Announcement sound source generation system, announcement sound source generation method, and announcement sound source generation program

Abstract

This invention provides an announcement audio source generation system, an announcement audio source generation method, and an announcement audio source generation program that can generate a single audio source data containing multiple announcement sentences. [Solution] The announcement sound source generation system comprises an announcement text generation unit and a sound source generation unit. The announcement text generation unit generates an announcement text AN which includes a first announcement text A1 and one or more second announcement texts A2. The sound source generation unit generates one sound source data DM which is an audio playback signal of the announcement text AN, based on source information including the announcement text AN.

Description

Technical Field

[0001] The present disclosure relates to an announcement sound source generation system, an announcement sound source generation method, and an announcement sound source generation program.

Background Art

[0002] An apparatus is known that generates announcement information including a plurality of pieces of information to be announced to a user. Examples of announcement information containing a plurality of pieces of information include multilingual announcement information. The multilingual announcement information includes a plurality of pieces of announcement information that have the same content but are in different languages. The multilingual broadcasting apparatus described in Patent Document 1 stores a plurality of pieces of announcement information in different languages. Then, the multilingual broadcasting apparatus rearranges the plurality of pieces of announcement information into a predetermined order based on a predetermined rule, and outputs the result as broadcast data.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When broadcasting a plurality of pieces of announcement information on a broadcasting apparatus, a continuous playback program for sound source data may be formed so that the plurality of pieces of announcement information are broadcast continuously. In addition, the sound source data related to the plurality of pieces of announcement information may be formed into a single piece of sound source data. Since the volume of the announcer's voice may differ across the plurality of sound source data, fine adjustment of each individual sound source data may be performed when creating the continuous playback program or the single piece of sound source data, so as not to cause discomfort to the listener. However, such work requires knowledge to operate the broadcasting apparatus because the broadcasting apparatus needs to be used, and also consumes time and labor.

Means for Solving the Problems

[0005] (1) An announcement sound source generation system that solves the problem comprises an announcement sentence generation unit that generates an announcement sentence including a first announcement sentence and one or more second announcement sentences, and a sound source generation unit that generates one sound source data which is an audio playback signal of the announcement sentence based on source information including the announcement sentence.

[0006] This configuration allows for the generation of a single audio file containing multiple announcements. Therefore, it eliminates the need to create a program for continuous playback of audio files to announce multiple pieces of information. Alternatively, it reduces the effort required to combine multiple audio files into a single one. In this way, it reduces the workload and effort involved when announcing multiple items.

[0007] (2) In the announcement sound source generation system described in (1) above, the source information includes the announcement text and sound data, and the sound source generation unit has a first processing unit that generates announcement voice data based on the announcement text and a second processing unit that generates one sound source data by synthesizing the announcement voice data and at least one of the sound data, and the sound data includes BGM data and sound effect data.

[0008] This configuration allows for the generation of audio data that combines announcement voice with background music or sound effects.

[0009] (3) In the announcement sound source generation system described in (2) above, the first processing unit forms the announcement voice data based on the announcement text and the pronunciation information to which the announcement text is added, and the pronunciation information includes intonation information and accent information.

[0010] This configuration allows for the generation of announcement voices with different intonations or accents from standard voices. This results in announcement voices that sound less unnatural.

[0011] (4) In the announcement sound source generation system described in (1) above, the announcement sentence generation unit has a first generation unit and a linking unit, wherein the first generation unit generates the first announcement sentence based on output information output from the large-scale language model by inputting first prompt information for forming the first announcement sentence into the large-scale language model, the first prompt information includes announcement source information which is the source information for generating the first announcement sentence, the announcement source information includes purpose information indicating the purpose of the announcement and announcement detail information indicating the detailed content of the announcement, and the linking unit links the first announcement sentence and the second announcement sentence. With this configuration, an announcement sentence can be generated based on the purpose and detailed content of the announcement.

[0012] (5) In the announcement sound source generation system described in (4) above, the announcement sentence generation unit further comprises a second generation unit, the second generation unit generates the second announcement sentence based on output information output from the sentence formation device by inputting second prompt information for forming the second announcement sentence based on the first announcement sentence to the sentence formation device.

[0013] With this configuration, a second announcement sentence can be generated based on a first announcement sentence. Therefore, an announcement sound source generation system having a first generation unit and a second generation unit can form sound source data including announcement audio related to the first announcement sentence and the second announcement sentence by inputting the purpose and detailed content of the announcement.

[0014] (6) In the announcement sound source generation system described in (5) above, the second announcement sentence is information having substantially the same content as the first announcement sentence and expressed in a language different from the language of the first announcement sentence, the sentence formation device includes a large-scale language model, and the second prompt information includes summarization command information that instructs to summarize the first announcement sentence and translation command information that instructs to translate the summarized sentence of the first announcement sentence into a specified language.

[0015] According to large-scale language models, when a large-scale language model is instructed to translate a sentence, it may produce a direct translation of the original first announcement sentence, resulting in a translation that is unsuitable for announcements. In this respect, the above configuration addresses this issue by summarizing the first announcement sentence. Then, the summarized version of the first announcement sentence is translated. This prevents the generation of an unsuitable second announcement sentence (i.e., a translated sentence) from the first announcement sentence A1.

[0016] (7) An announcement sound source generation method for solving the problem is an announcement sound source generation method for generating sound source data, comprising: an announcement sentence generation step in which the system generates an announcement sentence including a first announcement sentence and one or more second announcement sentences; and a sound source generation step in which the system generates one sound source data which is an audio playback signal of the announcement sentence based on source information including the announcement sentence.

[0017] This configuration allows for the generation of a single audio file containing multiple announcement messages. Therefore, it eliminates the need to create a program for continuous playback of audio files to announce multiple pieces of information. Alternatively, it reduces the effort required to combine multiple audio files into a single one. In this way, it reduces the workload and effort involved in announcing multiple items.

[0018] (8) An announcement sound source generation program for solving the problem is an announcement sound source generation program that generates sound source data, the program comprising: an announcement sentence generation step of causing a computer to generate an announcement sentence including a first announcement sentence and one or a plurality of second announcement sentences; and a sound source generation step of causing the computer to generate one piece of sound source data that is an audio playback signal of the announcement sentence, based on source information including the announcement sentence.

[0019] According to this configuration, it is possible to generate one piece of sound source data including a plurality of announcement sentences. Therefore, work such as creating a continuous playback program for sound source data to announce a plurality of pieces of announcement information can be omitted. Alternatively, the effort of synthesizing a plurality of pieces of sound source data into one piece of sound source data can be reduced. In this way, the work or effort required when announcing a plurality of items can be reduced. Effects of the Invention

[0020] According to the announcement sound source generation system, the announcement sound source generation method, and the announcement sound source generation program, it is possible to generate one piece of sound source data including a plurality of announcement sentences. Brief Description of the Drawings

[0021] [Figure 1] It is a block diagram of an announcement sound source generation system. [Figure 2] It is a block diagram of an announcement sentence generation unit in Figure 1. [Figure 3] It is a block diagram of first prompt information. [Figure 4] It is a diagram of a screen of an input / output interface for acquiring announcement source information in Figure 3. [Figure 5] It is a diagram of a screen of an input / output interface that displays a first announcement sentence. [Figure 6] It is a block diagram of second prompt information. [Figure 7] It is a block diagram of a sound source generation unit in Figure 1. [Figure 8]This is a schematic diagram of the sound source data. [Figure 9] This is a diagram showing a modified version of the input / output interface screen. [Figure 10] This is a diagram of the interface screen for changing the broadcast order. [Figure 11] This is a diagram of the translation interface screen. [Modes for carrying out the invention]

[0022] Referring to Figures 1 to 8, the announcement sound source generation system 1 according to the first embodiment will be described. The announcement sound source generation system 1 generates sound source data DM.

[0023] In this embodiment, the sound source data DM is the data of the sound source signal input to the broadcasting device. The broadcasting device broadcasts within the broadcasting area. Examples of broadcasting areas include areas set within a building including facilities, outdoor areas outside buildings on the premises, and wide-area broadcasting areas such as disaster prevention radio. The announcement includes various sounds and announcement voice related to the broadcast content. The broadcast content includes notices, weather information, greetings, news, etc. Notices include directions within the broadcasting area, event announcements, product announcements, lost child notices, disaster-related notices, event start times, and delay notices. The announcement text AN is a text corresponding to the broadcast content.

[0024] Announcements broadcast in a series of programs may contain multiple pieces of content. For example, an announcement may include a first announcement in the local language of the broadcasting region and a second announcement in another language with the same content. In another example, an announcement may include a greeting and an event announcement. Conventionally, announcements containing multiple pieces of audio are formed by combining multiple audio data. However, combining multiple audio data requires a certain level of knowledge in handling broadcasting equipment and the effort required to adjust the audio data. Therefore, we propose an announcement audio source generation system 1 that can easily generate a single audio source data DM containing multiple pieces of audio.

[0025] [Announcement Sound Source Generation System 1] The announcement sound source generation system 1 generates one sound source data DM containing multiple announcements. In this embodiment, the announcement sound source generation system 1 generates sound source data DM containing the audio related to the first announcement sentence A1 and the audio related to the second announcement sentence A2.

[0026] As shown in Figure 1, the announcement sound source generation system 1 comprises an announcement text generation unit 10 and a sound source generation unit 20.

[0027] [Announcement text generation unit 10] The announcement text generation unit 10 generates an announcement text AN which includes a first announcement text A1 and one or more second announcement texts A2. The second announcement text A2 may be a sentence related to the first announcement text A1, or it may be a sentence unrelated to the first announcement text A1.

[0028] In this embodiment, the second announcement A2 is related to the first announcement A1. Specifically, the second announcement A2 has substantially the same content as the first announcement A1, but is information expressed in a language different from the language of the first announcement A1. In short, the second announcement A2 is a translation of the first announcement A1 in a foreign language. Therefore, the announcement audio in this embodiment includes announcement audio in the language of the broadcasting region and announcement audio in a foreign language.

[0029] As shown in Figure 2, the announcement text generation unit 10 includes a first generation unit 11 and a connecting unit 13. The first generation unit 11 inputs the first prompt information P1 to the large-scale language model LLM and generates the first announcement sentence A1 based on the output information output from the large-scale language model LLM. The large-scale language model LLM is located in a data center 2 on the internet. The large-scale language model LLM may also be located on a server on a local network.

[0030] Examples of large-scale language models (LLMs) include ChatGPT®, Bard®, LaMDA®, LLaMA®, tsuzumi®, and CyberAgentLM®. Large-scale language models (LLMs) do not necessarily have to be hosted on the cloud; they can also be configured by the user themselves.

[0031] The first prompt information P1 is information for forming the first announcement sentence A1. In this embodiment, the first prompt information P1 is instruction information issued to the large-scale language model LLM.

[0032] As shown in Figure 3, the first prompt information P1 includes announcement text generation instruction information PA. Announcement text generation instruction information PA is information that instructs the large-scale language model LLM to create an announcement text suitable for broadcast. Announcement text generation instruction information PA includes the instruction information: "You are an assistant in creating announcement texts for broadcast," and "Generate an announcement text based on user input."

[0033] The first prompt information P1 further includes the source information PB, which is the source information for generating the first announcement sentence A1. In this embodiment, the source information PB is input to the large-scale language model LLM together with the announcement sentence generation instruction information PA.

[0034] As shown in Figure 4, the announcement source information PB includes purpose information NA indicating the purpose of the announcement and announcement detail information NB indicating the detailed content of the announcement. The announcement source information PB may also include location information NC regarding the location where the announcement will be made. The location information NC, purpose information NA, and announcement detail information NB are information input by the user via the input / output interface 30.

[0035] Figure 5 shows an example of the first announcement statement A1 generated based on the announcement source information PB. As shown in Figure 5, the first announcement statement A1 is displayed on the input / output interface 30. The first announcement statement A1 contains the contents of the announcement source information PB.

[0036] The announcement text generation unit 10 may also include an editing unit 14 that accepts editing of the first announcement text A1. In this case, the user can edit the first announcement text A1 at the input / output interface 30. The user can obtain a first announcement text A1 suitable for broadcasting by inputting the detailed content and purpose of the announcement to the announcement sound source generation system 1.

[0037] As shown in Figure 2, the announcement sentence generation unit 10 may further include a second generation unit 12. The second generation unit 12 generates a second announcement sentence A2 based on the output information output from the sentence formation device 3 by inputting second prompt information P2 to the sentence formation device 3. The second prompt information P2 is information for forming the second announcement sentence A2 based on the first announcement sentence A1. In this embodiment, the second announcement sentence A2 is a translation of the first announcement sentence A1.

[0038] In this embodiment, the sentence formation device 3 includes a large-scale language model (LLM). The sentence formation device 3 may be configured as a device that executes a translation program and a summarization program. The translation program may be composed of commercially available software or cloud-based software. The summarization program may be composed of commercially available software or cloud-based software.

[0039] As shown in Figure 6, the second prompt information P2 includes summary command information PC and translation command information PD. The summary command information PC is a command instructing to summarize the first announcement sentence A1. The translation command information PD is a command instructing to translate the summary of the first announcement sentence A1 into a specified language.

[0040] The second generation unit 12 obtains the specified language specified by the user via the first selection unit 31 of the input / output interface 30. Based on the obtained specified language and pre-set information, the second generation unit 12 forms translation command information PD.

[0041] The linking unit 13 links the first announcement sentence A1 and the second announcement sentence A2. The linking unit 13 may change the order in which the first announcement sentence A1 and the second announcement sentence A2 are linked based on a linking order command. The linking order command is formed based on user operations via the input / output interface 30. In one example, the reception of a linking order command by the user is configured as a linking order input box (not shown) provided on the input / output interface 30. A number indicating the linking order is entered into the linking order input box.

[0042] [Input / Output Interface 30] The input / output interface 30 is connected to the announcement sound source generation system 1.

[0043] As shown in Figure 5, the input / output interface 30 includes a first selection unit 31, a second selection unit 32, a third selection unit 33, a fourth selection unit 34, a playback unit 35, and a display unit 36.

[0044] The display unit 36 ​​displays the first announcement sentence A1 generated by the first generation unit 11. The display unit 36 ​​may display the announcement sentence AN generated by the first generation unit 11 and the second generation unit 12. The display unit 36 ​​accepts whether or not each character has an accent. In one example, the display unit 36 ​​displays the position of the accent with an accent mark MA. The accent position information in the first announcement sentence A1 is output to the sound source generation unit 20 as pronunciation information PE.

[0045] The first selection unit 31 is a reception unit that accepts the specification of the language for the translation of the first announcement text A1. The first selection unit 31 is configured to display multiple languages. The second selection unit 32 is an operation unit that accepts the specification of sound effects. The second selection unit 32 is configured to display multiple sound effects. The third selection unit 33 is an operation unit that accepts the specification of background music (BGM). The third selection unit 33 is configured to display multiple BGM tracks. The fourth selection unit 34 is an operation unit that accepts the specification of intonation. The fourth selection unit 34 is configured to add accent marks MA and to allow the accent marks MA to be moved by pointer operation or cursor keys. By adding accents and adjusting the accent positions, intonations that sound unnatural in the synthesized speech mechanically generated by the sound source generation unit 20 (described later) can be corrected. By adding accents and adjusting the accent positions, the announcement voice generated by the sound source generation unit 20 can be made closer to the voice of a human announcer.

[0046] The playback unit 35 is an operation unit that receives commands for playback and stopping of the sound source data DM. The user can check the audio of the sound source data DM by operating the playback unit 35.

[0047] [Sound source generation section 20] The sound source generation unit 20 generates one sound source data DM, which is the audio playback signal of the announcement sentence AN, based on source information including the announcement sentence AN. The source information includes the announcement sentence AN and the sound data.

[0048] As shown in Figure 7, the sound source generation unit 20 has a first processing unit 21 and a second processing unit 22. The first processing unit 21 generates announcement voice data DA based on the announcement sentence AN. In this case, announcement voice data DA pronounced with a standard intonation is generated.

[0049] The first processing unit 21 may form announcement voice data DA based on the announcement sentence AN and the pronunciation information PE to which the announcement sentence AN is added. The pronunciation information PE includes intonation information and accent information. In this case, announcement voice data DA is generated that is pronounced with a regional intonation different from the standard intonation.

[0050] The second processing unit 22 generates one sound source data DM by combining the announcement voice data DA with at least one sound data. The sound data includes BGM data DB and sound effect data DC.

[0051] The second processing unit 22 obtains the sound effect specified by the user from the sound effect server 4 via the second selection unit 32 of the input / output interface 30. The sound effect server 4 may be a server connected to the internet or a server located on a local network. The sound effect server 4 has a sound effect library.

[0052] The second processing unit 22 obtains the BGM specified by the user from the BGM server 5 via the third selection unit 33 of the input / output interface 30. The BGM server 5 may be a server connected to the internet or a server located on a local network. The BGM server 5 has a BGM library. The second processing unit 22 may also obtain BGM created by a large-scale language model (LLM).

[0053] The second processing unit 22 overlays the background music data DB onto the announcement voice data DA. Furthermore, the second processing unit 22 adds sound effect data DC to the head and end of the combined data of the announcement voice data DA and the background music data DB. In this way, the sound source data DM is formed.

[0054] Figure 8 is a schematic diagram of the sound source data DM. In this example, the length of the BGM data DB is greater than the length of the announcement voice data DA. The head region of the BGM data DB is located temporally before the head of the announcement voice data DA. The end region of the BGM data DB is located temporally after the end of the announcement voice data DA.

[0055] <First Action of the Embodiment> Sometimes, multiple announcements are broadcast in a sequence. For example, in a store, store information might be broadcast after a greeting to customers. Since store information changes daily, the broadcast program had to be restructured each time it was broadcast. This restructuring work is time-consuming. Also, when unexpected events occur, other announcements may be added to the planned announcements. For example, if an additional sale is to be held, a pre-prepared announcement announcing the additional sale is added after the planned announcement. In this case, the additional announcement is added by editing the broadcast program. However, this kind of work requires a certain level of knowledge to operate the broadcast program.

[0056] The announcement text generation unit 10 generates an announcement text AN which includes a first announcement text A1 and one or more second announcement texts A2. In other words, the announcement text generation unit 10 generates an announcement text AN which includes multiple announcement contents. The sound source generation unit 20 generates one sound source data DM based on the announcement text AN. This makes it possible to broadcast multiple announcement contents by inputting only one sound source data DM to the broadcasting device.

[0057] <Second action of the embodiment> The announcement text generation unit 10 further includes a second generation unit 12. The second generation unit 12 inputs second prompt information P2 to the sentence formation device 3 for forming a second announcement text A2 based on the first announcement text A1. The second generation unit 12 then generates the second announcement text A2 based on the output information output from the sentence formation device 3. Thus, the announcement text generation unit 10 can form a second announcement text A2 related to the first announcement text A1. For example, the announcement text generation unit 10 can form a translation of the first announcement text A1. Therefore, the announcement sound source generation system 1 can generate sound source data DM including the first announcement text A1 and the second announcement text A2 as a translation of the first announcement text A1. Therefore, by using the announcement sound source generation system 1, the user can manage the sound source data DM, which includes the announcement content and the announcement content translated into a foreign language, as a single file data by simply confirming the first announcement sentence A1.

[0058] <Effects of the Embodiment> (1) The announcement sound source generation system 1 comprises an announcement text generation unit 10 and a sound source generation unit 20. The announcement text generation unit 10 generates an announcement text AN which includes a first announcement text A1 and one or more second announcement texts A2.

[0059] This configuration allows for the generation of a single audio data DM containing multiple announcements. Therefore, it eliminates the need to create a continuous playback program for audio data DMs to announce multiple pieces of information. Alternatively, it reduces the effort required to combine multiple audio data DMs into a single one. In this way, it reduces the workload and effort involved when announcing multiple items.

[0060] (2) The sound source generation unit 20 has a second processing unit 22. The second processing unit 22 generates one sound source data DM by synthesizing the announcement voice data DA with at least one sound data. With this configuration, it is possible to generate sound source data DM which is a synthesis of the announcement voice and background music or sound effects.

[0061] (3) The first processing unit 21 forms announcement voice data DA based on the announcement sentence AN and the pronunciation information PE to which the announcement sentence AN is added. With this configuration, it is possible to generate announcement voices with different intonations or accents from standard voices. This makes it possible to generate announcement voices that sound less unnatural.

[0062] (4) The announcement text generation unit 10 has a first generation unit 11. The first generation unit 11 inputs first prompt information P1 for forming a first announcement text A1 to the large-scale language model LLM. The first generation unit 11 then generates the first announcement text A1 based on the output information output from the large-scale language model LLM. The first prompt information P1 includes announcement source information PB. The announcement source information PB includes purpose information NA indicating the purpose of the announcement and announcement detail information NB indicating the detailed content of the announcement. With this configuration, an announcement text AN can be generated based on the purpose and detailed content of the announcement.

[0063] (5) The announcement sentence generation unit 10 further includes a second generation unit 12. The second generation unit 12 inputs second prompt information P2 to the sentence formation device 3 for forming a second announcement sentence A2 based on the first announcement sentence A1. The second generation unit 12 then generates the second announcement sentence A2 based on the output information output from the sentence formation device 3.

[0064] With this configuration, a second announcement sentence A2 can be generated based on a first announcement sentence A1. Therefore, with the announcement sound source generation system 1 having a first generation unit 11 and a second generation unit 12, sound source data DM including announcement audio related to the first announcement sentence A1 and the second announcement sentence A2 can be formed by inputting the purpose and detailed content of the announcement.

[0065] (6) The second prompt information P2 includes summary command information PC which instructs to summarize the first announcement sentence A1, and translation command information PD which instructs to translate the summary of the first announcement sentence A1 into a specified language.

[0066] According to the Large-Scale Language Model (LLM), when a sentence is instructed to be translated, it may produce a direct translation of the original first announcement sentence A1, resulting in a sentence unsuitable for announcement. In contrast, with the above configuration, the first announcement is summarized. Then, the summarized version of the first announcement sentence A1 is translated. This prevents the generation of an unsuitable second announcement sentence A2 from the first announcement sentence A1.

[0067] (7) This embodiment discloses the following method for generating an announcement sound source. The announcement sound source generation method includes an announcement text generation step and a sound source generation step. In the announcement text generation step, the system generates an announcement text AN which includes a first announcement text A1 and one or more second announcement texts A2. In the sound source generation step, the system generates one sound source data DM which is the audio playback signal of the announcement text AN, based on source information including the announcement text AN.

[0068] The system may be an announcement sound source generation system 1, a general-purpose computer, or a workstation. Each step in the announcement sound source generation method may be performed by a separate computer. The computer includes a personal computer, a portable computer, a server, or a general-purpose computer. A portable computer includes a notebook computer and a smartphone. The computer may also be a control device incorporated into broadcasting equipment, etc. The control device has a CPU (Central Processing Unit) or an MPU (Micro Processing Unit) and a storage unit. The storage unit includes semiconductor memory, a hard disk, an optical disk, and a magnetic medium.

[0069] This configuration allows for the generation of a single audio data DM containing multiple announcements. Therefore, it eliminates the need to create a continuous playback program for audio data DMs to announce multiple pieces of information. Alternatively, it reduces the effort required to combine multiple audio data DMs into a single one. In this way, it reduces the workload and effort involved when announcing multiple items.

[0070] (8) This embodiment discloses the following announcement sound source generation program. The announcement sound source generation program includes an announcement text generation step and a sound source generation step. In the sound source generation step, the announcement sound source generation program causes the computer to generate an announcement text AN, which includes a first announcement text A1 and one or more second announcement texts A2. In the sound source generation step, the announcement sound source generation program causes the computer to generate one sound source data DM, which is an audio playback signal of the announcement text AN, based on source information including the announcement text AN.

[0071] Here, the definition of a computer is the same as the definition of a computer described in the announcement sound source generation method. Each step in the announcement sound source generation program may be executed by separate computers. In this case, the announcement sound source generation program causes the separate computers to cooperate in generating the target sound source data DM.

[0072] This configuration allows for the generation of a single audio data DM containing multiple announcements. Therefore, it eliminates the need to create a continuous playback program for audio data DMs to announce multiple pieces of information. Alternatively, it reduces the effort required to combine multiple audio data DMs into a single one. In this way, it reduces the workload and effort involved when announcing multiple items.

[0073] <Variation> The embodiments described above are examples of possible forms of the announcement sound source generation system 1, the announcement sound source generation method, and the announcement sound source generation program, and are not intended to limit their forms. The announcement sound source generation system 1, the announcement sound source generation method, and the announcement sound source generation program may take forms different from those exemplified in the embodiments above. Examples of such forms include forms in which some of the configurations of the embodiments are replaced, modified, or omitted, or forms in which new configurations are added to the embodiments. Modifications of the embodiments are shown below.

[0074] The announcement sound source generation system 1 may cause a computer to generate a first announcement sentence A1 in a rule-based manner. Then, a second announcement sentence A2 may be generated based on the first announcement sentence A1 generated in a rule-based manner.

[0075] The second announcement, A2, may not be directly related to the first announcement, A1. For example, the first announcement, A1, may be a message guiding people to a store, while the second announcement, A2, may be a message reporting a lost child. Alternatively, the first announcement, A1, may be a greeting, while the second announcement, A2, may be a message guiding people to a store. Alternatively, the first announcement, A1, may be a message guiding people to a store, while the second announcement, A2, may be a message reporting the weather conditions.

[0076] The input / output interface 30 may include a fifth selection unit (not shown) that can select a dialect intonation. The fifth selection unit is configured to display multiple types of dialect intonations on the display unit 36. The types of dialect intonations are defined by local place names. Information related to the specification of dialect intonation is output to the sound source generation unit 20 as pronunciation supplement information PE. This makes it possible to generate sound source data DM that includes announcement voices that match the local intonation for local broadcasts.

[0077] In this embodiment, examples of broadcast content include announcements, weather information, greetings, and news. When the first announcement text A1 is news, the announcement sound source generation system 1 generates the first announcement text A1 based on news information published on the internet or provided by news providers. Furthermore, the announcement sound source generation system 1 may perform news summarization processing, which involves having a large-scale language model (LLM) summarize the news information. The announcement sound source generation system 1 then uses the news summary generated by the news summarization processing as the first announcement text A1. Furthermore, the announcement sound source generation system 1 generates a second announcement text A2, which is a translation of the first announcement text A1, based on the first announcement text A1, through the processing shown in the embodiment. The subsequent processing is the same as in the embodiment.

[0078] • When the first announcement text A1 is designated as the first news item, the second announcement text A2 may be designated as the second news item related to the first news item. In this case, the announcement sound source generation system 1 generates the first announcement text A1 based on news information published on the internet or provided by news providers. Similarly, the announcement sound source generation system 1 generates the second announcement text A2 based on news information published on the internet or provided by news providers. The announcement sound source generation system 1 may also have a large-scale language model (LLM) extract news related to the first news item for the purpose of obtaining the second news item.

[0079] Figure 9 shows a modified example of the input / output interface 30. The input / output interface 30 shown in Figure 9 has a first tab screen 40 and is configured to allow the addition of an additional tab screen 41. The first tab screen 40 displays the base announcement text A3. The additional tab screen 41 is added by operating the additional operation tab 42 so that it visually overlaps the first tab screen 40. The additional operation tab 42 is configured to accept the language to which the base announcement text A3 will be translated when the additional operation tab 42 is selected. The additional announcement text (not shown) displayed on the additional tab screen 41 is a translation of the base announcement text A3. The additional announcement text is generated by the second generation unit 12 based on the base announcement text A3. The first tab screen 40 and the additional tab screen 41 are each provided with a sound source availability selection unit 45. The sound source selection unit 45 accepts the selection of whether or not to convert the base announcement text A3 or additional announcement text displayed on the first tab screen 40 and the additional tab screen 41 into sound. The linking unit 13 defines one of the group of base announcement text A3 or additional announcement texts selected to be converted into sound as the first announcement text A1, and defines one or more of the other as the second announcement text A2. In this way, the linking unit 13 sets the first announcement text A1 and the second announcement text A2. According to the sound source selection unit 45, instead of making the base announcement text A3 displayed on the first tab screen 40 the first announcement text A1, it is possible to make an additional announcement text the first announcement text A1. This allows, for example, when the base announcement text A3 is formed in Japanese, an English sentence to be defined as the first announcement text A1, and a Chinese sentence to be defined as the second announcement text A2. Therefore, audio data DM consisting of a first announcement sentence A1 in English and a second announcement sentence A2 in Chinese can be formed. In this way, the audio source selection unit 45 can form audio data DM from multiple additional announcement sentences without converting the base announcement sentence A3 into an audio source.

[0080] As shown in Figure 10, the announcement sound source generation system 1 may have a broadcast order change interface 50 for changing the broadcast order. The broadcast order change interface 50 displays a list of icons 51 for the first announcement sentence A1 and icons 52 for the second announcement sentence A2. The list of icons 51 for the first announcement sentence A1 and icons 52 for the second announcement sentence A2 are arranged from top to bottom in broadcast order. The broadcast order is equal to the order in which the first announcement sentence A1 is linked to one or more second announcement sentences A2. The broadcast order change interface 50 is configured as a GUI (Graphical User Interface). The icons 51 for the first announcement sentence A1 and icons 52 for the second announcement sentence A2 are configured to be moved on the screen by drag and drop using pointer operation. The linking unit 13 links the first announcement sentence A1 and the second announcement sentence A2 in the order of the icons 51 and 52 of the first announcement sentence A1 and second announcement sentence A2, respectively, as rearranged by the broadcast order change interface 50. Each icon 51, 52 may be provided with a sound source availability selection unit 45 and a deletion unit 53. The deletion unit 53 deletes the data of the first announcement sentence A1 or second announcement sentence A2 itself that has been moved to the deletion unit 53 by drag and drop.

[0081] As shown in Figure 11, the announcement sound source generation system 1 may have a translation interface 60. The translation interface 60 shows a first announcement sentence A1 and a second announcement sentence A2, which is the translation of the first announcement sentence A1. Furthermore, the translation interface 60 shows a back translation obtained by translating the second announcement sentence A2 into the language of the first announcement sentence A1. The back translation is generated by the second generation unit 12.

[0082] This disclosure discloses the following technologies. Appendix 2 relates to the technology of the input / output interface 30 shown in Figure 9. Appendix 1 is the technology that underlies the technology of Appendix 2. [Note 1] Note 1 is an announcement sound source generation system. The announcement sound source generation system comprises an announcement sentence generation unit that generates an announcement sentence including a first announcement sentence and one or more second announcement sentences, and a sound source generation unit that generates one sound source data which is an audio playback signal of the announcement sentence based on source information including the announcement sentence. [Note 2] In the announcement sound source generation system described in Appendix 1, the announcement text generation unit comprises a first generation unit, a linking unit, and a second generation unit. The first generation unit generates the base announcement sentence based on the output information output from the large-scale language model by inputting first prompt information for forming the base announcement sentence into the large-scale language model. The second generation unit generates the additional announcement sentence based on the output information output from the sentence formation device by inputting second prompt information for forming the additional announcement sentence based on the base announcement sentence to the sentence formation device. The first prompt information includes the source information, which is the original information from which the base announcement statement was generated. The aforementioned announcement source information includes purpose information indicating the purpose of the announcement and announcement detail information indicating the detailed content of the announcement. The concatenation unit defines one selected from the group including the base announcement statement and the additional announcement statement as the first announcement statement, defines another selected from the group as the second announcement statement, and concatenates the first announcement statement and the second announcement statement. [Explanation of Symbols]

[0083] A1...First announcement text, A2...Second announcement text, AN...Announcement text, DA...Announcement audio data, DB...BGM data, DC...Sound effect data, DM...Sound source data, LLM...Large-scale language model, NA...Purpose information, NB...Announcement details information, P1...First prompt information, P2...Second prompt information, PB...Announcement source information, PC...Summary command information, PD...Translation command information, 1...Announcement sound source generation system, 3...Sentence formation device, 10...Announcement text generation unit, 11...First generation unit, 12...Second generation unit, 13...Connecting unit, 20...Sound source generation unit, 21...First processing unit, 22...Second processing unit.

Claims

1. An announcement statement generation unit that generates an announcement statement including a first announcement statement and one or more second announcement statements, The system includes a sound source generation unit that generates one sound source data which is an audio playback signal of the announcement text based on source information including the announcement text, Announcement sound source generation system.

2. The aforementioned source information includes the announcement text and sound data. The sound source generation unit is A first processing unit that generates announcement voice data based on the aforementioned announcement text, The system includes a second processing unit that generates one sound source data by combining the aforementioned announcement voice data with at least one of the aforementioned sound data, The aforementioned sound data includes background music data and sound effect data. The announcement sound source generation system according to claim 1.

3. The first processing unit forms the announcement voice data based on the announcement text and the pronunciation information to which the announcement text is added. The aforementioned pronunciation information includes intonation information and accent information. The announcement sound source generation system according to claim 2.

4. The announcement text generation unit comprises a first generation unit and a connecting unit. The first generation unit generates the first announcement sentence based on the output information output from the large-scale language model by inputting first prompt information for forming the first announcement sentence into the large-scale language model. The first prompt information includes the source information, which is the source information for generating the first announcement statement. The aforementioned announcement source information includes purpose information indicating the purpose of the announcement and announcement detail information indicating the detailed content of the announcement. The connecting portion connects the first announcement sentence and the second announcement sentence. The announcement sound source generation system according to claim 1.

5. The aforementioned announcement text generation unit further has a second generation unit, The second generation unit generates the second announcement sentence based on the output information output from the sentence formation device by inputting second prompt information for forming the second announcement sentence based on the first announcement sentence to the sentence formation device. The announcement sound source generation system according to claim 4.

6. The second announcement statement has substantially the same content as the first announcement statement, and is information expressed in a language different from the language of the first announcement statement. The sentence formation apparatus includes a large-scale language model, The second prompt information includes summary command information instructing to summarize the first announcement sentence and translation command information instructing to translate the summary of the first announcement sentence into a specified language. The announcement sound source generation system according to claim 5.

7. A method for generating announcement sound sources that generates sound source data, The system includes an announcement statement generation step which generates an announcement statement that includes a first announcement statement and one or more second announcement statements, The system includes a sound source generation step of generating one sound source data which is an audio playback signal of the announcement text based on source information including the announcement text. Method for generating announcement audio.

8. An announcement sound source generation program that generates sound source data, An announcement statement generation step in which a computer generates an announcement statement that includes a first announcement statement and one or more second announcement statements, The process includes a sound source generation step of causing the computer to generate one sound source data which is an audio playback signal of the announcement text based on source information including the announcement text, Announcement sound source generation program.

Images