Voice communication device and voice communication program

Abstract

To improve the flexibility of voice response scenarios or the versatility of application in a voice communication system. [Solution] The voice call device of the present invention is a voice call device that assists a caller in achieving their goal through a voice call with a user, and has a first prompt input means for inputting a prompt to a language model to guide the caller to achieve the goal, an answer acquisition means for acquiring an answer including a question for the user from the language model into which the prompt has been input, a voice conversion means for converting a first text of the question into a first call voice, a call voice sending means for sending the first call voice to the user's telephone, a call voice receiving means for receiving a second call voice in response to the first call voice from the telephone, a text conversion means for converting the second call voice into a second text, and a second prompt input means for inputting a prompt including the second text into the language model.

Description

[Technical field]

[0001] The present invention relates to a voice communication device and a voice communication program. [Background technology]

[0002] For example, in the case of a manned call center system, there is a limit to the number of operators who can respond at the same time, and it may not be possible to respond to all incoming calls. To address this, automated voice response systems known as IVR or voicebots have been known. By playing voice guidance to customers and having them press telephone buttons, the system allows them to select requirements, answer questions, and reach the desired information or department in charge (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2010-258855 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the above conventional automated voice response systems, depending on the customer's reason for calling and the requirements of the inquiry, the question branching of the voice response (voice guidance / scenario) can become long, and the customer may have to press the button multiple times.In addition, since the only guidance is the pre-predicted voice response scenario, and the response to requirements is fixed and standardized, the flexibility of the scenario and versatility of use can be lacking.

[0005] The present invention has been proposed in view of the above-mentioned points, and an object of one aspect is to improve the flexibility of voice response scenarios or the versatility of application in a voice call system. [Means for solving the problem]

[0006] In order to solve the above problems, the telephone device according to the present invention is a voice telephone device that supports a caller in achieving his / her objective through a voice telephone call with a user, and includes: a first prompt input means for inputting a prompt for guiding the caller to the achievement of the objective into a language model; an answer acquisition means for acquiring an answer including a question for the user from the language model into which the prompt has been input; a voice conversion means for converting a first text of the question into a first telephone voice; a voice transmission means for transmitting the first telephone voice to a telephone set of the user; a voice reception means for receiving a second telephone voice in response to the first telephone voice from the telephone set; a text conversion means for converting the second telephone voice into a second text; and a second prompt input means for inputting a prompt including the second text into the language model. The prompt includes an instruction to limit the characters of an answer including a question for the user, and if the question exceeds the character limit, the question is revised to fall within the character limit. Effect of the Invention

[0007] According to the embodiment of the present invention, it is possible to improve the flexibility of voice response scenarios and the versatility of application in a voice call system. [Brief description of the drawings]

[0008] [Figure 1] 1 is a diagram illustrating an example of a network configuration of a voice call system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram illustrating an example of a hardware configuration of a voice call server according to the embodiment. [Diagram 3] FIG. 2 is a diagram illustrating an example of a software configuration of the voice call system according to the embodiment. [Figure 4] FIG. 2 is a sequence diagram relating to the voice call system. [Diagram 5] FIG. 2 is a diagram showing an example (part 1) of a text dialogue between the voice call server and the LLM according to the embodiment. [Figure 6] FIG. 2 is a diagram showing an example (part 1) of a voice call between a user and a voice call server according to the embodiment; [Figure 7] FIG. 13 is a diagram showing an example (part 2) of a text dialogue between the voice call server and the LLM according to the embodiment. [Figure 8] FIG. 11 is a diagram showing an example (part 2) of a voice call between a user and a voice call server according to the embodiment. [Figure 9] FIG. 2 is a sequence diagram of the voice call system according to the embodiment. [Figure 10] FIG. 11 is a diagram showing an example (part 3) of a text dialogue between the voice call server and the LLM according to the embodiment. [Figure 11] FIG. 11 is a diagram showing an example (part 4) of a text dialogue between the voice call server and the LLM according to the embodiment. [Figure 12] FIG. 11 is a diagram showing an example (part 3) of a voice call between a user and a voice call server according to the embodiment. [Figure 13] FIG. 11 is a diagram showing an example (part 4) of a voice call between a user and a voice call server according to the embodiment. [Figure 14] FIG. 11 is a sequence diagram of a voice call system according to the present modified example. [Figure 15] FIG. 13 is a diagram showing an example (part 5) of a text dialogue between a voice call server and an LLM according to this modified example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail with reference to the drawings. [Embodiment 1] <System configuration> (Network Configuration) 1 is a diagram showing an example of a network configuration of a voice call system according to this embodiment. The voice call system 100 in FIG.

[0010] The voice call server 10 is a server device that is used, for example, as a call center reception system and realizes voice calls with users. In particular, the voice call server 10 according to this embodiment is a so-called AI telephone server that, when receiving a voice inquiry from a user, responds to the inquiry by obtaining a response to the user's inquiry via the LLM 20.

[0011] LLM (Large Language Models) 20 is a language model that further learns (fine-tunes or RAGs) a knowledge base that aggregates knowledge documents (e.g., product manuals, service manuals, pamphlets, Q&A collections, etc.) for answering user inquiries, from a general-purpose natural language processing model (LLM: Large Language Models) that has been pre-trained on grammar, word meanings, etc. using a large amount of text data. Note that the general-purpose natural language processing model can be, for example, a language model (generative AI) based on BERT (Bidirectional Encoder Representations from Transformers), GPT (Generative Pre-trained Transformer)-3.5, GPT-4, etc.

[0012] The telephone 30 is a telephone capable of voice calls, such as a general telephone, a mobile phone, a smartphone, or an IP phone, and is a calling device for a user (caller) who uses the voice call system 100. A user can make an inquiry by voice call by calling a specific telephone number using the telephone 30 to call the voice call server 10. When the telephone 30 is a smartphone, a specific telephone application program is installed in advance.

[0013] The network 60 is a communication network including wired and wireless networks, and includes, for example, a public line network, the Internet (IP telephone network), and WiFi (registered trademark).

[0014] (Hardware configuration) 2 is a diagram showing an example of the hardware configuration of the voice call server according to the present embodiment. The voice call server 10 includes a central processing unit (CPU) 11, a read only memory (ROM) 12, a random access memory (RAM) 13, a hard disk drive (HDD) 14, and a communication device 15.

[0015] The CPU 11 executes various programs and performs arithmetic processing. The ROM 12 stores programs required at startup. The RAM 13 is a work area for temporarily storing the processing by the CPU 11 and storing data. The HDD 14 stores various data and programs. The communication device 15 communicates with other devices via the network 60.

[0016] (Software configuration) FIG. 3 is a diagram showing an example of the software configuration of the voice call system according to the present embodiment.

[0017] The voice call server 10 has a call receiving unit 101, a prompt input unit 102, a response acquisition unit 103, a text-to-speech conversion unit 104, and a call voice transmission / reception unit 105 as main functional units.

[0018] The call receiving unit 101 has a function of receiving a call from a user's telephone and making a call to the user's telephone.

[0019] The prompt input unit 102 has a function of inputting a goal-seek prompt to the language model, and also has a function of inputting a prompt including a text (user's answer sentence) to the language model.

[0020] The answer acquisition unit 103 has a function of acquiring an answer including a question sentence (AI question sentence) to the user from a language model to which a goal seek prompt is input.

[0021] The text-to-speech conversion unit 104 has a function of converting text into telephone voice or telephone voice into text.

[0022] The call voice transmitting / receiving unit 105 has a function of transmitting the call voice (AI question voice) to the telephone and receiving the call voice (user answer voice) from the telephone.

[0023] Each functional unit is realized by a program executed on hardware resources such as a CPU, ROM, and RAM of a computer constituting the voice call server 10. These functional units may be interpreted as "means," "module," "unit," or "circuit." Each functional unit may also be located in a separate device on the network 60. The computer program of the voice call server 10 may be stored in a computer-readable storage medium.

[0024] <System flow> FIG. 4 is a sequence diagram of the voice call system according to this embodiment. FIG. 5 is a diagram showing an example (part 1) of a text dialogue between the voice call server and the LLM according to this embodiment. FIG. 6 is a diagram showing an example (part 1) of a voice call between a user and the voice call server according to this embodiment. Below, a detailed explanation will be given with reference to FIG. 4 and, where appropriate, to FIG. 5 and FIG. 6, taking an example where a user makes a telephone inquiry to a call center about a malfunction of a printer. Note that with regard to the voice call server 10, each step (hereinafter abbreviated as "S") can be realized by having the CPU 11 load and execute a program capable of realizing this sequence.

[0025] S1: A user who is a caller uses the telephone 30 to call a predetermined telephone number (for example, the telephone number of a call center).

[0026] S2: The voice communication server 10 receives a call from the user and establishes a voice call, and inputs (transmits) to the LLM 20 a predetermined goal seek prompt 70a for causing the LLM 20 to respond to the user's inquiry / requirement.

[0027] For example, as shown in Fig. 5, the goal-seek prompt 70a is a type of prompt that issues instructions to the LLM 20, and can guide the user to achieve a goal that the user wants to achieve, such as resolving a request that the user wants to inquire about, by repeatedly asking the user questions (sometimes the LLM 20 creates the prompts itself) so that the user can be guided to achieve the goal. In other words, the goal-seek prompt 70a is a prompt that sets a goal to be achieved and issues an instruction to the LLM 20 to guide the user to that goal. Note that in the goal-seek prompt, the goal to be achieved is set, but the intermediate scenario (guidance) to the goal is not set.

[0028] When a goal-seek prompt 70a is input, the LLM 20 responds with a question (sentence) to guide the user to the goal as a response to the input goal-seek prompt. The user responds to the question, and this process is repeated until the user's purpose is finally achieved. Normally, when a user tries to reach a goal that he or she wants to achieve, it requires a considerable amount of work to create a prompt each time, but the goal-seek prompt does not require the user to think up a new prompt each time. By answering the questions output by the LLM 20 in the goal-seek prompt, it is possible to efficiently move toward the goal.

[0029] S3: When the goal seek prompt 70a is input, the LLM 20 immediately returns an answer including a question sentence (so to speak, an AI question sentence) 71a as an answer to the input goal seek prompt. In this embodiment, the answer to the goal seek prompt 70a includes, for example, a question sentence 71a for the user, such as "This is the XX call center. Please tell us the requirements of your inquiry." Furthermore, when the goal seek prompt instructs the LLM 20 to create a prompt, the answer may include a prompt generated by the LLM 20 (a prompt including a user answer sentence 72a in S8 described later or an example or template of the prompt).

[0030] S4: Since the answer question 71a received from the LLM 20 is text data, the voice call server 10 converts the text data of the question 71a into call voice data (so to speak, an AI question voice).

[0031] For example, the LLM 20 may give an unexpected or inappropriate answer, in which case the voice call server 10 may convert the text data of the question 71a into call voice data after modifying (processing) it. For example, even if a character limit for the answer is instructed with a prompt to answer within 100 characters, if the answer is 200 characters, the answer may be modified to within 100 characters. In this case, the voice call server 10 may have the LLM 20 modify or summarize the answer to within 100 characters again, and then convert the text data of the question 71a into call voice data.

[0032] S5: The voice call server 10 responds with the converted call voice data (AI question voice) 71b to the telephone 30. The call voice data 71b is a voice indicating an AI question, such as "This is the XX call center. Please tell us the details of your inquiry." Here, the voice data 71b asks the user the specific details of the inquiry and prompts the user to make the call (utterance).

[0033] S6: The user receives the AI ​​question voice 71b and freely speaks the requirements he / she wants to inquire about as the user answer voice 72b, for example, "I have a question about the product."

[0034] S7: The voice call server 10 receives the user's answer voice 72b to the AI ​​question voice 71b of S5 as call voice data, and converts the received call voice data into text data (user's answer sentence).

[0035] In addition, for example, the user may give an unexpected or inappropriate answer, and in that case, the voice call server 10 may convert the user's answer voice 72b into text data after correcting (processing) it. Alternatively, the voice call server 10 may prompt the user to make a correction again, and convert the user's answer voice 72b into text data after making the correction.

[0036] S8: The voice communication server 10 generates a prompt including the user answer sentence 72a (a prompt based on the user answer sentence 72a) as a response from the user to the AI ​​voice question 71b of S5, and inputs the generated prompt to the LLM 20.

[0037] S9: The LLM 20 receives the prompt including the user answer sentence 72a of S8 and immediately replies with an answer including a question sentence 73a. For example, the answer includes a question sentence 73a for the user, such as "This is the XX call center. Please tell us the requirements of your inquiry."

[0038] S10: The voice call server 10 converts the text data of the answer question sentence (AI question sentence) 73a received from the LLM 20 into call voice data (AI question voice).

[0039] S11: The voice call server 10 responds with the converted call voice data (AI question voice) 73b to the telephone 30. The call voice data 73b is, for example, an AI question voice saying, "Thank you for speaking. So, which product specifically are you asking about?", and prompts the user to make a further call (utterance) for a response.

[0040] Then, the exchange of the AI ​​question voice (AI question sentence) and the user answer voice (user answer sentence) is repeated until the user's purpose is achieved (S6 to S11). FIG. 7 is a diagram showing an example (part 2) of a text dialogue between the voice call server and the LLM according to this embodiment. FIG. 8 is a diagram showing an example (part 2) of a voice call between the user and the voice call server according to this embodiment. The LLM 20 according to this embodiment has studied, for example, a printer service manual, and is therefore capable of identifying the user's printer malfunction and providing a solution in accordance with the service manual. In addition, sometimes user answer voices vary from person to person and include unexpected answers, but the LLM 20 can flexibly respond with a question voice that corresponds to the user answer voice each time.

[0041] The voice call server 10 can also cooperate with various systems and DBs (FIG. 1). For example, it is possible to inquire the telephone number of the user who made the call in a customer management DB, identify the user information, and use the user information for the voice call. Also, for example, if it is necessary to identify past response history, it is possible to cooperate with a response history system or the like and use the past response history for the voice call.

[0042] [Embodiment 2] A description will be given of embodiment 2. In embodiment 1, an example of an inquiry response based on a call from a user (inbound response) was shown (S1 in FIG. 4). In embodiment 2, an example of a guidance response / sales response based on a call from the voice call server 10 to the user (outbound response) is shown.

[0043] FIG. 9 is a sequence diagram of the voice call system according to this embodiment. FIG. 10 is a diagram showing an example (part 3) of a text dialogue between the voice call server and the LLM according to this embodiment. FIG. 11 is a diagram showing an example (part 4) of a text dialogue between the voice call server and the LLM according to this embodiment. FIG. 12 is a diagram showing an example (part 3) of a voice call between a user and the voice call server according to this embodiment. FIG. 13 is a diagram showing an example (part 4) of a voice call between a user and the voice call server according to this embodiment. An example of providing sales guidance for an insurance product to a user will be described in detail with reference to FIG. 9 and, where appropriate, with reference to FIGS. 10-13.

[0044] S1-2: The voice communication server 10, which corresponds to the caller, calls the user, i.e., a predetermined telephone number (e.g., a telephone number listed in a customer list). Compared with S1 in FIG. 4, in S1, S2: When the user answers the phone, the voice communication server 10 establishes a voice call and inputs a predetermined goal seek prompt 80a to the LLM 20 so that the LLM 20 responds to inquiries and requirements from the user.

[0045] When the goal-seek prompt 80a is input, the S3:LLM 20 immediately returns an answer including a question (so to speak, an AI question) 81a as a response to the input goal-seek prompt. The answer includes a question 71a for the user, such as "This is the XX call center. Please tell us the requirements of your inquiry."

[0046] S4: The voice communication server 10 converts the text data of the answer question sentence 81a received from the LLM 20 into communication voice data (AI question voice).

[0047] S5: The voice call server 10 responds to the telephone set 30 with the converted call voice data (AI question voice) 81b.

[0048] S6: The user receives the AI ​​question voice 81b and freely speaks as the user answer voice 82b.

[0049] S7: The voice call server 10 receives the user's answer voice 82b to the AI ​​question voice 81b of S5 as call voice data, and converts the received call voice data into text data (user's answer sentence).

[0050] S8: The voice communication server 10 generates a prompt including the user answer sentence 82a (a prompt based on the user answer sentence 82a) as a response from the user to the AI ​​voice question 81b of S5, and inputs the generated prompt to the LLM 20.

[0051] S9: The LLM 20 receives the prompt including the user answer sentence 82a of S8 and immediately replies with a question sentence 83a.

[0052] S10: The voice call server 10 converts the text data of the answer question sentence (AI question sentence) 83a received from the LLM 20 into call voice data (AI question voice).

[0053] S11: The voice call server 10 responds to the telephone set 30 with the converted call voice data (AI question voice) 83b.

[0054] Then, the AI ​​question voice (AI question sentence) and the user answer voice (user answer sentence) are exchanged repeatedly until the user's goal is achieved (S6 to S11). Since the LLM 20 according to this embodiment has learned, for example, the insurance product pamphlet or sales manual, it is capable of outputting information related to the insurance product.

[0055] [Variations] Fig. 14 is a sequence diagram of the voice call system according to this modified example. Fig. 15 is a diagram showing an example (part 5) of a text dialogue between the voice call server and LLM according to this modified example. Fig. 15 is a diagram showing an example (part 5) of a text dialogue between the voice call server and LLM according to this modified example. A specific inquiry or requirement from the user may be explicitly set as a goal in the goal seek prompt as a goal that the user wishes to achieve.

[0056] S1: The user uses the telephone 30 to call a predetermined telephone number (eg, the telephone number of a call center).

[0057] S1-3: The voice call server 10 receives a call from the user, establishes a voice call, and responds with guidance to solicit the user's inquiry or requirements, such as, for example, "This is the XX call center. Please tell us the details of your inquiry."

[0058] S1-4: The user receives the guidance and freely talks about the requirements he / she wants to inquire about, for example, "I have a question about a product."

[0059] S1-5: The voice communication server 10 executes a voice analysis process and extracts keywords to identify the requirement that the user wishes to inquire about (identification means for identifying the user's desired purpose).

[0060] S2: The voice communication server 10 inputs a goal seek prompt 90a, in which the requirement that the user wishes to inquire about is set as the goal, to the LLM 20. In comparison with the goal seek prompt 70a in Fig. 5, the goal seek prompt 90a in Fig. 15 has a concrete goal set to the requirement that the user wishes to inquire about, such as "First, I would like to ask about a product. Please create a prompt with the goal seek prompt, ask me polite questions, and guide me to the achievement of my goal so that I can achieve my goal...". This can improve the possibility of the user's goal (for example, solving a request to inquire about) being achieved.

[0061] <Summary> According to the voice call system 100 of this embodiment, a manned operator is not required, so the number of users that can be handled simultaneously can be increased dramatically. In addition, because of the free voice call (answer) by the LLM (AI), a response voice (voice guidance) that was previously assumed is not required, and the response voice is not limited to a fixed and standardized requirement response, so that more flexible response voice guidance is possible until the goal is achieved according to the user's response. In addition, the range and versatility of the application fields and applications of the voice response system can be improved.

[0062] Although the present invention has been described by showing specific examples according to the preferred embodiment of the present invention, it is apparent that various modifications and changes can be made to these examples without departing from the broad spirit and scope of the present invention as defined in the claims. In other words, the details of the examples and the accompanying drawings should not be construed as limiting the present invention.

[0063] The voice call system according to this embodiment can be used for inbound correspondence based on calls from users and outbound correspondence based on calls to users, and can be used in, for example, various consultation and inquiry telephone helplines, procedures, product reservations, product purchase applications, product orders, complaint handling telephone helplines, guidance (sales), telephone research / surveys, etc. Multilingual support is also possible. [Explanation of symbols]

[0064] 10 Voice call server 20 LLM 30 telephone 60 Network 100 Voice Communication System 101 Telephone Receiving Unit 102 Prompt input section 103 Answer acquisition part 104 Text to speech conversion unit 105 Voice transmission and reception unit

Claims

1. A voice communication device that supports a caller in achieving a purpose through a voice call with a user, a first prompt input means for inputting a prompt to a language model to guide the user to achieve the objective; an answer acquisition means for acquiring an answer including a question sentence for the user from the language model to which the prompt is input; a speech conversion means for converting a first text of the question sentence into a first call speech; a call voice transmission means for transmitting the first call voice to the user's telephone; a call voice receiving means for receiving a second call voice in response to the first call voice from the telephone; a text conversion means for converting the second call voice into a second text; a second prompt input means for inputting a prompt including the second text to the language model; having The prompt includes a character limit instruction for an answer including a question for the user; If the question exceeds the character limit, modifying the question to fit within the character limit; A voice communication device comprising:

2. the prompt input by said first prompt input means is a goal seek prompt; 2. The voice communication device according to claim 1,

3. A means for identifying the purpose desired by the user through a voice analysis process via a voice call with the user; 3. The voice communication device according to claim 1, further comprising:

4. A computer that assists a caller in achieving his or her objectives through a voice call with a user, a first prompt input means for inputting a prompt to a language model to guide the user to achieve the objective; an answer acquisition means for acquiring an answer including a question sentence for the user from the language model to which the prompt is input; a speech conversion means for converting a first text of the question sentence into a first call speech; a call voice transmission means for transmitting the first call voice to the user's telephone; a call voice receiving means for receiving a second call voice in response to the first call voice from the telephone; a text conversion means for converting the second call voice into a second text; a second prompt input means for inputting a prompt including the second text to the language model; and make it work. The prompt includes a character limit instruction for an answer including a question for the user; If the question exceeds the character limit, the question is revised to fall within the character limit. Voice call program.

Images