JP6860472B2 - How to create a summary from meeting audio data - Google Patents

Description

The present invention relates to a method of creating a summary from audio data of a conference.

As a background technique of the present disclosure, for example, Japanese Patent Application Laid-Open No. 2011-28638 (Patent Document 1) is known. Japanese Unexamined Patent Publication No. 2011-28638 discloses a summary sentence generation technique. Specifically, a word registered in the concept word dictionary is extracted from the sentence to be summarized by using a concept word dictionary in which a plurality of sets in which the morphological element part and the word concept vector are associated with the word are registered. , The feature amount of the sentence to be summarized is calculated using the word concept vector corresponding to the extracted word, and it is included in the reference sentence group which is a set of reference sentences prepared in advance for use in creating the summary sentence. Calculate the similarity of the feature amount of each reference sentence to the feature amount of the sentence to be summarized, select the reference sentence with the highest similarity, and select the word contained in the selected reference sentence. Disclose that a summary is created by replacing a word with a word contained in the sentence to be summarized, which has a high degree of similarity based on the word concept vector (see summary).

Japanese Unexamined Patent Publication No. 2011-28638

However, in oral conversations such as meetings, the subject and object are often omitted, and it is difficult to generate a summary from syntactic features. Therefore, a technique capable of appropriately and automatically generating a summary of an oral conference is desired.

A typical example of the present disclosure is a method in which a computer system creates a summary from audio data of a conference, in which the computer system stores a processor, a program executed by the processor, and a dictionary. In the method, the processor acquires the voice data of the conference, forms a current text sentence from the voice data by voice recognition processing, and refers to the current text sentence and the dictionary. , It is determined whether or not the current text sentence contains a word in the dictionary, and when the current text sentence contains a word in the dictionary, the current text sentence is stored in the storage device. Including adding to the above abstract.

According to one aspect of the present disclosure, a meeting summary can be appropriately and automatically generated.

An example of the configuration of the conference summary generation system is shown. An example of the software configuration of the conference summary generation system is shown. The software configuration related to the dictionary registration process is shown. The flowchart of the dictionary registration process is shown. The software configuration related to speech recognition processing is shown. The software configuration related to the abstract generation process is shown. The flowchart of the abstract generation process by the abstract generation program is shown. The flowchart of the abstract generation process by the abstract generation program is shown. An example of the image displayed on the screen of the display device and the operation on the image in the modification of the abstract is shown. The image displayed on the screen of the display device in the modification of the abstract and other examples of operations on the image are shown. The image displayed on the screen of the display device in the modification of the abstract and other examples of operations on the image are shown.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. It should be noted that the present embodiment is merely an example for realizing the present invention and does not limit the technical scope of the present invention. The same reference numerals are given to common configurations in each figure.

The conference abstract generation system of the present disclosure generates abstracts from oral meetings. The conference summary generation system converts the spoken sentence into a sentence (text sentence) composed of characters by voice recognition. The conference summary generation system parses a sentence and then refers to a given dictionary. If a sentence contains words that are contained in a given dictionary, the conference abstract generation system determines that the sentence is included in the abstract. This allows you to select important sentences to include in the abstract.

FIG. 1 shows a configuration example of the conference summary generation system 100. The conference summary generation system 100 is, for example, a computer system that executes a program for generating a conference summary. FIG. 1 shows a hardware configuration example of the conference summary generation system 100, which is a computer system. The conference summary generation system 100 includes a processor 110, a main storage device 120, an auxiliary storage device 130, and an interface (I / F) 140. They are connected to the internal bus and can communicate with each other.

The conference summary generation system 100 further includes a display device 150, a microphone 161, a touch input device 162, a pen input device 163, and a mouse 164. These are connected to the internal bus via the I / F 140. The display device 150 is an output device, and the microphone 161, the touch input device 162, the pen input device 163, and the mouse 164 are input devices.

The processor 110 operates according to the program stored in the main storage device 120 to realize a predetermined function of the conference summary generation system 100. The main storage device 120 is, for example, a volatile storage device and stores a program executed by the processor 110 and data to be referred to. The auxiliary storage device 130 is, for example, a non-volatile storage device and stores data loaded in the main storage device 120. The main storage device 120, the auxiliary storage device 130, and a combination thereof are storage devices.

FIG. 2 shows a software configuration example of the conference summary generation system 100. The main storage device 120 stores the voice recognition program 201, the abstract generation program 202, the abstract modification program 203, and the dictionary registration program 204. These programs are loaded from an external device (including a device on the network) into the main storage device 120 via, for example, the auxiliary storage device 130 or the I / F 140.

As described above, the processor 110 operates as a specific functional unit according to the above program. Specifically, the processor 110 functions as a voice recognition unit, a summary generation unit, a summary correction unit, and a dictionary registration unit by operating according to the above program. At least some of these functions may be implemented by logic circuits other than the processor 110.

The main storage device 120 further stores the conference material dictionary 221 and the existing noun dictionary 222, the preceding sentence information 223, and the abstract 224. These are created by any of the above programs and stored in or updated in the main storage device 120. For example, the conference material dictionary 221 is generated by the dictionary registration program 204. The existing noun dictionary 222, the preceding sentence information 223, and the abstract 224 are generated and updated by the abstract generation program 202.

The auxiliary storage device 130 stores the conference material 301, the basic dictionary 302, the 5W1H related word dictionary 303, and the consent word dictionary 304. These are pre-installed in the conference abstract generation system 100 before the start of the conference to create the abstract. The data stored in the auxiliary storage device 130 is loaded into the main storage device 120 and used by the processor 110. The auxiliary storage device 130 may be omitted. The data storage position in FIG. 2 is for convenience of the description name, and the data storage position is arbitrary.

The voice recognition program 201 analyzes the voice data input from the microphone 161 and converts it into text data. The voice recognition program 201 refers to the basic dictionary 302 and the conference material dictionary 221 in analyzing the voice data. The basic dictionary 302 is a dictionary referred to for voice recognition of arbitrary utterances, and the conference material dictionary 221 is a dictionary dedicated to a conference for creating a summary. The conference material dictionary 221 is generated from the conference material 301 by the dictionary registration program 204.

The abstract generator 202 displays the abstract 224 on the display device 150 while updating it during the meeting. The abstract generation program 202 analyzes the text data generated by the speech recognition program 201 and updates the abstract 224 as needed. As will be described later, the abstract generation program 202 adds each sentence to the abstract 224, or makes a determination by referring to the 5W1H related word dictionary 303 and the consent word dictionary 304.

The 5W1H related word dictionary 303 is preset by the system designer. In the 5W1H-related word dictionary 303, words related to 5W1H selected by design are registered. The 5W1H related word dictionary 303 can include, for example, a word (interrogative word) indicating a question corresponding to 5W1H, a word indicating a date and time, a word indicating a place, and the like.

For example, the 5W1H related word dictionary 303 stores words other than interrogative words asking time, place, subject, object, reason, or method, and interrogative words indicating any of them. For example, "when", "where", "what", "who", "why", "how", "today", "tomorrow", "next week", "last week", "* day of the week", "* It can include "month * day", "so", "here", etc. Here, "*" represents an arbitrary character.

The consent word dictionary 304 is preset by the system designer. In the consent word dictionary 304, words representing consent selected by design are registered. For example, the consent word dictionary 304 includes words such as "yes", "yes", "OK", "accept", "know", "do", and "try".

The abstract modification program 203 enables the user to modify the abstract 224 generated by the abstract generator 202 on the display device 150. The abstract modification program 203 reflects the modification to the abstract 224 input from the touch input device 162, the pen input device 163, or the mouse 164 in the abstract 224.

The processing of each program will be described below. First, the dictionary registration process by the dictionary registration program 204 will be described with reference to FIGS. 3 and 4. FIG. 3 shows a software configuration related to the dictionary registration process. The dictionary registration program 204 refers to the conference material 301 and generates the conference material dictionary 221. The conference material 301 is, for example, a document file or a presentation file.

FIG. 4 shows a flowchart of the dictionary registration process. The dictionary registration program 204 acquires the conference material 301 (S101) and executes morphological analysis of the words contained in the conference material 301 (S302). The dictionary registration program 204 sequentially selects morphemes (S103) and executes steps S104 to S106.

The dictionary registration program 204 determines whether the selected morpheme is a verb or a noun (a predetermined subtype) (S104). When the selected morpheme is neither a verb nor a noun (S104: NO), the dictionary registration program 204 selects the next morpheme without including the morpheme in the conference material dictionary 221 (S103).

When the selected morpheme is a verb or a noun (S104: YES), the dictionary registration program 204 converts the morpheme into kana (reading) (S105) and registers the morpheme and the reading in the conference material dictionary 221. (S106). The subtype of the noun to be registered in the conference material dictionary 221 is set in advance in the dictionary registration program 204. For example, common nouns, proper nouns, and sa-variant nouns are set as subtypes of nouns registered in the conference material dictionary 221.

Next, the voice recognition process by the voice recognition program 201 will be described with reference to FIG. FIG. 5 shows a software configuration related to speech recognition processing. The voice recognition program 201 analyzes the voice data input from the microphone 161 using the conference material dictionary 221 and the basic dictionary 302, and sequentially generates a sentence (text sentence) composed of characters. The generated sentence (current sentence) is stored in the main storage device 120.

Speech recognition processing is a well-known technique, and detailed description thereof will be omitted. The voice recognition program 201 analyzes voice data, extracts acoustic features, and selects words having similar acoustic features from the words registered in the conference material dictionary 221 or the basic dictionary 302. The voice recognition program 201 refers to the conference material dictionary 221 with priority over the basic dictionary 302.

For example, when the homonyms exist in both the conference material dictionary 221 and the basic dictionary 302, the voice recognition program 201 selects a word in the conference material dictionary 221. When the voice recognition program 201 cannot find a word corresponding to (determined to be) voice data in the conference material dictionary 221, the voice recognition program 201 may search for the corresponding word in the basic dictionary 302.

As described above, the conference material dictionary 221 is generated from the conference material 301 prepared in advance. Since the word of the conference material 301 is likely to be used in the conference, more accurate voice recognition becomes possible by preferentially using the conference material dictionary 221. In particular, it makes it possible to accurately recognize proper nouns, technical terms, industry terms, and homonyms.

Next, the abstract generation process (summary update process) by the abstract generation program 202 will be described with reference to FIGS. 6, 7A and 7B. FIG. 6 shows a software configuration related to the abstract generation process.

The abstract generation program 202 creates the abstract 224, stores the data in the main storage device 120, and displays the data on the display device 150. The abstract generation program 202 sequentially acquires sentences from the voice recognition program 201, and updates the abstract 224 of the main storage device 120 according to the analysis result of the sentences. The update result of the abstract 224 is reflected in the image of the abstract 224 displayed by the display device 150.

The abstract generation program 202 acquires the sentence (current sentence) generated by the speech recognition program 201. As will be described later, the abstract generation program 202 refers to the preceding sentence information 223 and the existing noun dictionary 222 to determine whether the current sentence is included in a new context different from the immediately preceding sentence. Abstract 224 displays the boundaries of the context so that the sentences that make up each context are distinguished.

In the example of FIG. 6, the abstract generator 202 inserts a blank line between the two contexts in the abstract 224. Each context may be indicated by a dividing line, a surrounding line, or the like. The context distinction in the displayed abstract 224 allows the user to easily identify interrelated sentences on the same topic.

The abstract generation program 202 stores the immediately preceding sentence information 223 in the main storage device 120 at the time of analyzing the immediately preceding sentence. The immediately preceding sentence information 223 indicates the immediately preceding sentence (word string) and the utterance end time of the immediately preceding sentence. The utterance end time is, for example, the time when the voice recognition program 201 finishes receiving the voice data of the sentence, and the abstract generation program 202 acquires the value from the voice recognition program 201.

The existing noun dictionary 222 is a list of nouns uttered since the meeting was started. The abstract generation program 202 refers to the existing noun dictionary 222 in the analysis of the current sentence, and when a new noun not stored in the existing noun dictionary 222 is included in the current sentence, the existing noun dictionary 222 is used. Add to.

The subtypes of nouns to be included in the existing noun dictionary 222 are set in advance in the abstract generation program 202, and are, for example, common nouns, proper nouns, and sa-variant nouns. The existing noun dictionary 222 may be a list of nouns in the immediately preceding context instead of the list of nouns from the start of the conference.

The abstract generation program 202 refers to the 5W1H related word dictionary 303, the consent word dictionary 304, and the conference material dictionary 221 to analyze the current sentence and determine the sentence to be included in the abstract 224. Hereinafter, the abstract generation process by the abstract generation program 202 will be described with reference to the flowcharts of FIGS. 7A and 7B. FIG. 7A shows a flowchart of the context determination process, and FIG. 7B is a flowchart of the abstract update process.

The abstract generation program 202 acquires the current sentence (text data) generated by the voice recognition program 201 (S201). The abstract generation program 202 executes morphological analysis of the current sentence and identifies the part of speech of each word (morpheme) (S202).

Next, the abstract generation program 202 determines whether a long time has elapsed from the utterance end time of the immediately preceding sentence to the utterance start time of the current sentence (S203). For example, the abstract generation program 202 acquires the above two times from the voice recognition program 201, and determines that a long time has elapsed when the elapsed time exceeds a predetermined threshold value.

When a long time has passed since the utterance of the immediately preceding sentence (S203: YES), the abstract generation program 202 determines that the context of the current sentence has been changed from the immediately preceding context. The abstract generation program 202 adds a blank line indicating the change of context to the abstract 224 (S205). The passage of a long time system is likely to indicate that the previous sentence is less relevant to the current sentence. This step allows for more accurate context distinction.

If a long time has not passed since the utterance of the immediately preceding sentence (S203: NO), the abstract generation program 202 has many new nouns in the current sentence, and the number of directives preset in the abstract generation program 202. Is less in the immediately preceding sentence (S204).

Specifically, the abstract generation program 202 refers to the existing noun dictionary 222 and counts the number of new nouns in the current sentence. If the number of new nouns is greater than a predetermined threshold, the abstract generator 202 determines that there are many new nouns in the current sentence. The threshold is an integer of 0 or more.

Further, the abstract generation program 202 executes the morphological analysis of the immediately preceding sentence in the immediately preceding sentence information 223, and counts the number of preset directives included therein. When the number of demonstratives is less than a predetermined threshold value, the abstract generation program 202 determines that the number of demonstratives in the immediately preceding sentence is small. The threshold is an integer of 1 or more.

When there are many new nouns in the current sentence and few demonstratives in the previous sentence (S204: YES), the abstract generator 202 determines that the context has been changed, and changes the context in the abstract 224. A blank line is added to indicate (S205). If the determination results in steps S203 and S204 are both NO, the abstract generation program 202 determines that the current sentence is included in the same context as the immediately preceding sentence.

If there are many new nouns, it is highly possible that a context different from the previous context has been started. In addition, the fact that there are few demonstratives is likely to indicate that the previous sentence is less relevant to the current sentence. Therefore, this step allows for more accurate context distinction.

Next, the abstract update process will be described with reference to FIG. 7B. The abstract generator 202 determines whether the current sentence contains 5W1H related words (S206). Specifically, the abstract generation program 202 determines whether or not the current sentence contains a word registered in the 5W1H related word dictionary 303. When the current sentence contains 5W1H related words (S206: YES), the abstract generator 202 outputs the current sentence to the abstract 224 (S207). For example, the abstract generator 202 adds the current sentence to the last line of the abstract.

The abstract generator 202 further determines if the current sentence contains a questioning word in the 5W1H related words. Words that indicate a question are, for example, "why", "when", "who", "where", "how", and the like. For example, the 5W1H related word dictionary 303 contains information for identifying a word that indicates a question. If the questioning word is included in the current sentence, the abstract generator 202 determines that the sentence following the current sentence is included in the abstract 224.

The sentence following the sentence containing the word indicating the question is likely to be an important sentence containing the answer to the question, which can increase the possibility of including the important sentence in the summary. In addition, by including a sentence indicating a question and a sentence indicating the answer in the abstract 224, the content of the meeting can be shown more accurately. For example, the abstract generation program 202 internally retains information indicating that the following sentence is included in the abstract 224, or includes it in the generation of the preceding sentence information 223 indicating the information of the current sentence.

If the current sentence does not contain 5W1H related words (S206: NO), the abstract generator 202 determines whether the current sentence contains consent words (S208). Specifically, the abstract generation program 202 determines whether or not the current sentence contains a word registered in the consent word dictionary 304.

When the current sentence contains a consent word (S208: YES), the abstract generator 202 determines whether the immediately preceding sentence has been output to the abstract 224 (S209). The abstract generation program 202 compares the immediately preceding sentence information 223 with the abstract 224, and can determine whether or not the immediately preceding sentence has been output to the abstract 224. Immediate sentence information 223 may indicate whether the preceding sentence is included in the abstract 224.

When the immediately preceding sentence is not included in the abstract 224 (S209: NO), the abstract generation program 202 acquires the immediately preceding sentence from the immediately preceding sentence information 223 and outputs it to the abstract 224 (S212). The sentence immediately preceding the current sentence containing the consent word is likely to include the subject of consent. This step allows the abstract 224 to include important text that includes the subject of consent. If the preceding sentence is included in the abstract 224 (S209: YES), the process ends for the current sentence.

If the current sentence does not contain a consent word (S208: NO), the abstract generator 202 determines whether the current sentence contains a word of the conference material (S211). Specifically, the abstract generation program 202 compares the conference material dictionary 221 with the current sentence, and determines whether the word registered in the conference material dictionary 221 is included in the current sentence.

If the current sentence contains the words of the conference material (S211: YES), the abstract generator 202 outputs the current sentence to the abstract 224 (S212). Thereby, the sentence including the important word contained in the conference material 301 can be included in the abstract 224. If the current sentence does not include the word of the conference material (S211: NO), the processing for the current sentence ends.

When adding a sentence to the abstract 224 (S207, S210, S212), the abstract generator 202 deletes an existing sentence whose information overlaps with the sentence to be added from the abstract 224. This makes the abstract 224 simple and easy to read.

Specifically, the abstract generation program 202 compares the sentence added to the abstract 224 with each of the existing sentences in the same context. When the sentence to be added and the existing sentence include the same combination of the verb and the noun (S213: YES), the abstract generation program 202 deletes the existing sentence from the abstract 224 (S214).

For example, if the existing sentence contains a set consisting of any one verb included in the added sentence and any one noun of a predetermined subtype, the existing sentence is deleted. Alternatively, the condition for deleting the existing sentence may be that all the verbs and all the noun sets of the predetermined subtype are common to the current sentence and the existing sentence. For example, common nouns, proper nouns, and s-irregular nouns are preset as noun subtypes.

If none of the existing sentences contains the same verb and noun combination as the sentence to be added (S213: NO), all the existing sentences in the same context are maintained. After the step (S213, 214) for the existing sentence, the processing for the current sentence ends.

Hereinafter, the modification process of the abstract 224 by the abstract modification program 203 will be described. FIG. 8 shows an image displayed on the screen of the display device 150 and an example of an operation on the image in the abstract correction. The user selects the sentence (image) 241 to be corrected in the abstract 224 displayed on the screen. In the example of FIG. 8, the sentence 241 before correction is selected by the finger via the touch input device 162.

The abstract modification program 203 performs morphological analysis of the selected sentence 241 and extracts verbs and nouns (prescribed part of speech) of a predetermined subtype from the selected sentence 241. The abstract modification program 203 specifies the extracted word string (image) 251 by displaying it in a section different from the abstract 224. For example, common nouns, proper nouns, and s-irregular nouns are preset as noun subtypes. The part of speech to be extracted may be set by design.

In the example of FIG. 8, the order of the words in the word string 251 is the same as that of the selected sentence 241. Adjacent words in word string 251 are connected by a line. The abstract modification program 203 accepts the user's selection of words in the word string 251. In the example of FIG. 8, the user selects one word "patent" by touching with a finger.

Further, the user moves while touching the word "patent" with a finger, surrounds the two words "meeting" and "do", and then releases the finger from the screen. As a result, in addition to the word "patent", the words "meeting" and "do" are also selected. In the example of FIG. 8, the user can select a continuous word group including the first selected word.

The abstract modification program 203 changes the display image of the word string 251 according to the input from the touch input device 162, and detects that three words have been selected in the word string 251. In the example of FIG. 8, the abstract modification program 203 sets the selected (word “patent”) and enclosed (words “meeting”, “doing”) words by a single drag-and-drop operation as the selected words. Identify.

The abstract modification program 203 displays the word string (image) 252 of the selected word in place of the word string 251. The order of the words in the word string 252 is the same as that of the selected sentence 241. Adjacent words in word string 252 are connected by a line.

The abstract modification program 203 generates a modified sentence from the word string 252 of the selected word. Specifically, the abstract modification program 203 refers to the amended preamble 241 and acquires the particle "no" between the word "patent" and the next word "meeting" in the word string 252. Further, the abstract modification program 203 acquires the particle "o" between the word "meeting" and the next word "do" in the word string 252. The generated modified sentence 242 becomes "Patent meeting."

The abstract modification program 203 rewrites the sentence 241 before modification to the sentence 242 after modification in the abstract 224 of the main storage device 120. In the image of the abstract 224 displayed by the display device 150, the sentence 241 before the correction is replaced with the sentence 242 after the correction. The abstract modification program 203 may generate a copy file of the abstract 224 before modification and modify the copy file.

FIG. 9A shows the image displayed on the screen of the display device 150 in the abstract modification and another example of the operation on the image. This example generates one post-correction sentence from a plurality of pre-correction sentences. The user selects the sentence (image) 241 to be corrected in the abstract 224 displayed on the screen. In the example of FIG. 9, two uncorrected sentences 241 are selected by a finger via the touch input device 162.

The abstract modification program 203 performs morphological analysis of the selected sentence 241 and extracts verbs and nouns (prescribed part of speech) of a predetermined subtype from the selected sentence 241. The abstract modification program 203 specifies the extracted words in the selected sentence 241. In the example of FIG. 9A, the words extracted in the sentence (image) 244 being modified are each surrounded by a broken line rectangle. The part of speech to be extracted may be set according to the design.

In the example of FIG. 9A, the abstract modification program 203 accepts the user's word selection in the sentence 244 being modified. In the example of FIG. 9A, the user selects those words by tracing the words on the screen with a finger. In the example of FIG. 9A, the words "tomorrow" in the first sentence and the words "patent", "meeting", and "do" in the second sentence are selected. In the example of FIG. 9, the abstract modification program 203 accepts the user to select a discontinuous word group from the extracted words (verbs and nouns).

The abstract modification program 203 detects that four words have been selected in the sentence 244 being modified in response to input from the touch input device 162. In the example of FIG. 9A, the abstract modification program 203 identifies the words touched by a single drag-and-drop operation (“tomorrow”, “patent”, “meeting”, “do”) as selected words. To do.

The abstract modification program 203 generates the modified sentence 242 from the word string of the selected words. The word order in the modified sentence 242 matches the selected order. The abstract modification program 203 refers to the amended preamble 241 as the particle "no" between the word "patent" and the next extracted word "meeting", and the word "meeting" and the next extracted word "do". Restore the particle "o" between. Since the word "tomorrow" and the word "patent" are contained in different preambles, the abstract modification program 203 concatenates them with a comma. As a result, a natural modified sentence 242 can be generated.

FIG. 9B shows an image displayed on the screen of the display device 150 in the abstract modification and another example of the operation on the image. In the following, the differences from the example of FIG. 9A will be mainly described. In the example of FIG. 9B, the words "tomorrow", "plan", "confirm", and "will" in the first sentence, and the word "patent" in the second sentence are selected.

The abstract modification program 203 generates the modified sentence 242 from the word string of the selected words. The word order in the modified sentence 242 matches the selected order. The abstract modification program 203 refers to the modification preamble 241 and restores the collocation "about" between the word "plan" and the next extracted word "confirmation".

The word "tomorrow" and the word "patent" are contained in different preambles, and the word "patent" and the word "planned" are contained in different preambles. Therefore, the abstract modification program 203 connects the words "tomorrow" and "patent" with commas, and connects the words "patent" with commas. As a result, a natural modified sentence 242 can be generated.

The above example updates the summary displayed as the meeting progresses, while displaying the summary at the meeting. Unlike this, the system may create a summary from the audio data of the meeting after the meeting is over. The system may be configured to include only some of the above functions. For example, the processing of the context division and the reference information thereof may be omitted, or the function of modifying the summary sentence may be omitted. Only some dictionaries of the above dictionaries may be implemented.

The present invention is not limited to the above-described examples, and includes various modifications. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and is not necessarily limited to the one including all the configurations described. Further, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Further, it is possible to add / delete / replace a part of the configuration of each embodiment with another configuration.

Further, each of the above-mentioned configurations, functions, processing units and the like may be realized by hardware by designing a part or all of them by, for example, an integrated circuit. Further, each of the above configurations, functions, and the like may be realized by software by the processor interpreting and executing a program that realizes each function. Information such as programs, tables, and files that realize each function can be placed in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card or an SD card. In addition, control lines and information lines are shown as necessary for explanation, and not all control lines and information lines are shown in the product. In practice, it can be considered that almost all configurations are interconnected.

100 conference summary generator, 110 processor, 120 main memory, 130 auxiliary memory, 150 display, 161 microphone, 162 touch input device, 163 pen input device, 164 mouse, 201 voice recognition program, 202 summary generator , 203 abstract modification program, 204 dictionary registration program, 221 conference material dictionary, 222 existing nomenclature dictionary, 223 last sentence information, 224 summary, 241 amended preamble, 242 amended post sentence, 244 amended sentence, 251, 252 words Columns, 301 meeting materials, 302 basic dictionaries, 303 5W1H related word dictionaries 304 consent word dictionaries

Claims (6)

A way for computer systems to create summaries from conference audio data
The computer system
With the processor
Includes a program executed by the processor and a storage device that stores a dictionary.
In the method, the processor acquires the audio data of the conference.
From the voice data of the conference, a text sentence is currently formed by voice recognition processing.
With reference to the current text sentence and the dictionary, it is determined whether the current text sentence contains a word in the dictionary.
Wherein if it contains words currently in the dictionary text sentence, the current text sentence, to add to the abstract stored in the storage device, it viewed including the,
The dictionary contains words that indicate a question.
The method is that the processor adds the current text sentence and the next text sentence of the current text sentence to the abstract when the current text sentence contains a questioning word in the dictionary. including methods. A way for computer systems to create summaries from conference audio data
The computer system
With the processor
Includes a program executed by the processor and a storage device that stores a dictionary.
In the method, the processor
Acquire the audio data of the conference and
From the voice data of the conference, a text sentence is currently formed by voice recognition processing.
With reference to the current text sentence and the dictionary, it is determined whether the current text sentence contains a word in the dictionary.
Including adding the current text sentence to the abstract stored in the storage device when the current text sentence contains a word in the dictionary.
The storage device stores a consent word dictionary consisting of words indicating consent.
The method comprises adding the text sentence immediately preceding the current text sentence to the abstract when the current text sentence contains a word indicating acceptance in the consent word dictionary. A way for computer systems to create summaries from conference audio data
The computer system
With the processor
Includes a program executed by the processor and a storage device that stores a dictionary.
In the method, the processor
Acquire the audio data of the conference and
From the voice data of the conference, a text sentence is currently formed by voice recognition processing.
With reference to the current text sentence and the dictionary, it is determined whether the current text sentence contains a word in the dictionary.
Including adding the current text sentence to the abstract stored in the storage device when the current text sentence contains a word in the dictionary.
The computer system further includes a display device for displaying the abstract.
In the method, the processor
Display the abstract on the display device and display it.
Accepts user selection of uncorrected text in the abstract
In the uncorrected text sentence displayed on the display device, clearly indicate the word of the specified part of speech.
Accepting user selection of words from the words of the specified part of speech,
A new text sentence is generated by including the user-selected words from the words of the specified part of speech and excluding the words not selected by the user in the words of the specified part of speech.
Replace the uncorrected text sentence with the new text sentence,
A method comprising complementing a word with a part of speech other than the specified part of speech following the word selected by the user in the uncorrected text sentence in the new text sentence. A way for computer systems to create summaries from conference audio data
The computer system
With the processor
Includes a program executed by the processor and a storage device that stores a dictionary.
In the method, the processor
Acquire the audio data of the conference and
From the voice data of the conference, a text sentence is currently formed by voice recognition processing.
With reference to the current text sentence and the dictionary, it is determined whether the current text sentence contains a word in the dictionary.
Including adding the current text sentence to the abstract stored in the storage device when the current text sentence contains a word in the dictionary.
The computer system further includes a display device for displaying the abstract.
In the method, the processor
Display the abstract on the display device and display it.
Accepting user selection of the first uncorrected text sentence and the second uncorrected text sentence in the abstract
In the first uncorrected text sentence and the second uncorrected text sentence displayed on the display device, the word of the specified part of speech is clearly indicated.
In the first uncorrected text sentence and the second uncorrected text sentence displayed on the display device, the user selection of a word from the word of the specified part of speech is accepted.
Generate a new text sentence containing the user-selected word
Including that the first uncorrected text sentence and the second uncorrected text sentence are deleted from the abstract and the new text sentence is added to the abstract.
The generation of the new text sentence is
The user-selected words are arranged in the selected order to form an array.
In the array of words selected by the user, a comma is placed between a continuous word consisting of a word contained in the first uncorrected text sentence and a word contained in the second uncorrected text sentence. Complement and
In the array of user-selected words, the said in one of the consecutive words consisting of words contained in one of the first uncorrected text sentence or the second uncorrected text sentence. A method that includes complementing words with a part of speech other than the prescribed part of speech. The method according to any one of claims 1 to 4.
The storage device stores conference materials used in the conference and
The method is
A method comprising the processor performing a morphological analysis of the conference material and selecting a predetermined part of speech word to include in the dictionary before acquiring the audio data of the conference. The method according to any one of claims 1 to 4.
The dictionary comprises a word indicating time and a word indicating place.

Images