JP2024027395A - Dictionary registration program, dictionary registration method, and information processing device - Google Patents

Abstract

The present invention provides a dictionary registration program, a dictionary registration method, and an information processing device that streamline dictionary registration work. [Solution] A character recognition result sentence and reading information obtained by performing voice recognition on voice data using a dictionary are obtained, and a first modification to a morpheme included in the character recognition result sentence is detected. When the first correction is made based on the reading information for each character, the character recognition result of the morpheme is correlated with the reading information of each character included in the morpheme, and the morpheme is The correspondence between the first corrected character recognition result and reading information for is registered in the dictionary. [Selection diagram] Figure 4

Description

The present invention relates to a dictionary registration program, a dictionary registration method, and an information processing device.

In recent years, a number of systems have been proposed that automatically create text by converting speech, such as speech, into text to make it easier to create meeting minutes and write notes using mobile terminal devices. . Such systems that automatically generate text sentences generally use a method of converting speech input from a sound pickup device such as a microphone into text using a speech recognition tool.

However, since the analysis accuracy of speech analysis tools is not sufficient, the output text often contains erroneous translations. Therefore, in order to use automatically generated text sentences as accurate sentences, manual correction work is performed. Correcting erroneous conversions involves identifying the erroneously converted term and correcting it to the correct character string. If the analysis accuracy is low and there are many erroneous conversions, the number of man-hours will increase and the correction work for the operator will be complicated. becomes.

Therefore, in order to improve analysis accuracy and reduce erroneous conversions, terms unique to the usage environment that are not registered in the dictionary of the speech analysis tool are newly registered in the dictionary, and the priority of the registered words is increased. Examples of terms specific to the usage environment include personal names, company-specific terms, and new IT (Information Technology) terms. By performing such registration, it is possible to perform highly accurate voice analysis according to the usage environment, and it is possible to improve the probability of automatically generating accurate sentences.

However, when registering a new term in a dictionary manually, the operator must decide which word to register from among many possible candidates, which is a time-consuming task for the operator. Therefore, when registering words specific to the usage environment in a dictionary, a method is adopted in which the revised terms are automatically registered in the dictionary based on the operator's correction of the term to the text sentence generated by the speech analysis tool. . Since such modification of terms is likely to include many modifications of terms specific to the usage environment, new terms can be automatically registered according to the usage environment. This makes it possible to streamline the process of registering new terms in the dictionary.

In addition, as a voice recognition technology, voice recognition is performed to extract and display the keywords contained in the voice used for search, and the words that are one letter different from the keywords that the user has instructed to correct are displayed and selected as correction candidates. Techniques have been proposed to make this possible.

JP2005-275228A

However, when automatically registering terms based on input of corrections to a text sentence, an operator must correct the text sentence using an editor compatible with the speech analysis tool. As described above, since corrections to text sentences are made manually by an operator, typographical errors may occur. If a typo occurs, a device that automatically registers new terms may not be able to determine which word was to be registered. For example, if a two-letter term is corrected, and then one letter is corrected again due to a typo, the speech analysis tool will determine that the original term to be corrected is one letter. There is a risk of Therefore, with a method that simply automatically registers the correction results for text sentences, appropriate dictionary registration is not performed, and it is difficult to make the dictionary registration work more efficient.

The disclosed technology has been made in view of the above, and aims to provide a dictionary registration program, a dictionary registration method, and an information processing device that make dictionary registration work more efficient.

In one aspect of the dictionary registration program, dictionary registration method, and information processing device disclosed in the present application, a computer performs voice recognition on voice data using a dictionary, and generates character recognition result sentences and reading information. is obtained, and if a first modification to the morpheme included in the character recognition result sentence is detected, the character recognition result of the morpheme to which the first modification has been added based on the reading information for each character and the Correlate the reading information of each character included in the morpheme, and in response to the confirmation of the correction, store the correspondence between the character recognition result after the first correction and the reading information for the morpheme in the dictionary. Execute the process to register.

In one aspect, the present invention can streamline dictionary registration work.

FIG. 1 is a block diagram of a speech-to-text conversion system. FIG. 2 is a block diagram of the speech-to-text conversion system according to the embodiment. FIG. 3 is a diagram showing an example of text to be corrected. FIG. 4 is a diagram illustrating an example of text correction processing. FIG. 5 is a diagram showing a process of associating characters with readings. FIG. 6 is a flowchart of dictionary registration processing according to the embodiment. FIG. 7 is a hardware configuration diagram of the server.

Embodiments of a dictionary registration program, a dictionary registration method, and an information processing apparatus disclosed in the present application will be described in detail below with reference to the drawings. Note that the dictionary registration program, dictionary registration method, and information processing apparatus disclosed in the present application are not limited to the following embodiments.

FIG. 1 is a block diagram of a speech-to-text conversion system. In this embodiment, as shown in FIG. 1, a speech-to-text conversion system 1 constructed using a client-server system having a server 10 and clients such as user terminal devices 20A and 20B will be described. Here, the server 10 may be placed on a cloud. Furthermore, the speech-to-text conversion system 1 may be a single computer in which the functions of the server 10 and the functions of clients such as the user terminal devices 20A and 20B are integrated.

The server 10, which is an information processing device, is connected to a user terminal device 20A and a sound collection device 30A installed in a specific meeting room. Further, the server 10 is connected to a user terminal device 20B and a sound collection device 30B installed in another meeting room.

The sound collection device 30A is, for example, a microphone. The sound collection device 30A collects sounds emitted during a conference or the like in a meeting room. The sound collection device 30A then transmits the collected audio to the server 10. The sound collection device 30B is, for example, an IC recorder. The sound collection device 30B collects sounds emitted in other meeting rooms. Then, the sound collection device 30B transmits the collected audio to the server 10.

The user terminal device 20A receives text data of the audio collected by the sound collection device 30A from the server 10 and displays it on a monitor or the like. Next, the user terminal device 20A receives input of corrections made by the user to the displayed text sentence. The user terminal device 20A then transmits the correction information to the server 10.

Similarly, the user terminal device 20B receives data in the form of text of the audio collected by the sound collection device 30B from the server 10, and displays it on a monitor or the like. Next, the user terminal device 20B receives input of corrections made by the user to the displayed text sentence. The user terminal device 20B then transmits the correction information to the server 10. The user terminal device 20A and sound collection device 30A and the user terminal device 20B and sound collection device 30B have similar functions. Therefore, in the following, when the user terminal device 20A and the user terminal device 20B are not distinguished, they are referred to as the user terminal device 20, and when the sound collecting device 30A and the sound collecting device 30B are not distinguished, they are referred to as the sound collecting device 30. call.

The server 10 receives audio data from the sound collection device 30. Next, the server 10 performs phonetic recognition using a voice analysis tool, converts the voice data into text, and generates a text sentence. The server 10 then transmits the text to the user terminal device 20.

Thereafter, the server 10 receives modification information for the transmitted text from the user terminal device 20. Then, the server 10 registers the corrected terms in the dictionary and uses them for subsequent conversion of audio data into text. The details of the processing of the server 10 will be explained below.

FIG. 2 is a block diagram of the speech-to-text conversion system according to the embodiment. As shown in FIG. 2, the server 10 includes a communication control section 11, a speech analysis section 12, a dictionary 13, a correction management section 14, a storage section 15, a registration section 16, and a kanji conversion dictionary 17.

The communication control unit 11 controls communication with the user terminal device 20. The communication control unit 11 receives audio data from the sound collection device 30. The communication control unit 11 then outputs the received audio data to the audio analysis unit 12. Thereafter, the communication control section 11 receives from the speech analysis section 12 a text sentence obtained by converting the speech data into text. The communication control unit 11 then transmits the text to the user terminal device 20.

Further, the communication control unit 11 receives a notification of the start of editing a text sentence from the user terminal device 20 together with the data of the text sentence to be edited. The communication control unit 11 then outputs the data of the text sentence to be edited to the correction management unit 14. Further, the communication control unit 11 receives input of a text sentence divided into terms from the correction management unit 14 . The communication control unit 11 then transmits the text sentence divided into terms to the user terminal device 20. Furthermore, the communication control unit 11 receives operation information for correction processing for the text sentence to be edited from the user terminal device 20 . The communication control unit 11 then outputs operation information for the correction process to the correction management unit 14.

Thereafter, the communication control unit 11 receives a notification of completion of editing from the user terminal device 20. Then, the communication control unit 11 outputs a notification of completion of editing to the registration unit 16.

The dictionary 13 is a dictionary used in speech analysis. In the dictionary 13, for each term, its reading and characters including kanji whose readings have been converted are registered in association with each other.

The voice analysis section 12 receives input of voice data from the communication control section 11 . Then, the speech analysis unit 12 executes speech recognition of the speech data, generates a text sentence by converting the speech data into text using the dictionary 13, and acquires the reading of each character of the text sentence. This generated text sentence is an example of a "recognition result sentence." That is, the speech analysis unit 12 performs speech recognition on the speech data using the dictionary 13, and obtains a recognition result sentence obtained by transcribing the speech data and reading information. Thereafter, the voice analysis section 12 outputs text sentence data corresponding to the voice data to the communication control section 11.

The kanji conversion dictionary 17 is a dictionary in which kanji and one or more readings corresponding to the kanji are registered in association with each other.

When a notification to start editing a text sentence is sent from the user terminal device 20 to the server 10, the modification management unit 14 receives input of data of the text sentence to be edited from the communication control unit 11. Next, the modification management unit 14 divides the text sentence obtained by morphological analysis into parts of speech. This part of speech is an example of a "morpheme." This allows the modification management unit 14 to grasp the position of the term to be modified in the text sentence.

FIG. 3 is a diagram showing an example of text to be corrected. Text 101 is the correct text. The text 102 is a text created by erroneously recognizing a voice that should originally be recognized as the text 101 through voice analysis. The text 102 is divided into parts of speech by the modification management unit 14. Each part of speech in the text 102 is shown divided by slashes. In text 102, a term for which "Furuya" is the correct answer, as shown in text 101, is incorrectly recognized as "Furuya." In this case, the user will modify "Furuya" in the text 102 to "Furuya."

The modification management unit 14 transmits the text divided into parts of speech to the user terminal device 20 via the communication control unit 11 . The user makes corrections using the text sent from the correction management section 14, which is divided into parts of speech. However, the text sentence divided into parts of speech may not be transmitted, and the user may be allowed to make corrections using the text sentence before division without worrying about the division.

The modification management section 14 receives from the communication control section 11 the input of the operation information for the modification process transmitted from the user terminal device 20, and stores the term to be registered in the storage section 15 along with its reading. The details of the process of determining terms to be registered by the modification management unit 14 will be described below.

The modification management unit 14 receives input of the cursor position of the modification location. Here, when correcting a term, the user sets the cursor at the end of the divided part of speech that includes the term to be corrected. However, the correction management unit 14 may receive an input of an appropriate cursor position specified by the user, move the cursor position to the end of the term corresponding to the cursor position, and display it on the user terminal device 20. Furthermore, when the user makes a correction using the text sentence before division, the correction management unit 14 receives input of an appropriate cursor position specified by the user and moves the cursor to the end of the term corresponding to the cursor position. It may be displayed on the user terminal device 20 by moving the position.

For example, in the case of the text 102 shown in FIG. 3, "Furuya" is the correction target, so the modification management unit 14 recognizes that the cursor has been set at the end of the term "Furuya."

Next, the modification management unit 14 receives input of operation information from the user from the cursor position at the end of the term to be modified. For example, when the cursor is moved using the arrow keys, the modification management unit 14 identifies the movement position from the end. Further, when the backspace key or delete key is pressed, the modification management unit 14 recognizes that characters have been deleted. For example, if the backspace key is pressed twice in a row, the correction management unit 14 recognizes that the previous two characters from the immediately previous cursor position have been erased, and identifies them as the word to be corrected. Here, the explanation will be made on the assumption that the word to be corrected and other characters are included in the term. Then, the correction management unit 14 causes the storage unit 15 to store the position and number of characters of the word to be corrected in the text sentence as correction target information. That is, the first modification to a specific term is an example of a "first modification," and the modification management unit 14 edits the characters modified by the first modification based on the operation information for moving the cursor and deleting characters. Identify.

FIG. 4 is a diagram illustrating an example of text correction processing. First, since the word "Furuya" 111 has been deleted, the modification management unit 14 determines that the position of "Furuya" in the text 102 of FIG. 3 is the position of the word to be modified, and the number of characters in the word to be modified is 2. Recognizing that there is such information, the information is stored in the storage unit 15 as correction target information.

Next, the correction management unit 14 receives an input of the reading of the word to be corrected. Then, the correction management unit 14 causes the storage unit 15 to store the pronunciation of the term to be corrected, including the pronunciation of the word to be corrected. Furthermore, the correction management unit 14 receives input of characters whose reading has been converted. Then, the correction management unit 14 causes the storage unit 15 to store the term to be corrected, which includes the characters of the converted word of the word to be corrected. Further, the correction management unit 14 refers to the kanji conversion dictionary 17 and identifies the correspondence between each character and the reading based on the correspondence between the converted character and the reading of the word to be corrected. The converted characters are an example of "recognition results of modified morphemes." That is, the modification management unit 14 associates the recognition result of the morpheme that has been modified based on the reading information of each character with the reading information of each character included in the morpheme.

For example, in FIG. 4, the modification management unit 14 receives an input of the pronunciation 121 of "ko-ya". Further, the modification management unit 14 receives the word "hut" 112 as the converted characters through the conversion performed in step S101.

FIG. 5 is a diagram showing a process of associating characters with readings. The correction management unit 14 acquires the respective pronunciations 131 and 132 of the word "koya" 112 from the kanji conversion dictionary 17. Next, the correction management unit 14 refers to the reading of the word to be corrected held in the storage unit 15 and confirms that the reading is "ko-ya." Then, the correction management unit 14 confirms that "ko" is included in the pronunciations 131 of "ko", and associates the pronunciation 133 "ko" of the pronunciations "ko-ya" with "ko". In addition, the modification management unit 14 confirms that "ya" is included in the pronunciation 131 of "ya" and associates the pronunciation 134 of "ya" among the pronunciations "ko-ya" with "ya".

Here, for example, if the converted characters of the input word to be corrected are typographical errors, the user repeats the correction operation again. This second modification operation is an example of a "second modification." In that case, the modification management unit 14 specifies the movement position again by moving the cursor using the arrow keys. Then, when the backspace key or the delete key is pressed, the correction management unit 14 recognizes the deletion of the character again. At this time, the correction management unit 14 confirms that the correction position matches the position in the text sentence included in the correction target information stored in the storage unit 15, and recognizes that the correction is to a word that has already been targeted for correction. do. Then, the modification management unit 14 recognizes that a specific character of the word that has already been targeted for modification has been deleted. Furthermore, the correction management unit 14 identifies the erased portion of the reading of the word to be corrected from the reading corresponding to the corrected character, and selects the remaining readings of the word that has already been targeted for correction other than the erased portion. Recognize.

For example, in FIG. 4, the modification management unit 14 recognizes that "small" in the word "hut" 112 has been deleted. At this time, the correction management unit 14 identifies that "ko" has been deleted from the reading 121 as shown in step S102, and recognizes that the remaining reading has become "ya" shown in the reading 122.

Next, the correction management unit 14 receives the input of the re-corrected pronunciation of the character erased due to a typographical error, adds the re-corrected pronunciation to the remaining pronunciations, and completes the pronunciation of the word to be corrected. Thereafter, the modification management unit 14 completes the reading of the term to be modified, including the reading of the word to be modified, and stores it in the storage unit 15 for updating. Furthermore, the correction management unit 14 receives input of characters whose reading has been converted. Then, the correction management unit 14 stores and updates the term to be corrected, which includes the characters of the converted word of the word to be corrected, in the storage unit 15 in correspondence with the pronunciation of the term. That is, when the second modification to the character indicated by the position and number of characters in the recognition result sentence is detected, the modification management unit 14 selects the recognition result of the morpheme and the morpheme to which the first modification and the second modification have been added. and the reading information of each character included in the text.

For example, in FIG. 4, as shown in step S103, the correction management unit 14 receives an input of "ko" as the reading to be re-corrected in the reading 122 state, and then inputs the reading of the word to be corrected, "ko-ya". Complete 123 again. Further, the correction management unit 14 receives the input of the character "Ko" as the converted character by the conversion performed in step S104, and recognizes that the word 113 to be corrected is "Furuya". Then, the correction management unit 14 associates “Furuya” and “ko-ya” and stores them in the storage unit 15.

The modification management unit 14 repeats the above-described registration of the term to be corrected and the reading of the term until it receives a notification of completion of editing. Here, the user may notify the completion of editing after correcting a plurality of terms included in the text sentence. That is, the correction management unit 14 may register a plurality of terms to be corrected and how to read the terms from the start of editing to the completion of editing. Thereafter, when the modification management section 14 receives a notification of completion of editing via the communication control section 11, it deletes the modification target information stored in the storage section 15.

The storage unit 15 is a storage device. The storage unit 15 temporarily stores correction target information including the position and number of characters of a word to be corrected in a text sentence. Furthermore, the storage unit 15 holds the final revised recognition results and readings of the term to be revised.

The registration unit 16 receives a notification from the communication control unit 11 that the editing of the text sentence has been completed. Then, the registration unit 16 acquires the term to be corrected and the pronunciation of the term held in the storage unit 15. after that. The registration unit 16 registers the acquired term and its pronunciation in the dictionary 13 as a new term. That is, in response to the confirmation of the correction, the registration unit 16 registers the correspondence between the corrected recognition result and the reading information for the morpheme in the dictionary.

Next, the user terminal device 20 will be explained. The user terminal device 20 includes an input device 21, a communication control section 22, a display device 23, and an input/output control section 24. A user inputs data and commands to the user terminal device 20 using the input device 21 and display device 23 .

The communication control unit 22 controls communication with the server 10. The communication control unit 22 outputs information received from the server 10 to the input/output control unit 24. Furthermore, the communication control unit 22 transmits information input from the input/output control unit 24 to the server 10.

The input/output control unit 24 processes data and commands input from the input device 21 and controls the display of the table device 2. For example, the input/output control unit 24 obtains, from the communication control unit 22, text data obtained by converting audio data transmitted from the server 10 into text. The communication control unit 11 then outputs the acquired text sentence data to the display device 23 for display.

The input/output control unit 24 also receives a notification of the start of editing a text sentence from the input device 21 together with information on the text sentence to be edited. Then, the input/output control section 24 transmits a notification to start editing the text sentence to the server 10 via the communication control section 22 together with the data of the text sentence to be edited. Thereafter, the input/output control section 24 obtains the text sentence divided into parts of speech transmitted from the server 10 from the communication control section 22 . Then, the input/output control unit 24 outputs the text sentence divided into parts of speech to the display device 23 for display. Further, the input/output control unit 24 receives input of operation information for correction processing for the text sentence to be corrected from the input device 21 . Then, the communication control unit 22 outputs the operation result to the display device 23 to be reflected in the display, and transmits operation information for the correction process to the server 10. In addition, the input/output control unit 24 may output other information transmitted from the server 10 to the display device 23 for display. Thereafter, the input/output control unit 24 receives an input from the input device 21 of notification of completion of editing the text sentence. Then, the input/output control unit 24 transmits a notification of completion of editing the text sentence to the server 10 via the communication control unit 22.

The display device 23 is, for example, a monitor or a display. The display device 23 displays the data input from the input/output control section 24 and provides the data to the user as an image. For example, the display device 23 displays a text sentence obtained by converting audio data into text. Further, the display device 23 displays a text sentence divided into parts of speech. Further, the display device 23 updates the displayed image to reflect the operation result input from the input/output control unit 24.

The input device 21 is, for example, a keyboard or a mouse. The input device 21 receives a user's operation and outputs operation information to the input/output control unit 24 . For example, the input device 21 outputs a notification of the start of editing and a notification of the end of editing to the input/output control unit 24. In addition, the input device 21 outputs operation information of the operation performed by the user in the correction process to the input/output control unit 24 .

FIG. 6 is a flowchart of dictionary registration processing according to the embodiment. Next, with reference to FIG. 6, the flow of the land registration process according to this embodiment will be explained.

The audio analysis unit 12 receives input of audio data transmitted from the sound collection device 30. Then, the voice analysis unit 12 executes voice analysis using the acquired voice data (step S1).

Then, the speech analysis unit 12 uses the dictionary 13 to generate a text sentence by converting the speech data into text. Thereafter, the speech analysis unit 12 transmits the generated text sentence to the user terminal device 20 and displays the text sentence on the display device 23 (step S2).

Thereafter, the modification management unit 14 receives a notification of the start of editing transmitted from the user terminal device 20 (step S3).

Next, the correction management unit 14 divides the text sentence to be corrected into parts of speech, transmits the divided text sentences into parts of speech to the user terminal device 20, and displays them on the display device 23. Thereafter, upon receiving the input of the information on the cursor settings for the correction location, the correction management unit 14 causes the storage unit 15 to store correction target information including information on the position and number of characters of the word to be corrected in the text sentence (step S4 ).

After that, the modification management unit 14 receives the input of the operation information for the modification process, and modifies the characters of the word to be modified (step S5).

Then, the correction management unit 14 registers and updates the correction target term including the corrected word and the pronunciation of the term in the storage unit 15 (step S6).

Thereafter, the modification management unit 14 and the registration unit 16 determine whether the editing has been completed based on whether or not a notification of the completion of editing has been received (step S7). If the editing has not been completed (step S7: negative), the dictionary registration process returns to step S5.

On the other hand, if the editing has been completed (step S7: affirmative), the modification management section 14 deletes the modification target information from the storage section 15. Further, the registration unit 16 acquires the corrected recognition result of the term to be corrected and the reading of the term held in the storage unit 15. after that. The registration unit 16 registers the acquired term and its pronunciation as a new term in the dictionary 13 (step S8).

As described above, the information processing device according to the present embodiment divides the text sentence generated by analyzing the audio data into parts of speech and provides it to the user, and receives the user's operation to correct the term. Then, the corrected term and its pronunciation are newly registered in the dictionary. After that, the information processing device performs speech analysis using the dictionary in which the new term is registered. Thereby, dictionary registration work can be made more efficient. Furthermore, when users modify terms according to the usage environment of the speech-to-text system, new terms according to the usage environment are automatically registered in the dictionary, so the accuracy of speech analysis can be adjusted to match the usage environment. This makes it possible to improve the performance.

(Hardware configuration)
FIG. 7 is a hardware configuration diagram of the server. The server 10 shown in FIGS. 1 and 2 includes, for example, a CPU (Central Processing Unit) 91, a memory 92, a hard disk 93, and a network interface 94, as shown in FIG. The CPU 91 is connected to a memory 92, a hard disk 93, and a network interface 94 via a bus.

Network interface 94 is an interface for communication between server 10 and external devices. The network interface 94 relays communication between the CPU 91 and the user terminal device 20, for example. The network interface 94 realizes the functions of the communication control unit 11 illustrated in FIG.

The hard disk 93 is an auxiliary storage device. The hard disk 93 stores, for example, the dictionary 13 and the kanji conversion dictionary 17 illustrated in FIG. Further, the hard disk 93 realizes the function of the storage unit 15, for example. Furthermore, the hard disk 93 stores various programs including programs that implement the functions of the communication control section 11, voice analysis section 12, modification management section 14, and registration section 16 illustrated in FIG.

Memory 92 is a main storage device. For example, a DRAM (Dynamic Random Access Memory) can be used as the memory 92.

The CPU 91 reads various programs from the hard disk 93, expands them into the memory 92, and executes them. Thereby, the CPU 91 can realize the functions of the communication control section 11, voice analysis section 12, correction management section 14, and registration section 16 illustrated in FIG.

1 Speech to text system 10 Server 11 Communication control unit 12 Speech analysis unit 13 Dictionary 14 Correction management unit 15 Storage unit 16 Registration unit 17 Kanji conversion dictionary 20 User terminal device 21 Input device 22 Communication control unit 23 Display device 24 Input/output control Department

Claims (6)

obtaining recognition result sentences and reading information obtained by transcribing the audio data, which are obtained by performing speech recognition on the audio data using a dictionary;
When a first modification to a morpheme included in the recognition result sentence is detected, the recognition result of the morpheme to which the first modification has been added based on reading information for each character and each character included in the morpheme. Correlate with the reading information of
Dictionary registration characterized by causing a computer to execute a process of registering a correspondence between the recognition result after the first correction and the reading information for the morpheme in the dictionary in response to confirmation of the correction. program. Based on the operation information for moving the cursor and erasing characters, the characters modified by the first modification are identified, and the morphemes modified by the first modification are identified based on the reading information for each identified character. 2. The dictionary registration program according to claim 1, causing the computer to execute a process of associating a recognition result with reading information of each character included in the morpheme. dividing the recognition result sentence into the morphemes, retaining the position and number of characters of the morpheme to which the first modification has been applied in the recognition result sentence;
If a second modification to the character indicated by the position and the number of characters in the recognition result sentence is detected, the recognition result of the morpheme to which the first modification and the second modification are added and the morpheme included in the morpheme are determined. and the reading information of each character.
In response to the confirmation of the modification, causing the computer to execute a process of registering the correspondence between the recognition result after the first modification and the second modification and the reading information for the morpheme in the dictionary. The dictionary registration program according to claim 1, characterized in that: 2. The dictionary registration program according to claim 1, wherein the computer executes a process of associating the recognition result of the morpheme with reading information of each character included in the morpheme using a kanji conversion dictionary. The information processing device
obtaining recognition result sentences and reading information obtained by transcribing the audio data, which are obtained by performing speech recognition on the audio data using a dictionary;
When a first modification to a morpheme included in the recognition result sentence is detected, the recognition result of the morpheme to which the first modification has been added based on reading information for each character and each character included in the morpheme. Correlate with the reading information of
A dictionary registration method characterized by executing a process of registering a correspondence between the recognition result after the first modification and the reading information for the morpheme in the dictionary in response to confirmation of the modification. a speech analysis unit that obtains a recognition result sentence obtained by performing speech recognition on the speech data using a dictionary and transcribing the speech data and reading information;
When a first modification to a morpheme included in the recognition result sentence is detected, the recognition result of the morpheme to which the first modification has been added based on reading information for each character and each character included in the morpheme. a correction management unit that associates the reading information with the reading information;
and a registration unit that registers a correspondence between the recognition result after the first correction and the reading information for the morpheme in the dictionary in response to confirmation of the correction. Device.

Images