Information processing device, information processing method, and storage medium

Abstract

This information processing device divides utterance data, which includes time-series utterance contents by a plurality of speakers, into a plurality of pieces on the basis of clock time to generate divided utterance data. The information processing device generates prompt information including a directive to be stored in advance that instructs determination of establishment of a conversation for each combination of the plurality of speakers on the basis of the utterance content, and a piece of divided utterance data selected from the divided utterance data. The information processing device calculates a degree of establishment of conversation for each combination of the plurality of speakers, on the basis of the result of the determination of establishment of the conversation for each combination of the plurality of speakers output from a large-scale language model using the prompt information.

Description

Information processing device, information processing method, and storage medium

[0001] The present disclosure relates to an information processing device, an information processing method, and a storage medium.

[0002] There is a technology that calculates the degree of successful conversation for a combination of speakers by analyzing speech intervals based on speech data (audio data, scripts, etc.) containing time-series speech content from multiple speakers. Patent Document 1 discloses a technology that separates audio signals for each sound source in an environment where multiple voices are mixed, calculates the degree of successful conversation with a received signal, and extracts effective speech in which a conversation is established based on the calculation results. Specifically, as shown in FIG. 19 , the conversation analysis device described in Patent Document 1 includes a speech separation unit that separates multiple voices for each speaker, a conversation degree evaluation unit that calculates the degree of successful conversation for a combination of speakers, and a conversational voice combination unit that combines and outputs speech in which a conversation is established based on the results. The conversation degree evaluation unit includes a speech interval detection unit that detects speech intervals for each speaker from each separated voice, and a section analysis unit that calculates the degree of successful conversation from the results. The section analysis unit focuses only on the voices of the two speakers and calculates, as time-series data, sections in which both speakers are speaking, sections in which neither speaker is speaking, and sections in which only one speaker is speaking. Considering that conversation between two speakers is often conducted by alternating speech, the degree of conversation establishment is calculated from these time series data.

[0003] Japanese Patent Application Laid-Open No. 2004-133403

[0004] In the technology described above, it has been desired to calculate an evaluation result that can more accurately classify conversations into groups based on data of a plurality of voices.

[0005] An object of the present invention is to provide an information processing device, an information processing method, and a storage medium that solve the above-mentioned problems.

[0006] An information processing device according to one aspect of the present disclosure includes a division means for dividing speech data containing chronological speech content of multiple speakers into multiple pieces based on time to generate divided speech data; a generation means for generating prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a judgment on the success of a conversation for each combination of the multiple speakers based on the speech content, and selected divided speech data from the divided speech data; and a calculation means for calculating the degree of success of a conversation for each combination of the multiple speakers based on the results of the judgment on the success of a conversation for each combination of the multiple speakers output by the large-scale language model using the prompt information.

[0007] An information processing system according to one aspect of the present disclosure includes a division means for dividing speech data containing chronological speech content of multiple speakers into multiple pieces based on time to generate divided speech data; a generation means for generating prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a judgment on the success of a conversation for each combination of the multiple speakers based on the speech content, and selected divided speech data from the divided speech data; and a calculation means for calculating the degree of success of a conversation for each combination of the multiple speakers based on the results of the judgment on the success of a conversation for each combination of the multiple speakers output by the large-scale language model using the prompt information.

[0008] An information processing method according to one aspect of the present disclosure divides speech data containing chronological speech content of multiple speakers into multiple parts based on time to generate divided speech data, generates prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a determination of the success of a conversation for each combination of the multiple speakers based on the speech content, and selected divided speech data from the divided speech data, and calculates the degree of success of a conversation for each combination of the multiple speakers based on the results of the determination of the success of a conversation for each combination of the multiple speakers output by the large-scale language model using the prompt information.

[0009] A storage medium according to one aspect of the present disclosure stores a program that causes a computer of an information processing device to function as: a division means that divides speech data containing chronological speech content of multiple speakers into multiple parts based on time to generate divided speech data; a generation means that generates prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a determination of the success of a conversation for each combination of the multiple speakers based on the speech content, and selected divided speech data from the divided speech data; and a calculation means that calculates the degree of success of a conversation for each combination of the multiple speakers based on the results of the determination of the success of a conversation for each combination of the multiple speakers output by the large-scale language model using the prompt information.

[0010] According to the present invention, it is possible to calculate an evaluation result that allows for more accurate grouping of conversations based on data of a plurality of voices.

[0011] 1 is a first diagram showing functional blocks of an information processing device of the present disclosure. FIG. 2 is a diagram showing a hardware configuration of a computer of the information processing device of the present disclosure. FIG. 3 is a first diagram showing an example of utterance data of the present disclosure. FIG. 4 is a second diagram showing an example of utterance data of the present disclosure. FIG. 5 is a first diagram showing a prompt creation process of the present disclosure. FIG. 6 is a diagram showing a portion of prompt information generated by the present disclosure. FIG. 7 is a diagram showing an example of generated data of the present disclosure. FIG. 8 is a first diagram showing an example of an analysis result of generated data of the present disclosure. FIG. 9 is a second diagram showing an example of an analysis result of generated data of the present disclosure. FIG. 10 is a second diagram showing functional blocks of an information processing device of the present disclosure. FIG. 11 is a second diagram showing a processing flow of an information processing device of the present disclosure. FIG. 12 is a third diagram showing functional blocks of an information processing device of the present disclosure. FIG. 13 is a flowchart showing the operation of a conversation level integration unit of the present disclosure. FIG. 14 is a diagram showing the transition of data generated in the processing of the conversation level integration unit. FIG. 15 is a fourth diagram showing functional blocks of an information processing device of the present disclosure. FIG. 16 is a diagram showing functional blocks of an information processing device with another configuration of the present disclosure. FIG. 17 is a diagram showing a processing flow of an information processing device with another configuration of the present disclosure. FIG. 18 is a diagram showing the functional configuration of a conversation analysis device in a related technology.

[0012] Each embodiment will be described below with reference to the drawings.

[0013] [First embodiment] [Configuration description] Fig. 1 is a first diagram showing functional blocks of an information processing device of the present disclosure. As shown in Fig. 1, the information processing device 1 of the present disclosure includes a conversation level evaluation unit 10. The conversation level evaluation unit 10 includes a division unit 11, a prompt generation unit 12, and a conversation analysis unit 13.

[0014] FIG. 2 is a diagram illustrating the hardware configuration of a computer of an information processing device according to the present disclosure. As illustrated in this diagram, the information processing device 1 is a computer equipped with various hardware components, such as a central processing unit (CPU) 101, a read-only memory (ROM) 102, a random access memory (RAM) 103, a storage device 104, a communication module 105, and a sensor 106. The CPU 101 of the information processing device 1 executes a predetermined program to perform the functions of each processing unit shown in FIG. 1 . The information processing device 1 may also include other hardware components, such as a graphics processing unit (GPU). If the information processing device 1 includes a GPU, the functions of each processing unit shown in FIG. 1 may be performed by processing by the GPU. In another disclosure, multiple information processing devices 1 may be connected via a communication network to form an information processing system 100 having functions corresponding to the above-described processing units. In this case, each information processing device constituting the information processing system 100 may execute a predetermined program to perform the same functions as the information processing device 1 according to the present disclosure.

[0015] The division unit 11 divides utterance data including the time-series utterance content of the speakers into multiple pieces to generate divided utterance data. The prompt generation unit 12 generates prompt information including at least a pre-stored instruction statement that instructs determining whether a conversation has been established for each combination of multiple speakers based on the utterance content, and one piece of divided utterance data for each speaker that corresponds to a time period among the divided utterance data of each speaker. The conversation analysis unit 13 uses the prompt information to obtain a result of determining whether a conversation has been established for each combination of multiple speakers output by a large-scale language model (LLM), and calculates the degree of conversation establishment for each combination of multiple speakers.

[0016] FIG. 3 is a first diagram showing an example of speech data of the present disclosure. As shown in FIG. 3, the speech data includes speech section information (speech start time, speech end time) and a plurality of invention contents uttered by speakers between the speech start time and speech end time indicated by the speech section information, arranged in chronological order. FIG. 3 shows speech data indicating the contents of speeches made by two speakers, Speaker A (left in FIG. 3) and Speaker B (right in FIG. 3). Each line in the speech data indicates the speech start time, speech end time, and sentence of the speech contents, arranged from top to bottom in chronological order.

[0017] FIG. 4 is a second diagram illustrating an example of speech data according to the present disclosure. In the information processing device 1 according to the present disclosure, the division unit 11 divides the acquired speech data of each of the multiple speakers at predetermined time intervals. FIG. 4 illustrates divided speech data obtained by dividing the speech data of FIG. 3 into multiple pieces. For example, the division unit 11 identifies each piece of speech content included in a one-minute time interval from the start time of the first utterance in the acquired speech data as the first piece of divided speech data. The division unit 11 also identifies each piece of speech content included in a one-minute time interval from 30 seconds after the start time of the first utterance in the first piece of divided speech data as the second piece of divided speech data. In this manner, the division unit 11 may generate divided speech data by dividing the acquired speech data in sequence while partially overlapping the speech content included in each piece of divided speech data before and after the time series. The division unit 11 similarly generates divided speech data based on the speech data of each speaker. For example, assume that the acquired speech data includes 10 minutes of speech content. In this case, the division unit 11 may generate and output divided utterance data divided into one-minute segments without overlapping the utterance content. Note that the time from the start time of the first utterance to the end time of the last utterance included in each divided utterance data does not need to be the same for all divided utterance data. The above-described method of dividing the utterance data by the division unit 11 is one example, and divided utterance data may be divided into multiple segments in time series using other methods to generate divided utterance data.

[0018] [Explanation of Operation] Fig. 5 is a first diagram showing a processing flow of the information processing device of the present disclosure. Next, the processing flow of the information processing device 1 of the present disclosure will be described in order. First, the division unit 11 receives speech data of each of multiple speakers from the outside (step S101). The speech data may be, for example, multiple pieces of speech data generated externally based on the speech contents of the multiple speakers contained in audio data, script data, etc., so as to include the content of the speech of each of the multiple speakers. Script data is an example of text information.

[0019] 4, the division unit 11 divides the acquired utterance data at predetermined time intervals (step S102). The division unit 11 outputs the divided utterance data of each speaker to the prompt generation unit 12. The prompt generation unit 12 receives the divided utterance data for all speakers from the division unit 11.

[0020] 6 is a first diagram illustrating the prompt creation process. The prompt generation unit 12 acquires a total of m × 4 pieces of divided speech data from the segmentation unit 11, each of which contains the speech content of four speakers, A, B, C, and D, divided into m pieces. The prompt generation unit 12 counts the number of speakers, n, based on the number of speakers indicated by the acquired divided speech data (step S103). Since there are four speakers, A, B, C, and D, the number of speakers, n = 4.

[0021] The prompt generation unit 12 reads from the storage unit a template for an instruction statement to be included in the prompt. The template includes at least a first instruction and a second instruction. The first instruction is a sentence that instructs the LLM on what data generation process to perform. The second instruction is a sentence that instructs the LLM on the output format of the combination of divided utterance data used to determine the degree of successful conversation. The prompt generation unit 12 embeds divided utterance data identifiers, such as file names of the divided utterance data for each speaker (n = 4 speakers), in the characters "text_list" included in the first instruction of the read instruction statement (step S104). The prompt generation unit 12 also identifies all combinations of speakers (divided utterance data) for which the degree of successful conversation is to be determined, and embeds the output format of each combination as a second instruction (step S105). Because the number of speakers is n = 4, there are 11 combinations of divided utterance data corresponding to these speakers: AB, AC, AD, BC, BD, CD, ABC, ABD, ACD, BCD, and ABCD. Therefore, the instruction sentence of the combination indicated by the second instruction indicates the output format of the judgment result of the degree of success of the conversation for each of these 11 combinations.

[0022] 7 illustrates a portion of the generated prompt information. The prompt generation unit 12 generates prompt information including at least a first instruction 71, a second instruction 72, and a plurality of divided utterance data 74 to 77 selected based on the utterance start times of each speaker (step S106). That is, the prompt generation unit 12 generates a first prompt including a standard first instruction, a second instruction, a first piece of divided utterance data (TextA) 74 including the utterance content of speaker A that started earlier, a first piece of divided utterance data (TextB) 75 including the utterance content of speaker B that started earlier, a first piece of divided utterance data (TextC) 76 including the utterance content of speaker C that started earlier, and a first piece of divided utterance data (TextD) 77 including the utterance content of speaker D that started earlier. The prompt generator 12 may include in the prompt information a third instruction 73 that instructs the user to answer why the conversation between the pair of speakers specified in the second instruction 72 is successful.

[0023] The prompt generation unit 12 similarly generates prompt information including the divided utterance data of each speaker whose first utterance included in the divided utterance data has the next earliest utterance start time. The prompt generation unit 12 also generates prompt information repeatedly in the same manner, starting from the earliest utterance start time of the first utterance included in the divided utterance data. In other words, multiple pieces of prompt information are generated in the order of earliest utterance start time of the utterances included in the divided utterance data, such as first prompt information, second prompt information, and so on. For example, if the utterance data of each speaker is divided into ten parts, the prompt generation unit 12 sequentially generates ten pieces of prompt information including divided utterance data selected one for each speaker based on their utterance start time.

[0024] The prompt generator 12 sequentially inputs the generated prompt information to a computer such as a cloud server that performs generation processing of generated data using the LLM (step S107). As a result, the information processing device 1 acquires generated data corresponding to the prompt information from the computer (step S108). The LLM may be any type that generates generated data based on a directive using known technology.

[0025] FIG. 8 is a diagram illustrating an example of generated data. FIG. 9 is a first diagram illustrating an example of an analysis result of generated data. The information processing device 1 analyzes generated data acquired from a computer running an LLM. The generated data corresponding to one piece of prompt information includes, for example, information indicating a combination of segmented utterance data that constitutes a conversation. This combination indicates a combination of speakers estimated to have a conversation. Furthermore, if the prompt information includes a third instruction, the generated data may include a sentence indicating why a conversation based on the third instruction is established. In the example of FIG. 8 , responses indicating that a conversation is established between segmented utterance data including the utterance content of speaker A and segmented utterance data including the utterance content of speaker B, and a conversation is established between segmented utterance data including the utterance content of speaker C and segmented utterance data including the utterance content of speaker D, are included. Furthermore, the example of FIG. 8 also includes a reason why a conversation is established between the utterance content of these combinations. The conversation analysis unit 13 acquires and analyzes the generated data shown in FIG. 8 (step S109). Based on the generated data shown in FIG. 8, the conversation analysis unit 13 generates an analysis result (FIG. 9) for each combination of divided utterance data, indicating a value of "1" if a conversation is established and a value of "0" if the conversation is not determined to be established (step S110). The generated data and the analysis result may be data in a CSV file format. The LLM generates generated data that indicates the degree of establishment of the conversation with two values: "0" if the conversation is not determined to be established and "1" if the conversation is determined to be established. Depending on the form of the first instruction included in the prompt information, the LLM may generate generated data that indicates the degree of establishment of the conversation with a numerical value of two or more values.

[0026] FIG. 10 is a second diagram showing an example of the analysis results of generated data. In the above-described process, if the speech data of each speaker is divided into 10 pieces, the information processing device 1 inputs the generated 10 pieces of prompt information, in the order of first prompt information, second prompt information, third prompt information, ..., and tenth prompt information, into the LLM, and sequentially acquires 10 pieces of generated data corresponding to these 10 pieces of prompt information from the computer running the LLM. The conversation analysis unit 13 calculates the number of times the conversation was determined to be valid by adding up the number of "1"s assigned when the conversation was analyzed as valid based on each piece of prompt information. For example, FIG. 10 shows that generated data was acquired in which the conversation between the divided speech data containing the speech content of speaker A and the divided speech data containing the speech content of speaker B was determined to be valid twice. 10 shows that generated data was acquired in which it was determined six times that a conversation had been established between the utterance contents of divided utterance data including the utterance content of speaker A, divided utterance data including the utterance content of speaker B, and divided utterance data including the utterance content of speaker C. Similarly, the analysis result shown in FIG. 10 shows that it was acquired in which it was determined once that a conversation had been established based on divided utterance data including the utterance content of speakers C and D, once that a conversation had been established based on each of the divided utterance data including the utterance content of each of speakers A, B, and D, and once that a conversation had been established based on each of the divided utterance data including the utterance content of each of speakers A, B, C, and D.

[0027] The conversation analysis unit 13 calculates the degree of conversation establishment for each speaker combination based on the number of times it is determined that a conversation is established. For example, the conversation analysis unit 13 calculates the ratio of the number of times it is determined that a conversation is established for the speaker combination being calculated to the number m of divisions of the utterance data (= the number of generated prompts = 10) as the degree of conversation establishment for the speaker combination. In this case, the conversation analysis unit 13 calculates the degree of conversation establishment for the combination of speakers A and B as "0.2," the degree of conversation establishment for the combination of speakers C and D as "0.1," the degree of conversation establishment for the combination of speakers A, B, and C as "0.6," the degree of conversation establishment for the combination of speakers A, B, and D as "0.1," and the degree of conversation establishment for the combination of speakers A, B, C, and D as "0.1." The conversation analysis unit 13 outputs the calculated degree of conversation establishment for each speaker combination.

[0028] Based on the calculated conversation establishment degree and a threshold for determining whether a conversation has been established, the conversation analysis unit 13 may output a value of "1" indicating a high possibility that a conversation has been established if the conversation establishment degree is equal to or greater than the threshold, or "0" indicating a low possibility that a conversation has been established if the conversation establishment degree is less than the threshold. The output processing unit or device may determine which speakers belong to which group of speakers are conversing based on the conversation establishment degree. For example, in the analysis result shown in FIG. 10 , if the threshold is "0.2," it can be determined that the combination of speakers A and B (conversation establishment degree 0.2) and the combination of speakers A, B, and C (conversation establishment degree 0.6) are conversing groups. Note that, in this way, if the group of speakers A and B and the group of speakers A, B, and C contain overlapping speakers, the group of speakers A, B, and C with the highest conversation establishment degree value may be determined to be the conversing group.

[0029] [Explanation of Effects] According to the above-described process, the information processing device 1 in the first embodiment acquires generated data including whether a conversation has been established for each combination of speakers generated by the LLM based on the speech start time, speech end time, and speech content, and calculates the degree of conversation establishment based on the generated data. Also, as shown in Figure 7, an example is handled in which the timing (speech start time, speech end time) of the first utterances of TextA and TextC, and TextB and TextD, are identical for each of the utterances of four speakers (TextA, TextB, TextC, TextD). Furthermore, in the first embodiment, a conversation has been established between TextA and TextB, and between TextC and TextD, and the conversations can be grouped into two groups. Here, the method of Patent Document 1, which calculates the degree of successful conversation using only speech section information, does not use the content of the speech, and therefore is unable to distinguish between Text A and Text C, or Text B and Text D, i.e., between speaker A and speaker C, or speaker B and speaker D, and is therefore unable to perform appropriate grouping. On the other hand, the information processing device 1 in the first embodiment specifies the content of each speaker's speech in the prompt, instructs the LLM to take into account the content of each speaker's speech when determining whether or not a conversation has been established, and calculates the degree of successful conversation based on the result of the determination of whether or not a conversation has been established, thereby making it possible to more accurately determine whether or not a conversation has been established between speakers.

[0030] Furthermore, in the information processing device 1 according to the first embodiment, the conversation level evaluation unit 10 performs a process of dividing each speaker's utterance content based on the duration of the utterance to generate divided utterance data, and then performs conversation analysis using LLM multiple times for each of the divided utterance data. This process enables more accurate grouping. For example, in audio at an event venue, multiple speakers may discuss the same topic during a certain time period, regardless of whether a conversation is established or not. If conversation analysis is performed without performing the division process, there is a possibility that the influence of the time period during which multiple speakers discuss the same topic may lead to the incorrect determination that all speakers belong to the same conversation group. In the first embodiment, the division process for each speaker's utterance content creates multiple divided sections that are affected by the time period during which multiple speakers discuss the same topic and multiple divided sections that are not affected by the time period. If the time period during which multiple speakers discuss the same topic is sufficiently short compared to the overall duration of the utterance content included in the utterance data, the number of unaffected sections will increase. By calculating the determination probability based on whether a conversation is established for each such time period and determining whether a section belongs to a conversation group, appropriate grouping is possible.

[0031] In the above process, multiple pieces of prompt information are sequentially input into one LLM, and the resulting generated data is used to calculate the degree of successful conversation for each speaker combination. However, one piece of prompt information may be sequentially input into multiple different LLMs, and the process may be repeated for each piece of prompt information. The generated data obtained from each different LLM may be used to calculate the degree of successful conversation for each speaker combination. In this case, each LLM is implemented on a different computer, and the information processing device 1 is communicatively connected to each computer. This process allows the degree of successful conversation to be calculated based on the generated data output by multiple different LLMs, and an evaluation result can be calculated that can more accurately classify conversation groups from data on multiple conversation groups of multiple people.

[0032] Furthermore, in the above process, if the time from the start time to the end time of the entire utterance content included in the utterance data of each speaker is short, one utterance data may be copied for each speaker to generate multiple identical utterance data for each speaker, and the prompt information corresponding to each copy may be input sequentially to different LLMs. This allows the degree of successful conversation to be calculated based on the generated data output from multiple different LLMs, and even if the duration of the utterance content is short, it is possible to calculate an evaluation result that can more accurately group the conversations into groups based on the data of multiple conversation groups of multiple people.

[0033] [Second Embodiment] [Configuration Description] Fig. 11 is a second diagram showing functional blocks of an information processing device of the present disclosure. Fig. 12 is a second diagram showing a processing flow of the information processing device of the present disclosure. As shown in Fig. 11, the information processing device 1 of the present disclosure may further include the functions of a speech recognition unit 20 in addition to a conversation level evaluation unit 10. The conversation level evaluation unit 10 is the same as in the first embodiment. Functions that are the same as in the first embodiment are assigned the same reference numerals as in Fig. 1, and detailed description will be omitted. The speech recognition unit 20 inputs speech data for each speaker from an external device and generates speech data by extracting the speech content, speech start time, and speech end time included in the speech data.

[0034] [Explanation of Operation] The speech recognition unit 20 acquires speech data for each speaker (step S201). The speech recognition unit 2 inputs the speech data and detects the speech start time and speech end time in the speech data (step S202). The speech recognition unit 2 generates and outputs speech data including speech section information (speech start time, speech end time) and a list of the speech content in that section (step S203). This allows, when there are multiple speakers, to transfer to the conversation level evaluation unit 10 a file (speech data) listing the speech content independently for each speaker. Figure 3 can also be seen as the speech recognition result of a conversation between two speakers (speaker A and speaker B). The subsequent processing of the information processing device 1 is the same as in the first embodiment.

[0035] [Explanation of Effects] The information processing device 1 in the second embodiment extracts the speech start time, speech end time, and speech content required for input to the conversation level evaluation unit 10 by the speech recognition unit 2. In most cases, conversations are made up of verbal exchanges, and by directly inputting the speech, it is possible to save the effort and time required to create a transcription.

[0036] [Third Embodiment] [Configuration Description] Fig. 13 is a third diagram showing functional blocks of an information processing device of the present disclosure. As shown in Fig. 13, the information processing device 1 of the present disclosure differs from the second embodiment in the conversation level evaluation unit 10, but the other components are the same as those of the second embodiment. Components that are the same as those of the second embodiment are assigned the same reference numerals as those in Fig. 9, and detailed description thereof will be omitted.

[0037] The conversation level assessment unit 10 of the third embodiment includes a segmentation unit 11, a prompt generation unit 12, and a conversation analysis unit 13, as well as an utterance period detection unit 14, a section analysis unit 15, and a conversation level integration unit 16. That is, the third embodiment differs from the second embodiment in that it includes the utterance period detection unit 14, the section analysis unit 15, and the conversation level integration unit 16.

[0038] The speech section detection unit 14 detects speech section information indicating at least the speech start time and speech end time of the speech included in the audio data. The section analysis unit 15 calculates a second degree of validity of the conversation using the voiced and silent states of the speech identified based on the speech section information. The conversation degree integration unit 16 defines the degree of validity of the conversation calculated by the conversation analysis unit 13 as a first degree of validity and calculates an integrated degree of validity based on the first degree of validity and the second degree of validity.

[0039] [Explanation of Operation] The speech interval detection unit 14 receives input of voice data from multiple speakers, detects speech intervals (speech start time, speech end time) included in each piece of voice data using, for example, VAD (Voice Activity Detection) technology, and generates a list of speech intervals including information indicating those intervals. The speech interval detection unit 14 may also use other technologies to detect speech intervals. Assume that one input piece of voice data includes voice extracted from only one person's speech. If there are multiple speakers and multiple corresponding pieces of voice data, they are input to the speech interval detection unit 14 one by one. If the voice includes noise, noise suppression processing may be applied in advance.

[0040] The section analysis unit 15 receives the list of speech sections from the speech section detection unit 14 and outputs the degree of successful conversation between two speakers. The difference is that the conversation analysis unit 13 also considers conversations between three or more speakers, while the section analysis unit 15 is limited to conversations between two speakers. For example, the analysis means described in Patent Document 1 focuses on the voice / silence patterns of the conversational voice between two speakers. Specifically, the degree of successful conversation is set to be high when one voice is voiced and the other voice is silent. The section analysis unit 15 of the present disclosure similarly calculates the degree of successful conversation such that the degree of successful conversation is high when one voice is voiced and the other voice is silent. Any calculation process may be used to calculate the degree of successful conversation such that the degree of successful conversation is high when one voice is voiced and the other voice is silent. For example, as in Patent Document 1, the degree of successful conversation may be calculated such that the degree of successful conversation is high when one voice is voiced and the other voice is silent. The content of the conversation is not used in calculating the degree of successful conversation.

[0041] Fig. 14 is a flowchart showing the operation of the conversation level integration unit 16 of the present disclosure. Fig. 15 is a diagram showing the transition of data generated in the processing of the conversation level integration unit 16. The conversation level integration unit 16 receives the degree of conversation establishment between two speakers from the section analysis unit 15 (step S161). The data of the conversation establishment degree received here indicates the value of the degree of conversation establishment between the two speakers, as shown in (A) of Fig. 15. For example, suppose there are four speakers, speakers A, B, C, and D, and based on the voice data of each speaker, the section analysis unit 15 calculates the degree of conversation success between each pair of speakers as follows: the degree of conversation success between speaker A and speaker B is "0.8", the degree of conversation success between speaker A and speaker C is "0.6", the degree of conversation success between speaker A and speaker D is "0.1", the degree of conversation success between speaker B and speaker C is "0.7", the degree of conversation success between speaker B and speaker D is "0.3", and the degree of conversation success between speaker C and speaker D is "0.3".

[0042] In this case, the conversation degree integration unit 16 calculates the conversation establishment degree for all combinations of speakers (FIG. 15(B)) based on the received conversation establishment degree between two speakers (FIG. 15(A)) (step S162). In other words, the conversation degree integration unit 16 cannot calculate the conversation establishment degree for three-party and four-party conversations other than the conversation establishment degree between two speakers. Therefore, the conversation degree integration unit 16 calculates the conversation establishment degree for three-party and four-party conversations based on the conversation establishment degree between two parties calculated by the section analysis unit 15, and outputs the conversation establishment degree for all combinations of speakers.

[0043] Specifically, the conversation establishment degree for a combination of three or more speakers is calculated by averaging the conversation establishment degrees of all pairs of two speakers that make up the combination, as calculated by the section analysis unit 15. For example, the conversation establishment degree integration unit 16 calculates the conversation establishment degree for the combination of speakers A, B, and C as "0.7" by averaging the conversation establishment degree of speakers A and B (0.8), the conversation establishment degree of speakers A and C (0.6), and the conversation establishment degree of speakers B and C (0.7) received from the section analysis unit 15. The conversation establishment degree integration unit 16 also calculates the conversation establishment degree for the combination of speakers A, B, and D as "0.4" by averaging the conversation establishment degree of speakers A and B (0.8), the conversation establishment degree of speakers A and D (0.1), and the conversation establishment degree of speakers B and D (0.3) received from the section analysis unit 15. The conversation degree integration unit 16 also calculates the conversation establishment degree for the combination of speakers A, C, and D as "0.3" by averaging the conversation establishment degree for speakers A and C (0.6), the conversation establishment degree for speakers A and D (0.1), and the conversation establishment degree for speakers C and D (0.3) received from the section analysis unit 15. Note that this calculation is rounded down to the second decimal place. The conversation degree integration unit 16 similarly calculates the conversation establishment degree for the combination of speakers B, C, and D and the combination of speakers A, B, C, and D. This allows the conversation degree integration unit 16 to calculate the conversation establishment degree for each combination of all speakers (second conversation establishment degree) as shown in FIG. 15B. Note that the section analysis unit 15 may also perform a process of calculating the conversation establishment degree for all combinations of speakers based on the conversation establishment degrees between two people calculated by the section analysis unit 15.

[0044] The conversation degree integration unit 16 receives the conversation establishment degrees (first conversation establishment degrees) for all speaker combinations from the conversation analysis unit 13 (step S163). The conversation establishment degrees for all speaker combinations received from the conversation analysis unit 13 are the results calculated by the conversation analysis unit 13 in the processing of the first and second embodiments. The data shown in FIG. 15(C) shows the conversation establishment degrees (first conversation establishment degrees) for all speaker combinations acquired from the conversation analysis unit 13. The conversation degree integration unit 16 calculates an integrated conversation establishment degree ( FIG. 15(D)) using the conversation establishment degrees for all speaker combinations acquired from the conversation analysis unit 13 (first conversation establishment degrees shown in FIG. 15(C)) and the conversation establishment degrees calculated based on the information obtained from the section analysis unit 15 (second conversation establishment degrees shown in FIG. 15(B)) (step S164). 15 shows an example in which the average of the first conversation establishment degrees and the second conversation establishment degrees for matching speaker combinations is calculated as the integrated conversation establishment degree. This average may be calculated using a weighted average. The conversation degree integration unit 16 outputs the calculated integrated conversation establishment degree (step S165).

[0045] The information processing device 1 then determines whether a conversation is established, for example, using a threshold value for the degree of conversation establishment. For example, the threshold value is 0.5. The conversation establishment degrees of combinations of multiple speakers whose integrated conversation establishment degrees in FIG. 15(D) are equal to or greater than the threshold value of 0.5 are the conversation establishment degrees of the combination of speakers A and B and the conversation establishment degrees of speakers A, B, and C. In this case, the information processing device 1 may select and output the combination of speakers A and B or the combination of speakers A, B, and C as a group with an established conversation. Alternatively, if a speaker included in one of the multiple groups selected based on the conversation establishment degrees overlaps with a speaker in another group, the information processing device 1 may select the group of speakers with the highest conversation establishment degree and output it as information on the group with an established conversation. In the process shown in FIG. 15, there are two groups with a threshold value of 0.5 or greater: the combination of speakers A and B and the combination of speakers A, B, and C. In this case, the group of the combination of speakers A, B, and C, which has a high conversation success rate of "0.7", may be selected as a group with which conversation is successful, and information about this may be output.

[0046] [Explanation of Effects] The information processing device 1 in the third embodiment calculates the degree of established conversation independently from the presence or absence of speech in the speech section information, without reference to the content of the speech, using the speech section detection unit 14, the section analysis unit 15, and the conversation degree integration unit 16. Although the content of the speech acquired by the speech recognition unit 20 may contain speech recognition errors, this process calculates the degree of established conversation without taking the content of the speech into consideration. Therefore, it is possible to calculate evaluation results that can be grouped more accurately by adding elements that can be used to classify conversations relatively unaffected by the analysis results of the invention based on speech recognition errors.

[0047] [Fourth Embodiment] [Configuration Description] Fig. 16 is a fourth diagram showing functional blocks of an information processing device of the present disclosure. As shown in Fig. 16, an information processing device 1 of the present disclosure includes a conversation level evaluation unit 10, a speech recognition unit 2, a speech separation unit 3, and a conversational speech combination unit 4. The components of the conversation level evaluation unit 10 and the speech recognition unit 2 are the same as those in the third embodiment. The same components as those in the third embodiment are denoted by the same reference numerals as in Fig. 13, and detailed description thereof will be omitted.

[0048] The fourth embodiment differs from the third embodiment in the following respects. First, processing by the speech separation unit 3 is added as preprocessing for the speech recognition unit 2 and the conversation degree evaluation unit 10. Second, conversation speech creation processing by the conversation speech combination unit 4 is added as postprocessing for the conversation degree evaluation unit 10.

[0049] The speech separation unit 3 separates the speech of multiple speakers contained in the input speech data to generate speech data for each of the multiple speakers. The conversation speech combination unit 4 outputs combined speech data that combines the speech data of multiple speakers identified based on the degree of success obtained from the conversation degree integration unit 16.

[0050] [Explanation of Operation] The speech separation unit 3 receives input of speech data containing speech from multiple speakers, separates the speech of each speaker, and outputs speech data containing the individual speech content of each speaker. A commonly known speech separation method for generating speech data of a single speaker from speech data containing multiple speakers may be used. The input speech data may be single-channel or multi-channel, and may include auxiliary information such as microphone position, speech arrival direction, and other recording environment information and person information (e.g., attributes such as male, female, child, adult, etc.). The output speech data is speech extracted from only one person's speech, which was assumed as the speech data to be input to the speech recognition unit 2 and conversation level evaluation unit 10 in the second and third embodiments. The speech recognition unit 2 and conversation level evaluation unit 10 then perform the processing described in the other embodiments above.

[0051] The conversational voice combination unit 4 receives the degree of conversation establishment from the conversation degree integration unit 16. The conversational voice combination unit 4 also receives the voice data of each speaker from the voice separation unit 3. The conversational voice combination unit 4 identifies the identifiers of the speakers included in the conversation group based on the degree of conversation establishment of the received voices of each speaker. The speaker identifiers may be identifiers of the voice data corresponding to the speakers. For example, if the selected group includes speakers A, B, and C, the conversational voice combination unit 4 identifies the voice data of speakers A, B, and C. The conversational voice combination unit 4 generates and outputs combined voice data by combining the identified voice data. The conversational voice combination unit 4 may also be configured to output voices that do not belong to any conversation group as is. The number of output voices is determined according to the number of conversation groups.

[0052] [Explanation of Effects] The information processing device 1 in the fourth embodiment is equipped with the audio separation unit 3, which makes it possible to separate and analyze audio data containing speech from multiple speakers even when the audio for each speaker is not available. Furthermore, the conversational audio combination unit 4 makes it possible to play back audio for each conversation of speakers included in a group, improving the user experience.

[0053] [Fifth embodiment] [Configuration description] Fig. 17 is a diagram showing functional blocks of an information processing device according to another configuration of the present disclosure. Fig. 18 is a diagram showing a processing flow of an information processing device according to another configuration of the present disclosure. As shown in Fig. 16, the information processing device 1 may include at least a dividing means 61, a generating means 62, and a calculating means 63.

[0054] The dividing means 61 divides speech data including time-series speech content of multiple speakers into multiple pieces based on time to generate divided speech data (step S301). The generating means 62 generates prompt information including a pre-stored instruction statement that instructs determining whether a conversation is established for each combination of multiple speakers based on the utterance content, and selected divided utterance data from the divided utterance data (step S302). The calculating means 63 uses the prompt information to calculate the degree of establishment of a conversation for each combination of multiple speakers based on the results of the conversation establishment determination for each combination of multiple speakers output by the large-scale language model (step S303).

[0055] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure.

[0056] Some or all of the above-described embodiments can be described as, but are not limited to, the following supplementary notes.

[0057] (Supplementary Note 1) An information processing device comprising: a division means for dividing speech data including chronological speech content of a plurality of speakers into a plurality of parts based on time to generate divided speech data; a generation means for generating prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a judgment on the success of a conversation for each combination of the plurality of speakers based on the utterance content, and divided utterance data selected from the divided utterance data; and a calculation means for calculating the degree of success of a conversation for each combination of the plurality of speakers based on the result of the judgment on the success of a conversation for each combination of the plurality of speakers output by the large-scale language model using the prompt information.

[0058] (Supplementary Note 2) The information processing device according to Supplementary Note 1, wherein the splitting means inputs, for each of the plurality of speakers, speech data including the time-series speech content of one speaker extracted from the time-series speech content of the plurality of speakers, and generates split speech data by splitting the speech data into a plurality of pieces based on the times indicated by the time series; the generation means generates the prompt information including the instruction sentence and one piece of split utterance data for each of the plurality of speakers selected from the split utterance data based on the time series; and the information processing device comprises an input means for inputting, to the large-scale language model in the order of the time series, the plurality of pieces of prompt information each including the split utterance data selected based on the time series.

[0059] (Supplementary Note 3) The information processing device according to Supplementary Note 1 or Supplementary Note 2, comprising: a voice recognition means for inputting voice data and extracting the speech content, speech start time, and speech end time contained in the voice data to generate the speech data.

[0060] (Supplementary Note 4) The information processing device according to Supplementary Note 3, wherein the calculation means calculates a first degree of conversation success for each combination of the plurality of speakers based on a result of a conversation success determination for each combination of the plurality of speakers output by the large-scale language model using the prompt information; speech section detection means detects speech section information indicating at least an utterance start time and an utterance end time of the speech included in the audio data; section analysis means calculates a second degree of conversation success using a sound and silence state of the utterance identified based on the speech section information; and conversation degree integration means calculates an integrated degree of conversation success based on the first degree of conversation success and the second degree of conversation success.

[0061] (Supplementary Note 5) An information processing device as described in Supplementary Note 3 or Supplementary Note 4, comprising: a voice separation means for extracting the time series speech content of each speaker from the time series speech content of the multiple speakers included in the voice data, and generating the speech data of each of the multiple speakers; and a combination means for outputting combined data that combines the speech content of the speech data of the multiple speakers identified based on the degree of validity based on the time series.

[0062] (Supplementary Note 6) The information processing device according to any one of Supplementary Note 1 to Supplementary Note 5, wherein the speech data is document data indicating characters of the speech.

[0063] (Supplementary Note 7) The information processing device according to any one of Supplementary Note 1 to Supplementary Note 5, wherein the speech data is voice data representing a speech sound.

[0064] (Supplementary Note 8) An information processing system comprising: a division means for dividing speech data including chronological speech content of a plurality of speakers into a plurality of parts based on time to generate divided speech data; a generation means for generating prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a judgment on the success of a conversation for each combination of the plurality of speakers based on the utterance content, and selected divided utterance data from the divided utterance data; and a calculation means for calculating the degree of success of a conversation for each combination of the plurality of speakers based on the result of the judgment on the success of a conversation for each combination of the plurality of speakers output by the large-scale language model using the prompt information.

[0065] (Supplementary Note 9) The information processing system described in Supplementary Note 8, wherein the splitting means inputs, for each of the multiple speakers, speech data including the time-series speech content of one speaker extracted from the time-series speech content of the multiple speakers, and generates split speech data by splitting the speech data into multiple pieces based on the times indicated by the time series; the generation means generates the prompt information including the instruction sentence and one piece of split utterance data for each of the multiple speakers selected from the split utterance data based on the time series; and the information processing system includes an input means for inputting, to the large-scale language model in the order of the time series, the multiple pieces of prompt information each including the split utterance data selected based on the time series.

[0066] (Supplementary Note 10) An information processing system according to Supplementary Note 8 or Supplementary Note 9, comprising: a voice recognition means for inputting voice data and extracting the speech content, speech start time, and speech end time contained in the voice data to generate the speech data.

[0067] (Supplementary Note 11) The information processing system according to Supplementary Note 10, wherein the calculation means calculates a first degree of conversation success for each combination of the plurality of speakers based on a result of a conversation success determination for each combination of the plurality of speakers output by the large-scale language model using the prompt information; speech section detection means detects speech section information indicating at least an utterance start time and an utterance end time of the speech included in the audio data; section analysis means calculates a second degree of conversation success using a sound and silence state of the utterance identified based on the speech section information; and conversation degree integration means calculates an integrated degree of conversation success based on the first degree of conversation success and the second degree of conversation success.

[0068] (Supplementary Note 12) An information processing system as described in Supplementary Note 10 or Supplementary Note 11, comprising: a voice separation means for extracting the time series speech content of each speaker from the time series speech content of the multiple speakers included in the voice data, and generating the speech data of each of the multiple speakers; and a combination means for outputting combined data in which the speech content of the speech data of the multiple speakers identified based on the degree of validity is combined based on the time series.

[0069] (Supplementary Note 13) The information processing system according to any one of Supplementary Note 8 to Supplementary Note 12, wherein the speech data is document data indicating characters of the speech.

[0070] (Supplementary Note 14) The information processing system according to any one of Supplementary Note 8 to Supplementary Note 12, wherein the speech data is voice data representing the voice of a speech.

[0071] (Supplementary Note 15) An information processing method comprising: dividing speech data containing chronological speech content of multiple speakers into multiple parts based on time to generate divided speech data; generating prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a determination of the success of a conversation for each combination of the multiple speakers based on the utterance content, and selected divided utterance data from the divided utterance data; and using the prompt information to calculate the degree of success of a conversation for each combination of the multiple speakers based on the results of the determination of the success of a conversation for each combination of the multiple speakers output by the large-scale language model.

[0072] (Supplementary Note 16) An information processing method according to Supplementary Note 15, comprising the steps of: inputting speech data for each of the plurality of speakers, the speech data including the time-series speech content of one speaker extracted from the time-series speech content of the plurality of speakers; generating split speech data by dividing the speech data into a plurality of pieces based on the times indicated by the time series; generating prompt information including the instruction sentence and one piece of split speech data for each of the plurality of speakers selected from the split utterance data based on the time series; and inputting the plurality of pieces of prompt information, each including the split utterance data selected based on the time series, to the large-scale language model in the order of the time series.

[0073] (Supplementary Note 17) The information processing method according to Supplementary Note 15 or Supplementary Note 16, comprising: inputting voice data, and extracting the speech content, speech start time, and speech end time included in the voice data to generate the speech data.

[0074] (Supplementary Note 18) The information processing method according to Supplementary Note 17, further comprising: calculating a first degree of conversation success for each combination of the plurality of speakers based on a result of a conversation success determination for each combination of the plurality of speakers output by the large-scale language model using the prompt information; detecting speech section information indicating at least an utterance start time and an utterance end time of the speech included in the audio data; calculating a second degree of conversation success using the voiced and silent states of the speech identified based on the speech section information; and calculating an integrated degree of conversation success based on the first degree of conversation success and the second degree of conversation success.

[0075] (Supplementary Note 19) An information processing method described in Supplementary Note 17 or Supplementary Note 18, which extracts the time-series speech content of each speaker from the time-series speech content of the multiple speakers included in the audio data to generate the speech data for each of the multiple speakers, and outputs combined data by combining the speech content of the speech data of the multiple speakers identified based on the degree of validity based on the time series.

[0076] (Supplementary Note 20) The information processing method according to any one of Supplementary Note 15 to Supplementary Note 19, wherein the speech data is document data indicating characters of the speech.

[0077] (Supplementary Note 21) The information processing method according to any one of Supplementary Notes 15 to 19, wherein the speech data is voice data representing the voice of a speech.

[0078] (Supplementary Note 22) A storage medium storing a program that causes a computer of an information processing device to function as: a division means that divides speech data containing chronological speech content of multiple speakers into multiple pieces based on time to generate divided speech data; a generation means that generates prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a judgment on the success of a conversation for each combination of the multiple speakers based on the utterance content, and selected divided utterance data from the divided utterance data; and a calculation means that calculates the degree of success of a conversation for each combination of the multiple speakers based on the result of the judgment on the success of a conversation for each combination of the multiple speakers output by the large-scale language model using the prompt information.

[0079] (Supplementary Note 23) A storage medium that stores a program according to Supplementary Note 22, which functions as an input means, wherein the splitting means inputs, for each of the plurality of speakers, speech data including the time-series speech content of one speaker extracted from the time-series speech content of the plurality of speakers, and generates split speech data by splitting the speech data into a plurality of pieces based on the times indicated by the time series; the generation means generates the prompt information including the instruction sentence and one piece of split utterance data for each of the plurality of speakers selected from the split utterance data based on the time series; and inputs the plurality of pieces of prompt information each including the split utterance data selected based on the time series to the large-scale language model in the order of the time series.

[0080] (Supplementary Note 24) A storage medium storing the program according to Supplementary Note 22 or Supplementary Note 23, which functions as a speech recognition means that receives voice data as input and extracts the speech content, speech start time, and speech end time contained in the voice data to generate the speech data.

[0081] (Supplementary Note 25) A storage medium storing a program according to Supplementary Note 24, wherein the calculation means functions as: calculates a first degree of conversation success for each combination of the plurality of speakers based on a result of a conversation success determination for each combination of the plurality of speakers output by the large-scale language model using the prompt information; speech section detection means detects speech section information indicating at least an utterance start time and an utterance end time of the speech included in the audio data; section analysis means calculates a second degree of conversation success using a sound and silence state of the utterance identified based on the speech section information; and conversation degree integration means calculates an integrated degree of conversation success based on the first degree of conversation success and the second degree of conversation success.

[0082] (Supplementary Note 26) A storage medium storing a program according to Supplementary Note 24 or Supplementary Note 25, which functions as: a voice separation means for extracting the time series speech content of each speaker from the time series speech content of the multiple speakers included in the voice data, and generating the speech data of each of the multiple speakers; and a combination means for outputting combined data by combining the speech content of the speech data of the multiple speakers identified based on the degree of validity, based on the time series.

[0083] (Supplementary Note 27) A storage medium storing the program according to any one of Supplementary Note 22 to Supplementary Note 26, wherein the speech data is text data indicating characters of the speech.

[0084] (Supplementary Note 28) A storage medium storing the program according to any one of Supplementary Note 22 to Supplementary Note 26, wherein the speech data is voice data representing the voice of a speech.

[0085] REFERENCE SIGNS LIST 1... Information processing device 2... Speech recognition unit 3... Speech separation unit 10... Conversation level evaluation unit 11... Segmentation unit (segmentation means 61) 12... Prompt generation unit (generation means 62) 13... Conversation analysis unit (calculation means 63) 14... Speech section detection unit 15... Section analysis unit 16... Conversation level integration unit 20... Speech recognition unit

Claims

1. An information processing device comprising: a division means for dividing speech data containing chronological speech content of multiple speakers into multiple pieces based on time to generate divided speech data; a generation means for generating prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a judgment on the success of a conversation for each combination of the multiple speakers based on the utterance content, and selected divided utterance data from the divided utterance data; and a calculation means for calculating the degree of success of a conversation for each combination of the multiple speakers based on the result of the judgment on the success of a conversation for each combination of the multiple speakers output by the large-scale language model using the prompt information.

2. The information processing device according to claim 1, wherein the splitting means inputs, for each of the multiple speakers, speech data including the time-series speech content of one speaker extracted from the time-series speech content of the multiple speakers, and generates split speech data by splitting the speech data into multiple pieces based on the times indicated by the time series; the generation means generates the prompt information including the instruction sentence and one piece of split speech data for each of the multiple speakers selected from the split utterance data based on the time series; and the information processing device comprises an input means for inputting, to the large-scale language model in the order of the time series, the multiple pieces of prompt information each including the split utterance data selected based on the time series.

3. The information processing device according to claim 2, further comprising: a voice recognition means for inputting voice data and extracting the speech content, speech start time, and speech end time contained in the voice data to generate the speech data.

4. The information processing device according to claim 3, wherein the calculation means calculates a first degree of conversation success for each combination of the plurality of speakers based on the result of a conversation success determination for each combination of the plurality of speakers output by the large-scale language model using the prompt information; speech section detection means detects speech section information indicating at least the speech start time and speech end time of the speech included in the audio data; section analysis means calculates a second degree of conversation success using the voiced and silent states of the speech identified based on the speech section information; and conversation degree integration means calculates an integrated degree of conversation success based on the first degree of conversation success and the second degree of conversation success.

5. An information processing device as described in claim 3 or claim 4, comprising: a voice separation means for extracting the time series speech content of each speaker from the time series speech content of the multiple speakers included in the voice data and generating the speech data of each of the multiple speakers; and a combination means for outputting combined data by combining the speech content of the speech data of the multiple speakers identified based on the degree of validity based on the time series.

6. The information processing device according to claim 5, wherein the speech data is text data representing characters of the speech.

7. The information processing device according to claim 5, wherein the speech data is voice data representing the voice of a speech.

8. An information processing system comprising: a division means for dividing speech data containing chronological speech content of a plurality of speakers into a plurality of parts based on time to generate divided speech data; a generation means for generating prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a judgment on the success of a conversation for each combination of the plurality of speakers based on the speech content, and selected divided speech data from the divided utterance data; and a calculation means for calculating the degree of success of a conversation for each combination of the plurality of speakers based on the results of the judgment on the success of a conversation for each combination of the plurality of speakers output by the large-scale language model using the prompt information.

9. An information processing method comprising: dividing speech data containing chronological speech content of multiple speakers into multiple parts based on time to generate divided speech data; generating prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a judgment on the success of a conversation for each combination of the multiple speakers based on the utterance content, and selected divided speech data from the divided utterance data; and using the prompt information to calculate the degree of success of a conversation for each combination of the multiple speakers based on the results of the judgment on the success of a conversation for each combination of the multiple speakers output by the large-scale language model.

10. A storage medium storing a program that causes a computer of an information processing device to function as: a division means that divides speech data containing chronological speech content of multiple speakers into multiple pieces based on time to generate divided speech data; a generation means that generates prompt information to be input to a large-scale language model, the prompt information including a pre-stored instruction statement that instructs a judgment on the success of a conversation for each combination of the multiple speakers based on the speech content, and selected divided utterance data from the divided utterance data; and a calculation means that calculates the degree of success of a conversation for each combination of the multiple speakers based on the results of the judgment on the success of a conversation for each combination of the multiple speakers output by the large-scale language model using the prompt information.

Images