to output classification information as to whether the telephone call voice is a synthetic voice when the telephone call voice is input, and outputting the classification information; Output the output classification information to the listener A program that causes a computer to carry out processing.

2. 2. The program according to claim 1, wherein when the call voice of the speaker is determined to be a synthetic voice, a warning is output to the listener.

3. 2. The program according to claim 1, further comprising: a confidence level indicating a probability that the call voice of the sender is a synthetic voice, which is output to the recipient.

4. 2. The program according to claim 1, further comprising: recording a call voice of the caller.

5. 2. The program according to claim 1, further comprising: notifying a predetermined notification destination when it is determined that the call voice of the sender is a synthetic voice.

6. 6. The program according to claim 5, further comprising: a display unit displaying a confidence level indicating the probability that the call voice of the sender is a synthetic voice; a recording reception unit that receives an instruction to record the call voice of the sender; and a notification reception unit that receives an instruction to execute the notification.

7. If it is determined that the call voice of the sender is a synthetic voice, response data is generated using data related to the call voice of the synthetic voice and a language generation model; A response based on the generated response data is output to the sender. The program according to any one of claims 1 to 6.

8. A control unit is provided, The control unit: Acquire the call voice of the caller, An information processing device that inputs acquired telephone call voice into a learning model that has been trained to output classification information as to whether or not the telephone call voice is synthetic voice when the telephone call voice is input, and outputs the output classification information to a recipient.

9. Acquire the call voice of the caller, inputting the acquired telephone call voice into a learning model that has been trained to output classification information as to whether the telephone call voice is a synthetic voice when the telephone call voice is input, and outputting the classification information; An information processing method for causing a computer to execute a process of outputting the output classification information to a listener.

Images