Loudspeaker broadcasting system

Abstract

To provide a loudspeaker broadcasting system capable of acquiring an announcement sentence corresponding to the state of a broadcasting area.SOLUTION: The public address broadcasting system 10 publically broadcasts sound in a broadcasting area BA. A loudspeaker broadcasting system 10 includes an extraction part 21 for extracting event information from the situation of a broadcasting area BA, a generation part 22 for generating prompt information for instructing the generation of an announcement sentence for the broadcasting area BA based on the event information, and an acquisition part 23 for acquiring the announcement sentence generated by a large-scale language model 24 by inputting the prompt information to the large-scale language model 24.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to public address systems. [Background technology]

[0002] There is known a system that loudspeakers broadcast announcements generated in real time. For example, an automatic public address system disclosed in Patent Document 1 loudspeakers broadcast announcements containing specific keywords. In the automatic public address system disclosed in Patent Document 1, template text is prepared in advance, and the announcements are generated by adding specific keywords to the template text. In the example disclosed in Patent Document 1, the specific keywords are information related to the operation of railway vehicles. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-190832 Summary of the Invention [Problem to be solved by the invention]

[0004] If announcements are generated by preparing templates for announcements in advance and adding specific keywords, as in the automatic broadcasting device of Patent Document 1, the announcements will be uniform. Uniform broadcasts may not be effective enough. [Means for solving the problem]

[0005] (1) A loudspeaker system that solves the problem is a loudspeaker system that loudspeakers broadcasts audio to a broadcast area, and includes an extraction unit that extracts event information from the situation in the broadcast area, a generation unit that generates prompt information that instructs the creation of an announcement to be loudspeaked to the broadcast area based on the event information, and an acquisition unit that inputs the prompt information into a large-scale language model and thereby acquires the announcement generated by the large-scale language model.

[0006] According to this configuration, announcements are generated from prompt information generated based on event information. Because announcements that reflect the situation in the broadcast area are generated using a large-scale language model, announcements that correspond to the situation in the broadcast area can be obtained.

[0007] (2) In the loudspeaker system described in (1) above, the event information includes at least one of first event information, second event information, and third event information, the first event information being information about an event that occurred in the broadcast area, the second event information being information about the attributes of people involved in the event, and the third event information being information about at least one of the location of the broadcast area and the broadcast time.

[0008] According to this configuration, it is possible to acquire an announcement text corresponding to at least one of the first event information, the second event information, and the third event information.

[0009] (3) In the loudspeaker system described in (1) or (2) above, the generation unit generates the prompt information by combining the event information with base information that instructs the creation of the announcement text, and the base information is set in advance.

[0010] According to this configuration, since the base information is set in advance, the prompt information can be easily generated by combining the base information with the event information.

[0011] (4) In the loudspeaker system according to any one of (1) to (3) above, the generating unit generates the prompt information including purpose information relating to the purpose of loudspeaker broadcasting to the broadcast area.

[0012] According to this configuration, since the prompt information includes purpose information, it is possible to obtain announcement sentences that match the purpose of the loudspeaker broadcast using a large-scale language model.

[0013] (5) In the loudspeaker system described in (4) above, the generation unit selects the target information corresponding to the event information extracted by the extraction unit from a target database in which the target information corresponding to the event information is stored.

[0014] According to this configuration, since the purpose information corresponding to the event information is selected, the purpose of the broadcast can be set according to the situation in the broadcast area.

[0015] (6) In the loudspeaker broadcasting system described in (4) above, the generation unit inputs the situation of the broadcast area to the large-scale language model, thereby causing the large-scale language model to generate the target information.

[0016] According to this configuration, the purpose information is generated using a large-scale language model, so that the purpose of broadcasting can be set appropriately according to the situation in the broadcasting area.

[0017] (7) In the loudspeaker system described in any one of (4) to (6) above, an effect determination unit is provided that determines the broadcasting effect of the loudspeaker broadcast to the broadcast area based on the announcement text, and the generation unit generates the prompt information based on the broadcasting effect.

[0018] According to this configuration, since the prompt information is generated based on the broadcast effect, it is possible to obtain an announcement text with a higher broadcast effect than prompt information that is not based on the broadcast effect.

[0019] (8) In the loudspeaker system according to any one of (4) to (7) above, a broadcast database is generated in which data linked to the objective information, the prompt information, and the announcement text is stored.

[0020] According to this configuration, the broadcast database can be used as a knowledge base regarding loudspeaker broadcasting.

[0021] (9) In the loudspeaker system described in any one of (1) to (8) above, the loudspeaker system includes a sound source generation unit that generates audio data based on the announcement text, and a broadcasting unit that loudspeaks and broadcasts audio based on the audio data to the broadcast area.

[0022] According to this configuration, the broadcasting unit can broadcast audio based on the announcement text to the broadcasting area.

[0023] (10) In the loudspeaker system described in any one of (1) to (9) above, a detection unit is provided that detects the status of the broadcast area, and the extraction unit extracts the event information from the status of the broadcast area detected by the detection unit.

[0024] According to this configuration, event information can be extracted according to the situation in the broadcast area.

[0025] (11) In the loudspeaker system described in (10) above, the detection unit includes at least one of a camera device that captures images of the broadcast area and a microphone device that captures audio from the broadcast area.

[0026] According to this configuration, event information can be extracted from at least one of the video of the broadcast area and the audio of the broadcast area. [Effects of the Invention]

[0027] The loudspeaker system makes it possible to obtain announcement text that is appropriate for the situation in the broadcast area. [Brief explanation of the drawings]

[0028] [Figure 1] 1 is a block diagram showing an example of the configuration of a voice dialogue system according to a first embodiment. [Figure 2] FIG. 10 is a block diagram showing an example of the configuration of prompt information. [Figure 3] 1 is an example of a target database. [Figure 4] This is the first example of the broadcast area situation. [Figure 5] 5 is a diagram showing an example of prompt information corresponding to FIG. 4 and an example of an announcement generated from the prompt information by a large-scale language model. FIG. [Figure 6] This is the second example of the broadcast area situation. [Figure 7] 7 is a diagram showing an example of prompt information corresponding to FIG. 6 and an example of an announcement generated from the prompt information by a large-scale language model. FIG. [Figure 8] This is the third example of the broadcast area situation. [Figure 9] 9 is a diagram showing an example of prompt information corresponding to FIG. 8 and an example of an announcement generated from the prompt information by a large-scale language model. FIG. [Figure 10] This is the fourth example of the broadcast area situation. [Figure 11] 11 is a diagram showing an example of prompt information corresponding to FIG. 10 and an example of an announcement generated from the prompt information by a large-scale language model. FIG. [Figure 12] FIG. 10 is a diagram showing an example of intent generation prompt information and an example of intent information generated by a large-scale language model from the intent generation prompt information in the voice dialogue system according to the second embodiment. [Figure 13] 13 is a diagram showing an example of prompt information including the objective information of FIG. 12, and an example of an announcement sentence generated from the prompt information by a large-scale language model. FIG. [Figure 14] FIG. 10 is a block diagram showing an example of the configuration of a voice dialogue system according to a third embodiment. [Figure 15] 10 is a flowchart showing the flow of processing executed in the voice dialogue system according to the third embodiment. [Figure 16] 10 is an example of a database of the voice dialogue system according to the fourth embodiment. [Figure 17] 10A and 10B are diagrams showing examples of prompt information and examples of announcements generated from the prompt information by a large-scale language model in a voice dialogue system according to a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0029] First Embodiment A loudspeaker system 10 according to an embodiment will be described with reference to FIGS. 1 to 11. FIG.

[0030] <Public address system> As shown in Figures 1 and 2, the public address system 10 broadcasts audio to a broadcast area BA. Examples of the broadcast area BA include an area set up inside a building, such as a facility, an outdoor area outside a building on a premises, and a wide area such as a disaster prevention radio. The public address system includes background music, sound effects, and audio related to the broadcast content. The broadcast content includes warnings, announcements, weather information, greetings, etc. The announcements include information about the broadcast area, event information, product information, lost child notifications, disaster notifications, event times, and schedule delays.

[0031] <Configuration of public address system> The loudspeaker system 10 includes a computer 11. The computer 11 is a server. The computer 11 may be a general-purpose computer. The computer 11 may also be configured by combining a plurality of computers.

[0032] The computer 11 has a control unit and a storage unit. The control unit includes a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). The storage unit includes a semiconductor memory, a hard disk, an optical disk, and a magnetic medium. The control unit executes various programs stored in the storage unit, thereby causing various programs to be executed on the computer 11.

[0033] The computer 11 is communicably connected to a broadcasting device 12 located in the broadcast area BA via a network. The network is, for example, the Internet. The computer 11 may be located in the broadcast area BA or may be located away from the broadcast area BA. The computer 11 may be incorporated into facilities such as broadcasting equipment in the broadcast area BA.

[0034] The loudspeaker system 10 includes a sound source generating unit 13 and a broadcasting unit 14. The sound source generating unit 13 is provided in a computer 11, and the broadcasting unit 14 is provided in a broadcasting device 12.

[0035] The sound source generation unit 13 generates voice data based on the announcement text AT. The announcement text AT is text corresponding to the broadcast content of the loudspeaker broadcast. The sound source generation unit 13 includes a voice synthesis engine such as a TTS (Text to Speech) model. The voice data generated by the sound source generation unit 13 is transmitted to the broadcasting device 12.

[0036] One or more broadcasting units 14 are arranged in the broadcasting area BA. The broadcasting units 14 loudly broadcast audio based on the audio data into the broadcasting area BA. The broadcasting units 14 include a playback device that plays back the audio data, an amplifier that amplifies the played-back audio signal, a speaker device, etc. The broadcasting units 14 loudly broadcast the contents of the announcement text AT.

[0037] The loudspeaker system 10 includes a detection unit 15. The detection unit 15 detects the status of the broadcast area BA. The detection unit 15 includes at least one of a camera device 16 that captures images of the broadcast area BA and a microphone device 17 that acquires audio from the broadcast area BA.

[0038] The status of broadcast area BA includes at least one of video of broadcast area BA and audio of broadcast area BA. Detection unit 15 transmits information related to the detected status of broadcast area BA to computer 11. When transmitting the information related to the status of broadcast area BA to computer 11, detection unit 15 may also transmit information such as the identification number of detection unit 15 that detected the information, the location where detection unit 15 that detected the information is located, the temperature at that location, weather, detection time, etc.

[0039] <Announcement text generation> The announcement AT is generated by the computer 11. In the computer 11, the announcement AT is generated based on the situation of the broadcast area BA detected by the detection unit 15.

[0040] The loudspeaker system 10 includes an extraction unit 21, a generation unit 22, and an acquisition unit 23. The extraction unit 21 extracts event information E from the situation in the broadcast area BA. The generation unit 22 generates prompt information P1 based on the event information E. The acquisition unit 23 inputs the prompt information P1 into the large-scale language model 24, thereby acquiring the announcement sentence AT generated by the large-scale language model 24.

[0041] The extraction unit 21, the generation unit 22, and the acquisition unit 23 are realized by a program installed in the computer 11. The program that realizes the extraction unit 21, the generation unit 22, and the acquisition unit 23 is stored in the memory unit of the computer 11. The computer 11 generates the announcement text AT from the situation in the broadcast area BA by the control unit of the computer 11 executing the program that realizes the extraction unit 21, the generation unit 22, and the acquisition unit 23. Each of the extraction unit 21, the generation unit 22, and the acquisition unit 23 may be a function of a program realized on the computer 11. Each of the extraction unit 21, the generation unit 22, and the acquisition unit 23 may be configured by a computer 11 that is independent of one another.

[0042] The large-scale language model (also abbreviated as LLM) 24 is a machine learning model trained based on a huge amount of linguistic information. The large-scale language model 24 is installed in, for example, the computer 11. The large-scale language model 24 may be installed in a computer other than the computer 11, a cloud server, or the like. The large-scale language model 24 may be operated by an operator other than the operator that operates the computer 11 or the broadcasting device 12.

[0043] Examples of the large-scale language model 24 include ChatGPT (registered trademark), Bard (registered trademark), LaMDA (registered trademark), LLaMA (registered trademark), tsuzumi (registered trademark), CyberAgentLM (registered trademark), etc. The large-scale language model 24 may be configured by the user of the public address system 10.

[0044] The large-scale language model 24 generates a response to the input prompt information P1. For example, if the prompt information P1 includes an instruction to create an announcement AT, the large-scale language model 24 generates the announcement AT.

[0045] The computer 11 starts generating an announcement sentence AT when it extracts a specific event from the situation in the broadcast area BA. If the specific event is included in the extracted event information E, the extraction unit 21 outputs the event information E to the generation unit 22. The specific event is, for example, set in advance.

[0046] The specific event may be, for example, an event that requires the attention of the person involved or bystanders, such as the outbreak of a fire, a person falling, a fight, etc. The specific event may also be an event that requires the promotion of facility use, such as customers entering a retail store.

[0047] The extraction unit 21 extracts event information E from the situation in the broadcast area BA detected by the detection unit 15. The extraction unit 21 extracts the behavior of people, animals, vehicles, etc. in the broadcast area BA from the video or audio of the broadcast area BA using behavior recognition technology. The extraction unit 21 also extracts attributes such as the faces, ages, and genders of people in the broadcast area BA from the video or audio of the broadcast area BA using face recognition technology. In addition, the extraction unit 21 extracts the occurrence of fires, fallen trees, etc. in the broadcast area BA from the video or audio of the broadcast area BA using a machine learning model such as deep learning.

[0048] <prompt information> The prompt information P1 is information that instructs the large-scale language model 24 to create an announcement sentence AT to be broadcast to the broadcast area BA. The prompt information P1 includes base information B1 that instructs the creation of the announcement sentence AT, and event information E. The generation unit 22 combines the base information B1 with the event information E to generate the prompt information P1.

[0049] The base information B1 includes text that instructs the large-scale language model 24 to create an announcement sentence AT. Examples of the base information B1 are text such as "(Event information E) has occurred. Please generate a sentence to be broadcast to alert those in the vicinity" and "(Event information E) has occurred. Please generate a sentence to be broadcast to resolve the problem." In the above example, text related to the event information E is inserted where it says (event information E).

[0050] The base information B1 may include elements such as the number of characters, number of words, tone, ending, first person, etc. of the announcement text AT. The various elements of the announcement text AT are set, for example, according to the broadcast area BA. The base information B1 may also include example sentences of the announcement text AT. The example sentences of the announcement text AT are, for example, announcements made in loudspeaker broadcasts previously broadcast in the broadcast area BA. By including the various elements and example sentences of the announcement text AT in the base information B1, it is possible to create a sense of unity in the loudspeaker broadcast of the announcement text AT.

[0051] The base information B1 is set in advance. Different text of the base information B1 is prepared for each broadcast area BA. The base information B1 may be common to each piece of prompt information P1 generated by the generation unit 22.

[0052] For example, a plurality of pieces of base information B1 are preset, such as "(Event information E) has occurred. Please generate a message to be broadcast to alert those in the vicinity" and "(Event information E) has occurred. Please generate a message to be broadcast to resolve the problem." The plurality of pieces of base information B1 are stored, for example, in a storage unit. The generation unit 22 selects one of the plurality of pieces of base information B1 according to the event information E extracted by the extraction unit 21. For example, the event information E is classified into one of a plurality of event groups, and each of the plurality of pieces of base information B1 is associated with one of the plurality of event groups. The generation unit 22 selects the base information B1 corresponding to the event group into which the extracted event information E is classified.

[0053] The event information E is, for example, information related to a specific event. The event information E includes, for example, keywords such as "fire outbreak," "person falling," "man in his 40s," and "street." The event information E includes at least one of first event information E1, second event information E2, and third event information E3.

[0054] The first event information E1 is information about an event that occurred in the broadcast area BA. Examples of the first event information E1 include information about phenomena that occur in the broadcast area BA, such as a fire or a fallen tree, or information about human behavior, such as a person sleeping, a person falling down, a fight, or a person picking up an item in a retail store.

[0055] The second event information E2 is information about the attributes of people involved in the event. Examples of the second event information E2 include personal characteristics of people involved in the event, such as gender, age, and health condition, and membership information for retail stores, etc. People involved in the event include not only the people involved in the event, but also people around the event.

[0056] The third event information E3 is information relating to at least one of the location in the broadcast area BA and the broadcast period. Examples of the third event information E3 include an address, a facility name, a season, weather, temperature, time, etc. The extraction unit 21 may extract the third event information E3 not from the video or audio, but based on information such as the identification information, location information, and detection time of the detection unit 15 that detected the video or audio.

[0057] The generation unit 22 generates prompt information P1 including purpose information G. The purpose information G relates to the purpose of loudspeaker broadcasting to the broadcast area BA. The purpose of loudspeaker broadcasting can be to draw attention and to promote facility use. The purpose of drawing attention can be, for example, to get the parties involved to stop a fight, to get people around to report the fight, to wake up a sleeping person, or to get people around a sleeping person to call out to them. The purpose of promoting facility use can be, for example, to guide people to a specific facility or to get people to buy a specific product.

[0058] 3 stores destination information G corresponding to event information E. Destination database DB1 is stored in the storage unit of computer 11. Generation unit 22 selects, from destination database DB1, destination information G corresponding to event information E extracted by extraction unit 21.

[0059] The objective database DB1 stores at least one of a first objective and a second objective for each event. The first objective is the objective of loudspeaker broadcasting to the parties involved in the event, and the second objective is the objective of loudspeaker broadcasting to people around the event. For example, an event such as a fight has two objectives set: a first objective to get the parties involved to stop the fight, and a second objective to get people around to report the fight. For example, an event such as someone falling has no first objective set for the parties involved, and only a second objective to get people around to help the person who fell is set.

[0060] <Announcement text generation> 1 to 11, examples of announcement sentences AT generated based on the status of the broadcast area BA will be described. In each of the first to fourth examples below, the generation unit 22 generates prompt information P1 for each of inputs 1 to 4. The acquisition unit 23 also acquires announcement sentences AT for each of outputs 1 to 4.

[0061] [Example 1] Fig. 4 is an example of a video of a fire breaking out in a plaza, which is broadcast area BA. Extraction unit 21 extracts "fire breaking out" as first event information E1 from the video in Fig. 4, as indicated by the dashed line.

[0062] In the prompt information P1 of input 1 shown in Figure 5, the base information B1 is "(First event information E1) has been detected at this location. Please create a sentence to be broadcast on loudspeaker to alert those in the vicinity." The above base information B1 is combined with the first event information E1, "fire outbreak." In response to the prompt information P1 of input 1, the large-scale language model 24 generates the announcement sentence AT of output 1.

[0063] In one example, the event information E, "fire outbreak," is classified into a first event group. The first event group includes event information E, such as "fire outbreak" and "fallen tree," which require alerting people in the vicinity. The first event group is associated with base information B1, which reads, "(first event information E1) has been detected at this location. Please create a message to be broadcast on loudspeaker to alert people in the vicinity." When the extraction unit 21 extracts "fire outbreak" as the first event information E1, the generation unit 22 selects the base information B1 associated with the first event group. The base information B1 of the input 1 includes objective information G, which is to alert people in the vicinity.

[0064] In the prompt information P1 of FIG. 5, the second event information E2 and the third event information E3 are omitted.

[0065] [Example 2] Figure 6 is an example of a video of a person falling at the entrance of a facility in broadcast area BA. From the video in Figure 6, the extraction unit 21 extracts "person falling" as first event information E1, as indicated by the dashed line. The extraction unit 21 also extracts the person's age "40s" and gender "male" as second event information E2, and the location "entrance" as third event information E3.

[0066] In the prompt information P1 in Figure 7, the base information B1 is "(First event information E1) has been detected at this location. Please create a sentence to be broadcast over loudspeaker to solve the problem. The person is (Second event information E2). The location is (Third event information E3)." Prompt information P1, which combines each event information E with the above base information B1, is input as input 2. In response to the prompt information P1 of input 2, the large-scale language model 24 generates announcement sentence AT of output 2.

[0067] In one example, the event information E, "person falls," is classified into the second event group. The second event group includes event information E that requires resolution of the event through loudspeaker broadcasting, such as "person falls" and "fight breaks out." The second event group is associated with base information B1 that reads, "(First event information E1) has been detected at this location. Please create a message to be broadcast to resolve the problem. The person is (second event information E2). The location is (third event information E3)." When the extraction unit 21 extracts "person falls" as the first event information E1, the generation unit 22 selects the base information B1 associated with the second event group.

[0068] 7, the base information B1 includes the goal information G of solving the problem. The prompt information P1 may further include the goal information G of helping a person who has fallen.

[0069] [Example 3] Figure 8 is an example of a video of two men fighting. From the video in Figure 8, the extraction unit 21 extracts "fight" as first event information E1, as indicated by the dashed line. The extraction unit 21 also extracts the parties' ages ("40s" and "20s") and gender ("male") as second event information E2, and the location ("on the street") as third event information E3.

[0070] The generation unit 22 selects from the objective database DB1 a first objective to have the parties involved stop the fight and a second objective to have people nearby report the fight, which correspond to the event "fight."

[0071] In the prompt information P1 of input 3 shown in FIG. 9, the base information B1 is "(First event information E1) has been detected at this location. Please create a sentence for loudspeaker broadcast with the purpose of (first purpose of purpose information G). Please create a sentence for loudspeaker broadcast with the purpose of (second purpose of purpose information G). The party involved in (first event information E1) is (second event information E2). The location is (third event information E3)." Prompt information P1, which combines event information E and purpose information G with the above base information B1, is input as input 3. In response to the prompt information P1 of input 3, the large-scale language model 24 outputs announcement sentences AT as shown in outputs 3-1 and 3-2. Output 3-1 is an announcement sentence corresponding to the first purpose, and output 3-2 is an announcement sentence corresponding to the second purpose.

[0072] [Example 4] Fig. 10 is an example of a video of a sales floor in a retail store. From the video in Fig. 10, the extraction unit 21 extracts "the target is in the product sales floor" as first event information E1, as indicated by the dashed line. The extraction unit 21 also extracts the target's age "20s" and gender "male" as second event information E2, and the date and time "June 29th" and location "Fukuoka city" as third event information E3.

[0073] In the prompt information P1 of input 3 shown in FIG. 11, the base information B1 is "The target is (second event information E2). The date and time is (third event information E3). The location is (third event information E3). Under these conditions, please create a promotional slogan for (objective information G)." The objective information G is "to motivate the target to buy a product." In the example of FIG. 11, the product is chocolate. If the first event information E1 is extracted as being in a sales area selling a specific product (e.g., "the target is in a sales area selling chocolate products"), it is preferable that the objective information G be to motivate the target to buy the specific product (e.g., "to motivate the target to buy chocolate"). In response to the prompt information P1 of input 4, the large-scale language model 24 outputs an announcement sentence AT as shown in output 4.

[0074] <Operation of the embodiment> The operation of this embodiment will be described. In the loudspeaker system 10, event information E is extracted from the situation in the broadcast area BA, and an announcement AT is generated by the large-scale language model 24 based on the extracted event information E. Because the announcement AT is generated based on the event information E, it is possible to generate announcements AT that suit a variety of situations. Because the announcement AT is generated by the large-scale language model 24, the same announcement AT may not be generated even if the situation in the broadcast area BA is the same. In the loudspeaker system 10, various variations of announcements AT are generated rather than a uniform one, so it is possible to attract the attention of people who are the targets of the loudspeaker broadcast.

[0075] <Effects of the embodiment> The effects of this embodiment will be described. (1) The loudspeaker system 10 includes an extraction unit 21, a generation unit 22, and an acquisition unit 23. The extraction unit 21 extracts event information E from the situation in the broadcast area BA. The generation unit 22 generates prompt information P1 that instructs the creation of an announcement sentence AT to be broadcast to the broadcast area BA, based on the event information E. The acquisition unit 23 inputs the prompt information P1 into the large-scale language model 24, thereby acquiring the announcement sentence AT generated by the large-scale language model 24.

[0076] According to this configuration, the announcement sentence AT is generated from prompt information P1, which is generated based on event information E. Since the large-scale language model 24 generates the announcement sentence AT that reflects the situation in the broadcast area BA, it is possible to obtain the announcement sentence AT that corresponds to the situation in the broadcast area BA.

[0077] (2) The event information E includes at least one of first event information E1, second event information E2, and third event information E3. The first event information E1 is information about an event that occurred in the broadcast area BA. The second event information E2 is information about the attributes of people involved in the event. The third event information E3 is information about at least one of the location in the broadcast area BA and the broadcast time.

[0078] According to this configuration, it is possible to acquire an announcement AT corresponding to at least one of the first event information E1, the second event information E2, and the third event information E3.

[0079] (3) The generator 22 generates prompt information P1 by combining base information B1, which instructs the creation of an announcement AT, with event information E. The base information B1 is set in advance.

[0080] According to this configuration, since the base information B1 is set in advance, by combining the base information B1 and the event information E, the prompt information P1 can be easily generated.

[0081] (4) The generation unit 22 generates prompt information P1 including objective information G. The objective information G relates to the objective of the loudspeaker broadcast to the broadcast area BA.

[0082] According to this configuration, since the prompt information P1 includes the objective information G, the large-scale language model 24 can acquire an announcement sentence AT that matches the objective of the loudspeaker broadcast.

[0083] (5) The generating unit 22 selects the object information G corresponding to the event information E extracted by the extracting unit 21 from the object database DB1 in which the object information G corresponding to the event information E is stored.

[0084] According to this configuration, the objective information G corresponding to the event information E is selected, so that the objective of the broadcast can be set according to the situation in the broadcast area BA.

[0085] (6) The loudspeaker system 10 includes a sound source generator 13 and a broadcaster 14. The sound source generator 13 generates audio data based on the announcement text AT. The broadcaster 14 loudspeakers broadcast audio based on the audio data to the broadcast area BA.

[0086] According to this configuration, the broadcasting unit 14 can broadcast the sound based on the announcement text AT in the broadcasting area BA.

[0087] (7) The loudspeaker system 10 includes a detection unit 15. The detection unit 15 detects the status of the broadcast area BA. The extraction unit 21 extracts event information E from the status of the broadcast area BA detected by the detection unit 15.

[0088] According to this configuration, it is possible to extract event information E that is appropriate for the situation in the broadcast area BA.

[0089] (8) The detection unit 15 includes at least one of a camera device 16 that captures images of the broadcast area BA and a microphone device 17 that captures audio from the broadcast area BA.

[0090] According to this configuration, the event information E can be extracted from at least one of the video of the broadcast area BA and the audio of the broadcast area BA.

[0091] Second Embodiment A loudspeaker system 10 according to the second embodiment will be described with reference to Figures 1, 2, 12, and 13. In this embodiment, components common to the first embodiment are assigned the same reference numerals as those in the first embodiment, and descriptions of overlapping components will be omitted.

[0092] In the loudspeaker system 10 of the first embodiment, the target information G is selected from the target database DB1. In contrast, in the loudspeaker system 10 of the present embodiment, the target information G is generated by the large-scale language model 24.

[0093] In this embodiment, the event information E is a part of the video or audio of the broadcast area BA. The event information E is, for example, an image of a part of the video.

[0094] The generation unit 22 of this embodiment inputs the situation of the broadcast area BA to the large-scale language model 24, thereby causing the large-scale language model 24 to generate goal information G. The generation unit 22 inputs goal generation prompt information P2 related to the situation of the broadcast area BA to the large-scale language model 24. The goal generation prompt information P2 includes an instruction to cause the large-scale language model 24 to generate goal information G.

[0095] The generation unit 22 generates goal generation prompt information P2 based on the situation in the broadcast area BA. The goal generation prompt information P2 includes goal generation base information B2 and a portion of the video or audio of the broadcast area BA. The goal generation base information B2 includes, for example, text such as "Please explain the situation in the attached image. Please tell me the dangers lurking in the situation in this image." The portion of the video or audio of the broadcast area BA includes an image P, which is an attached image. Image P is an example of event information E. The goal generation base information B2 is set in advance.

[0096] Input 5 in Figure 12 shows an example of goal-generating prompt information P2. Image P in Figure 5 is an example of an image of a child climbing a fence. The goal-generating prompt information P2 in input 5 causes the large-scale language model 24 to generate a situation in image P and a potential danger in the situation in image P.

[0097] As output 5, a list of "dangers lurking in the situation in the image" is generated as a situation description, and objective information G, "encourage the child to get off the fence and play in a safe place."

[0098] 13, the generation unit 22 generates prompt information P1 of input 6 based on the objective information G of output 5 of FIG. 12. The prompt information P1 of input 6 includes objective information G of "encourage the child to get down from the fence and instruct them to play in a safe place." As a result, the acquisition unit 23 acquires the announcement sentence AT of output 6.

[0099] <Effects of the embodiment> The effects of this embodiment will be described. The generation unit 22 inputs the situation of the broadcast area BA into the large-scale language model 24, thereby causing the large-scale language model 24 to generate target information G.

[0100] According to this configuration, the objective information G is generated by the large-scale language model 24, so that the objective of broadcasting can be set appropriately according to the situation in the broadcasting area BA.

[0101] <Third embodiment> A loudspeaker system 10 according to the third embodiment will be described with reference to Figures 2, 14, and 15. In this embodiment, components common to the first embodiment are assigned the same reference numerals as those in the first embodiment, and descriptions of overlapping components will be omitted.

[0102] The loudspeaker system 10 of this embodiment includes, in addition to the configuration of the first embodiment, an effect determining unit 25. The effect determining unit 25 is realized by a program installed in the computer 11.

[0103] The effect determination unit 25 determines the broadcast effect of the loudspeaker broadcast to the broadcast area BA based on the announcement text AT. The broadcast effect is, for example, the degree to which the purpose of the loudspeaker broadcast has been achieved. Specifically, the broadcast effect is whether or not the purpose of the objective information G contained in the prompt information P1 that formed the basis of the announcement text AT for the loudspeaker broadcast has been achieved after the loudspeaker broadcast. The determination of the effectiveness is made, for example, based on the status of the broadcast area BA detected by the detection unit 15.

[0104] In this embodiment, the generation unit 22 generates prompt information P1 based on the broadcast effect. Specifically, after loudspeaker broadcasting based on the announcement text AT to the broadcast area BA, the effect determination unit 25 determines the broadcast effect, and the generation unit 22 generates improved prompt information P1 based on the broadcast effect determined by the effect determination unit 25.

[0105] For example, if a fight is extracted as event information E and "get the parties to stop fighting" is selected as objective information G, the effect determination unit 25 determines whether the parties to the fight stop fighting after the loudspeaker broadcast. If the parties do not stop fighting, the generation unit 22 improves the prompt information P1 by generating prompt information P1 that generates an announcement sentence AT in a style or tone that is thought to be more effective in stopping the fight. A more effective announcement sentence AT is one that more strongly encourages the parties and those around them to act, for example, by using a tone that is more coercive and more likely to elicit sympathy from the other party. The generation of such prompt information P1 is realized by improving the base information B1 to generate an announcement sentence AT that is more effective. For example, the generation unit 22 improves the base information B1 to generate an announcement sentence AT in a stronger tone. For example, the generation unit 22 improves the base information B1 by adding text such as "Please use an imperative tone" or "Please use a stronger tone" to the base information B1. Base information B1 including the above text may be set in advance. The generator 22 uses the improved base information B1 to generate more effective prompt information P1.

[0106] The generation unit 22 may improve the base information B1 by adding a past announcement sentence AT as text to the base information B1. For example, the base information B1 may be improved by adding text to the base information B1 that instructs that the style or tone of the past announcement sentence AT be made stronger, such as "Last time, (past announcement sentence AT) was broadcast by loudspeaker. Please make this loudspeaker speech stronger." Note that the text of the announcement sentence AT broadcast in the past is inserted into (past announcement sentence AT).

[0107] If the parties have stopped fighting, the generation unit 22 does not generate new prompt information P1 based on the improved base information B1, and the loudspeaker system 10 ends the loudspeaker broadcast. If the parties have stopped fighting, the generation unit 22 may generate prompt information P1 including objective information G such as to inform people around that the fight has stopped and to express gratitude to the people around for their cooperation.

[0108] The flow of processing performed in the loudspeaker system 10 of this modified example will be described with reference to Fig. 15. The processing in Fig. 15 is executed by the computer 11.

[0109] In S1, the computer 11 determines whether a specific event has been extracted. When the extraction unit 21 extracts event information E from the situation in the broadcast area BA, the control unit determines that a specific event has been extracted. If the extraction unit 21 has not extracted a specific event, the process of FIG. 15 ends.

[0110] When a specific event is extracted, the computer 11 sets the objective information G and an achievement status in S2. The achievement status is information for determining whether the objective information G has been achieved. The achievement status is set in advance for the objective information G. For example, for the objective information G of "getting the parties to stop fighting," the achievement status set is "no fight is extracted as event information E from the situation in the broadcast area BA."

[0111] In S3, the computer 11 generates an announcement AT based on prompt information P1 including objective information G. The prompt information P1 includes base information B1 corresponding to the event information E extracted in S1. The event information E and objective information G are combined with this base information B1 to generate the prompt information P1.

[0112] In S4, the computer 11 executes loudspeaker broadcasting with the announcement text AT generated in S3 as the broadcast content. In S5, after a predetermined period of time has elapsed, the computer 11 acquires the status of the broadcast area BA from the detection unit 15. The predetermined period of time is set to a period during which the broadcast area BA is considered to be in an attained state after the loudspeaker broadcasting.

[0113] In S6, the computer 11 determines whether the broadcast area BA is in an achieved state based on the objective information G. If the broadcast area BA is in an achieved state, the computer 11 ends the processing of Fig. 15. If the broadcast area BA is not in an achieved state, the computer 11 improves, in S7, the prompt information P1 used to generate the announcement text AT to be broadcast amplified in S4.

[0114] In S7, the prompt information P1 is improved by improving the base information B1 to include instructions for changing the style and tone of the announcement AT to achieve the objective indicated by the objective information G. The computer 11 repeats the process of FIG. 15 from S3 using the prompt information P1 improved in S7. In S3 after S7, an announcement AT is generated based on the improved prompt information P1, and then in S4, a loudspeaker broadcast is carried out using this announcement AT as the broadcast content. Thereafter, in S6, the computer 11 again determines whether the broadcast area BA is in an achieved state. That is, the process of S3 to S6 after S7 is repeated until the broadcast area BA is in an achieved state. The prompt information P1 is improved each time the process is repeated.

[0115] If the broadcast area BA is in an achieved state in S6, the computer 11 may end the processing of FIG. 15 after making a loudspeaker broadcast to thank people nearby who reported the fight.

[0116] <Effects of the embodiment> The effects of this embodiment will be described. The loudspeaker system 10 includes an effect determination unit 25. The effect determination unit 25 determines the broadcasting effect of loudspeaker broadcasting to the broadcast area BA based on the announcement text AT. The generation unit 22 generates improved prompt information P1 based on the broadcasting effect.

[0117] According to this configuration, the prompt information P1 is improved based on the broadcast effect, so that an announcement AT with a higher broadcast effect can be obtained than prompt information P1 that is not based on the broadcast effect.

[0118] <Fourth embodiment> A loudspeaker system 10 according to the fourth embodiment will be described with reference to Figures 1, 2, and 16. In this embodiment, components common to the first embodiment are assigned the same reference numerals as those in the first embodiment, and descriptions of overlapping components will be omitted.

[0119] The loudspeaker system 10 of this embodiment generates a broadcast database DB2. Data linked to objective information G, prompt information P1, and announcement text AT is accumulated in the broadcast database DB2. The control unit of the computer 11 stores data related to loudspeaker broadcasts in the storage unit each time a loudspeaker broadcast is made. The broadcast database DB2 may be generated in the storage unit of the computer 11, or may be generated in an external computer separate from the computer 11. When the broadcast database DB2 is generated in an external computer, the control unit of the computer 11 outputs the data to the external computer each time a loudspeaker broadcast is made. Each piece of loudspeaker broadcast data may be linked to event information E.

[0120] Furthermore, each piece of loudspeaker broadcast data may be linked to information regarding the broadcast effectiveness. The broadcast effectiveness can be determined by the effect determination unit 25 of the third embodiment. The information regarding the broadcast effectiveness is, for example, whether or not the announcement text AT had a broadcast effect. The broadcast database DB2 may include only data regarding the announcement text AT that had a broadcast effect.

[0121] <Effects of the embodiment> The effects of this embodiment will be described. The loudspeaker system 10 generates a broadcast database DB2. The broadcast database DB2 stores data that is linked to the objective information G, the prompt information P1, and the announcement text AT.

[0122] According to this configuration, the broadcast database DB2 can be used as a knowledge base regarding loudspeaker broadcasting.

[0123] <Modification> The above-described embodiments are merely examples of possible forms of the public address system 10 and are not intended to limit the forms. The public address system 10 may take forms different from those illustrated in the above-described embodiments. Examples of such forms include forms in which part of the configuration of each embodiment is replaced, modified, or omitted, or forms in which new configurations are added to each embodiment. Modified examples of each embodiment are shown below.

[0124] In the embodiment, the computer 11 is described as a server, but the computer 11 may be a personal computer, a portable computer, or the like. In this modification, the large-scale language model 24 may be configured in a cloud server connected to the computer 11.

[0125] In the loudspeaker system 10 of the first embodiment, the large-scale language model 24 generates the announcement AT from prompt information P1 containing objective information G. Alternatively, as shown in FIG. 17 , the large-scale language model 24 may generate the announcement AT directly from the situation in the broadcast area BA. The extraction unit 21 extracts a portion of video or audio as event information E from the situation in the broadcast area BA. In the example of FIG. 17 , the portion of video or audio is an image P. The generation unit 22 generates prompt information P1 by combining base information B1 with the image P serving as event information E. In FIG. 17 , the base information B1 is "Please let us know if there is anything we should address in the loudspeaker broadcast to address the problem inherent in the situation in this image." By attaching the image P to the base information B1, the prompt information P1 of input 7 is generated. The large-scale language model 24 outputs the announcement AT as output 7.

[0126] In each embodiment, the camera device 16 and the microphone device 17 are exemplified as the detector 15, but the form is not important as long as the situation in the broadcast area BA can be acquired. For example, an optical sensor, a weight sensor, an infrared sensor, etc. may be used as a sensor for detecting human behavior. A GPS (Global Navigation Satellite System) may be used to acquire the current location of a person, a vehicle, etc.

[0127] At least one of the sound source generation unit 13 and the broadcast unit 14 may be omitted from the loudspeaker system 10. The loudspeaker system 10 of this modified example outputs the generated announcement text AT to an external broadcasting device. The external broadcasting device broadcasts the announcement text AT based on the received announcement text AT. The external broadcasting device may broadcast a voice generated by a voice synthesis engine, or may broadcast a voice of a person reading the announcement text AT.

[0128] The detection unit 15 may be omitted from the loudspeaker system 10. In the loudspeaker system 10 of this modified example, the extraction unit 21 extracts event information E based on the status of the broadcast area BA input by the user. The status of the broadcast area BA input by the user is input as data such as images, video, audio, text, etc. [Explanation of symbols]

[0129] 10...Public address broadcasting system, 13...Sound source generation unit, 14...Broadcasting unit, 15...Detection unit, 16...Camera device, 17...Microphone device, 21...Extraction unit, 22...Generation unit, 23...Acquisition unit, 24...Large-scale language model, 25...Effect determination unit, DB1...Objective database, DB2...Broadcast database.

Claims

1. A loudspeaker system for loudly broadcasting audio in a broadcast area, comprising: an extraction unit that extracts event information from the status of the broadcast area; a generating unit that generates prompt information instructing the creation of an announcement to be broadcast to the broadcast area based on the event information; an acquisition unit that acquires the announcement sentence generated by a large-scale language model by inputting the prompt information into the large-scale language model.

2. the event information includes at least one of first event information, second event information, and third event information; the first event information is information about an event that occurred in the broadcast area, the second event information is information about attributes of people involved in the event, 2. The loudspeaker system according to claim 1, wherein the third event information is information relating to at least one of a location of the broadcast area and a broadcast time.

3. the generation unit generates the prompt information by combining the event information with base information that instructs creation of the announcement text; The public address system according to claim 1 , wherein the base information is preset.

4. The loudspeaker system according to claim 1 , wherein the generating unit generates the prompt information including purpose information relating to a purpose of loudspeaker broadcasting to the broadcast area.

5. 5. The loudspeaker system according to claim 4, wherein the generating unit selects the object information corresponding to the event information extracted by the extracting unit from an object database in which the object information corresponding to the event information is stored.

6. The loudspeaker system according to claim 4 , wherein the generation unit inputs a situation of the broadcast area into the large-scale language model, thereby causing the large-scale language model to generate the target information.

7. an effect determination unit that determines a broadcast effect of loudspeaker broadcasting to the broadcast area based on the announcement sentence; The loudspeaker system according to claim 4 , wherein the generating unit generates the prompt information based on the broadcast effect.

8. The loudspeaker system according to claim 4 , further comprising: a broadcast database that stores data linked to the objective information, the prompt information, and the announcement text.

9. a sound source generating unit that generates voice data based on the announcement; The loudspeaker system according to claim 1 , further comprising: a broadcasting unit that loudly broadcasts audio based on the audio data to the broadcast area.

10. a detection unit for detecting a status of the broadcast area; The loudspeaker system according to claim 1 , wherein the extracting unit extracts the event information from the state of the broadcast area detected by the detecting unit.

11. The loudspeaker system according to claim 10, wherein the detection unit includes at least one of a camera device that captures images of the broadcast area and a microphone device that captures audio in the broadcast area.

Images