JP6244993B2 - Encouraging sound determination device, encouraging sound determination method, and encouraging sound determination computer program - Google Patents

Description

  The present invention relates to, for example, a prompt sound determination device, a prompt sound determination method, and a prompt sound determination computer program for determining whether or not a sound included in a phrase reading is a stutter sound or a prompt sound.

  In recent years, services that provide information by voice have been used. For example, an apparatus that provides such a service generates speech data of information by applying speech synthesis processing to text data representing a sentence describing information to be provided.

  In order to generate audio data of the information to be provided, describe "reading" that represents the audio of the word for "notation" that represents each word that appears in the sentence describing the information using kanji The dictionary is used. If a phrase that is not registered in the dictionary is included in the input text data or speech, the pronunciation of the phrase may be unknown, and thus synthesized speech that is misread may be generated. Therefore, it is preferable that as many words as possible be registered in the dictionary. For example, tens of thousands to hundreds of thousands of phrases are registered in such a dictionary.

  For this reason, when a new dictionary is created, if an operator tries to manually register words in the dictionary, the amount of work becomes enormous. Therefore, various words and phrases are extracted by a computer from an existing database that contains information on the reading of the words as well as the wording of the words, for example, various web pages, and the notation and readings of the extracted words are automatically stored in the dictionary. It is preferable to register with.

  However, the existing database cannot use the stuttering sounds such as “a”, “u”, “yo”, and the prompting sounds such as “t” as the character code, and the stuttering sound and the prompting sound are not distinguished from the direct sound. Or, there is a reading that includes a sound and is written in all capital letters. In order to exclude such a database from a database used for creating a dictionary, for example, an operator needs to visually confirm the reading and is complicated. In addition, since databases that can be used for creating a dictionary are limited, the number of words that are not registered in the dictionary may increase as a result. Therefore, it is preferable that a database in which stuttering sounds and prompt sounds are not correctly written can also be used for creating a dictionary.

  On the other hand, a technique is proposed for determining whether or not a prompt sound candidate character included in an input character string can be made a prompt sound, and displaying the character string that is stuttered or sounded when it is determined that it is possible (For example, refer to Patent Document 1). The technique described in Patent Literature 1 determines whether or not the character of the prompt sound candidate can be converted to a prompt sound by referring to a table or the like that represents a combination of the immediately preceding sound candidate and the character that can be stuttered.

JP-A-6-4520

  However, in the technique described in Patent Document 1, only a portion that can be urged to sound is displayed in the input character string, and the operator finally determines whether or not the portion should be sounded. It is necessary to judge it.

  Accordingly, an object of one aspect of the present specification is to provide a prompt sound determination device that can appropriately determine a sound to be sounded or sounded during reading of a phrase.

  According to one embodiment, a prompt sound determination device is provided. The prompt sound determination device includes notation data representing a notation of a phrase including at least one kanji and an acquisition unit that acquires reading data representing the reading of the phrase without distinguishing the stuttering sound and the prompt sound from the direct sound, and a plurality of A storage unit that stores a single kanji dictionary in which each reading of kanji and the learning type of the reading are registered, and each of at least one kanji included in the phrase represented in the notation data with reference to the single kanji dictionary Reading candidate detection unit for detecting reading candidates, and sounds included in the reading of the word or phrase represented in the reading data, among sounds that may be a prompting sound or a stuttering sound, and a sound that may include a prompting sound or a stuttering sound For each of at least one kanji included in the word represented in the notation data, and the candidate for the word A matching part that identifies the candidate that most closely matches the word and the reading of the word corresponding to the specified candidate is the reading of the kanji, and at least one kanji included in the word represented in the notation data Whether or not the prompting sound candidate is either a stuttering sound or a sounding sound according to the position of the prompting sound candidate in the reading of the first Chinese character corresponding to the prompting sound candidate and the tone type of the reading of the first Chinese character A determination unit.

The objects and advantages of the invention will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims.
It should be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are not restrictive of the invention as claimed.

  The prompt sound determination device disclosed in the present specification can appropriately determine a sound to be stuttered or sounded in reading a phrase.

It is a schematic block diagram of the dictionary registration apparatus which is an example of the prompt sound determination apparatus. It is a functional block diagram of a processing part. It is a figure which shows the arrangement pattern of the sound which may contain a roaring sound. It is a figure which shows the arrangement pattern of the sound which may contain a prompt sound. It is a figure which shows an example of the positional relationship between a stuttering candidate and the reading of corresponding kanji. It is a figure which shows an example of the positional relationship with a prompt sound candidate, the reading of corresponding kanji, and a word boundary. It is a figure which shows an example of a co-occurrence probability database. It is an operation | movement flowchart of the dictionary registration process containing a prompt sound determination process. It is a functional block diagram of the process part by a modification.

Hereinafter, the prompt sound determination device will be described with reference to the drawings.
The inventor has focused on the fact that the sound to be stuttered or sounded is included in a specific sound sequence and satisfies a specific condition regarding the position in the reading of kanji corresponding to the sound. Therefore, this prompt sound determination device acquires reading data representing the reading of a phrase expressed without distinguishing the stuttering sound and the prompt sound from the direct sound. The prompt sound determination device includes a sequence of specific sounds that may include a stuttering sound or a prompting sound included in the reading of the word of interest, and a corresponding kanji reading. Stuttering or stimulating those that satisfy the location requirements.

  FIG. 1 is a schematic configuration diagram of a dictionary registration device, which is an example of a prompt sound determination device. In the present embodiment, the dictionary registration device 1 includes a communication unit 2, a storage unit 3, a display unit 4, and a processing unit 5. The communication unit 2, the storage unit 3, and the display unit 4 are connected to the processing unit 5 via a bus.

  The communication unit 2 includes an interface circuit for connecting the dictionary registration device 1 to a communication network. The communication unit 2 is an example of an acquisition unit, and acquires notation data and reading data of a word to be registered from another device connected to the dictionary registration device 1 via a communication network. Then, the communication unit 2 passes the notation data and the reading data to the processing unit 5.

In addition, the notation of a phrase is described, for example, with a combination of kanji and kana or only kanji, while the reading of a phrase is expressed, for example, with kana. In the following, for convenience of explanation, the phrase notation is expressed as a kanji-kana mixed sentence, while the phrase reading is expressed as a katakana. In addition, the stuttering sound and the prompt sound are written in lower case letters.
In the acquired reading data, the stuttering sound and the prompt sound are described without being distinguished from the direct sound. For example, in the acquired reading data, the stuttering sound and the prompt sound are not written in lowercase kana, but are written in uppercase kana like straight sounds.
The notation data and the reading data may be represented in any file format. For example, the notation data and the reading data are represented by a text file.

  In addition, the communication unit 2 outputs the notation data of the phrase received from the processing unit 5 and the reading data corrected for the stuttering sound or the prompt sound to the device storing the dictionary in which the phrase is registered via the communication network. May be.

  The storage unit 3 includes, for example, at least one of a semiconductor memory circuit, a magnetic storage device, and an optical storage device. The storage unit 3 stores various computer programs used by the processing unit 5 and various data used for dictionary registration processing including prompting sound determination processing. For example, the storage unit 3 stores a single kanji dictionary in which various kanji notations and readings and phonetic type information indicating whether the reading is a reading or a reading are registered.

  Further, the storage unit 3 may store notation data and reading data of a word to be registered in the dictionary. Furthermore, the memory | storage part 3 may memorize | store the co-occurrence probability database showing the co-occurrence probability in which two or more Chinese characters are used side by side as one word. Or the memory | storage part 3 may memorize | store the search corpus utilized in order to determine such a co-occurrence probability database. Details of the co-occurrence probability database and the search corpus will be described later. Furthermore, the memory | storage part 3 may memorize | store the dictionary which should register the notation and reading of a phrase. The dictionary in which the notation and reading of words are registered may be a word dictionary used in speech synthesis or speech recognition, for example.

  The display unit 4 includes a display device such as a liquid crystal display, for example. Then, the display unit 4 displays, for example, the notation data of the word to be registered and the reading data of the word in which the notation of stuttering and prompting has been corrected.

  The processing unit 5 includes one or a plurality of processors, a memory circuit, and a peripheral circuit. Then, the processing unit 5 determines whether or not the sound that can be a stuttering sound or a prompting sound included in the reading data of the word to be registered is a stuttering sound or a prompting sound. If the sound is a stuttering sound or a sounding sound, the processing unit 5 corrects the notation of the sound determined to be a sounding sounding sound or sounding sound included in the reading data, and then responds to the reading of the word. Register it with the notation in the dictionary.

FIG. 2 is a functional block diagram of the processing unit 5. The processing unit 5 includes a reading candidate detection unit 11, a prompt sound candidate extraction unit 12, an association unit 13, a determination unit 14, and a registration unit 15.
Each of these units included in the processing unit 5 is, for example, a functional module realized by a computer program that operates on a processor included in the processing unit 5. Alternatively, these units included in the processing unit 5 may be mounted on the dictionary registration device 1 as one integrated circuit that realizes the functions of the units.

  The reading candidate detection unit 11 decomposes the word / phrase to be registered for each kanji included in the word / phrase notation. Then, the reading candidate detection unit 11 refers to the single kanji dictionary and detects the reading of each kanji included in the notation of the word as a reading candidate for the kanji.

  For example, when the notation of the word to be registered is “sterilization”, the kanji characters “Kan” and “fungi” are included. It is assumed that the reading of the Chinese character “Kan” registered in the single kanji dictionary is “Metsu” and “Mech”, and the reading of the Chinese character “Bacteria” is “Kin” and “Gon”. In this case, the reading candidate detection unit 11 detects “Metsu” and “Mech” as reading candidates for the Chinese character “Kan”, and detects “Kin” and “Gon” as reading candidates for the Chinese character “Bacteria”. Furthermore, the reading candidate detection unit 11 refers to the single Chinese character dictionary and acquires the phonology type information of each reading candidate.

  The reading candidate detection unit 11 outputs the reading candidates for each Chinese character included in the word / phrase to be registered and the sound lesson type information of the reading candidates to the associating unit 13.

  The prompting sound candidate extraction unit 12 extracts a sound that can be a stuttering sound or a prompting sound from the reading data of a word to be registered, and the extracted sound includes a sequence of specific sounds that may include the stuttering sound or the prompting sound In addition, the extracted sound is used as a candidate for stuttering or prompting sound. In the present embodiment, a sound candidate that can be a stuttering is a stuttering candidate, and a sound candidate that can be a prompting sound is a prompting candidate.

  For example, the prompting sound candidate extraction unit 12 extracts sounds “ya”, “yu”, and “yo” that can be a stuttering sound from the reading of a word to be registered. Then, if the sound immediately before the extracted sound is one of a plurality of specific sounds, the prompting sound candidate extraction unit 12 sets the extracted sound as a stuttering candidate.

  FIG. 3 is a diagram illustrating a sound arrangement pattern that may include stuttering. The list 300 shows a list of sounds that can be roaring. Specifically, the list 300 includes “ya”, “yu”, and “yo”. In addition, the list 310 shows a list of sounds that may be a stuttering sound when the sound immediately after is a sound included in the list 300. Specifically, the list 310 includes “ki / gi”, “shi / ji”, “chi / ji”, “ni”, “hi / bi / pi”, “mi”, “li”. .

For example, if the reading is “Jiyu”, the sound immediately before the sound “Yu” that can be a roaring sound is the sound “Ji” included in the list 310, so the prompting sound candidate extraction unit 12 changes the reading “JiYu” to The included sound “Yu” is a candidate for stuttering. If the reading is “kiyoiku”, the sound immediately before the sound “yo” that can be a roaring sound is the sound “ki” included in the list 310, and therefore the prompting sound candidate extraction unit 12 changes the reading to “kiyouiku”. The included sound “yo” is set as a stuttering candidate.
On the other hand, the reading “yaku” includes a sound “ya” that can be a roar. However, the sound “I” immediately before “YA” is not included in the list 310. For this reason, the prompting sound candidate extraction unit 12 does not set the sound “ya” included in the reading “yaku” as a stuttering candidate.

  Further, the prompting sound candidate extraction unit 12 extracts a sound “tsu” that can be a prompting sound from the reading of the word to be registered. The prompting sound candidate extraction unit 12 extracts the sound immediately before the extracted sound if it is any of the specific sounds, and the sound immediately after the extracted sound is any of the plurality of specific sounds. Sound is a candidate for prompting sound.

  FIG. 4 is a diagram illustrating a sound arrangement pattern that may include a prompt sound. The list 400 shows a list of sounds that can be prompting sounds. Specifically, the list 400 includes “tsu”. The list 410 represents a list of sounds that may be located immediately before the prompt sound. On the other hand, the list 420 represents a list of sounds that may be located immediately after the prompt sound. That is, any sound included in the list 410 is positioned immediately before the sound “tsu” included in the list 400, and any sound included in the list 420 is positioned immediately after the sound “tsu”. In this case, the sound “tsu” becomes a prompt sound candidate.

  Specifically, the list 410 includes all sounds other than “wo” and “n”, and the list 420 includes “ka line”, “sa line”, “ta line”, and “pa line”. Sounds are included.

For example, if the reading is “GATSUKO”, the sound immediately before the sound “TSU” that can be a prompt sound is the sound “GA” included in the list 410, and the sound “KO” immediately after “TSU” It is included in the list 420. Therefore, the prompting sound candidate extraction unit 12 sets the sound “tsu” included in the reading “Gatsuko” as the prompting sound candidate.
On the other hand, the reading “Foot” includes a sound “T” that can be a prompt sound. However, the sound “ U ” immediately after “ TSU ” is not included in the list 420. For this reason, the prompting sound candidate extraction unit 12 does not set the sound “tsu” included in the reading “futsu” as the prompting sound candidate.

  The prompting sound candidate extraction unit 12 outputs the stuttering sound candidate and the prompting sound candidate to the associating unit 13 and the determination unit 14.

  The associating unit 13 identifies, for each kanji included in the word to be registered, the candidate that most closely matches the reading for the word among the kanji reading candidates, and reads the word corresponding to the identified candidate. Is the reading of the kanji.

In this embodiment, the associating unit 13 replaces the sound “tu”, which is a prompting sound candidate included in the reading of the target phrase, with any of the sounds “U / K / SU / CH”, and then reads each kanji. Compare with Note that the sound “U / K / S / C” to be replaced with the prompting sound candidate is a sound that is used in reading a certain Chinese character and may be replaced with the prompting sound.
Further, the associating unit 13 regards the stuttering sound or the prompt sound included in the kanji reading candidates as being matched with the direct sound, and compares the reading of the phrase with the kanji reading candidates. For example, in a kanji reading candidate, stuttering and prompting sounds are written in lower case letters and straight sounds are written in upper case letters, the associating unit 13 distinguishes between lowercase and upper case letters of a phrase reading and a kanji reading candidate. Compare without. By performing such replacement, the associating unit 13 can associate the reading of the phrase with the reading of the kanji even for the kanji whose reading has been changed by being used in the phrase. Therefore, the associating unit 13 can accurately determine the part corresponding to each kanji included in the phrase in the phrase reading.

The associating unit 13 compares the kanji reading candidates included in the phrase with the readings of the phrase in order from the beginning of the target phrase, and determines the best matching candidate among the kanji reading candidates as the target phrase. The part corresponding to the kanji in the reading.
Alternatively, the associating unit 13 associates each kanji reading candidate combination included in the target phrase with the reading of the phrase by dynamic programming (DP) matching, thereby obtaining the best match. You may specify the combination to do. Then, the associating unit 13 may use the reading portion of the target phrase that is associated with the candidate for reading each kanji character in the combination as the reading corresponding to the kanji character.

  The associating unit 13 notifies the determination unit 14 of information indicating the correspondence between the reading of the word to be registered and each kanji included in the word, for example, the position of the reading boundary for each kanji.

  The determination unit 14 determines whether the stuttering candidate is stuttered according to the position of the stuttering candidate in the reading of the kanji included in the phrase corresponding to the stuttering candidate included in the reading of the word to be registered and the learning type of the reading of the kanji. It is determined whether or not. Similarly, the determination unit 14 determines the sound prompt according to the position of the sound prompt candidate in the reading of the Chinese character included in the word corresponding to the sound prompt candidate included in the reading of the word to be registered and the type of sound learning of the reading of the Chinese character. It is determined whether the candidate is a prompt sound.

  In the present embodiment, the determination unit 14 determines that the stuttering candidate is not a stuttering when the stuttering candidate is positioned at the head of the corresponding kanji reading or the corresponding kanji reading is kanji reading. On the other hand, the determination unit 14 determines that the stuttering candidate is a stutter if the stuttering candidate is located at a position other than the head of the corresponding kanji reading and the corresponding kanji reading is aloud.

  FIG. 5 is a diagram illustrating an example of a positional relationship between stuttering candidates and corresponding kanji readings. FIG. 5 shows three types of combinations 501 to 503 of kanji included in the registration target phrase and readings corresponding to the kanji. In the combination 501, the reading “Jiyu” is associated with the Chinese character “gun”, and the sound “Yu” is a stuttering candidate. This stuttering candidate is not the head of the reading of the corresponding Chinese character “gun”, and the reading “Jiyu” is reading aloud. Therefore, the determination unit 14 sets the stuttering candidate “Yu” as a stuttering.

  In addition, in the combination 502, the reading “Kiyo” is associated with the Chinese character “Kyo”, and the reading “Iku” is associated with the Chinese character “Education”. Of these, the sound “yo” is a stuttering candidate. This stuttering candidate is not the beginning of the reading of the corresponding Chinese character “Kyo”, and the reading “Kiyo” is a reading aloud. Therefore, the determination unit 14 sets the stuttering candidate “yo” as a stuttering.

  On the other hand, in the combination 503, the reading “di” is associated with the Chinese character “O”, and the reading “Yu” is associated with the Chinese character “Y”. Of these, the sound “Yu” is a stuttering candidate. This stuttering candidate is located at the beginning of the reading of the corresponding kanji “Yu”. Therefore, the determination unit 14 determines that the stuttering candidate “Yu” is not a stuttering.

  In addition, the determination unit 14 determines whether the prompt sound candidate is a prompt sound if the prompt sound candidate is located at a position other than the end of the corresponding kanji reading, is in a position in contact with the word boundary, or the corresponding kanji reading is a knot reading. It is determined that it is not. On the other hand, the determination unit 14 determines that the prompting candidate is a prompting sound if the prompting candidate is located at the end of the corresponding kanji reading, does not touch the word boundary, and the corresponding kanji reading is aloud reading. judge. The word boundary is a boundary between words used independently of each other.

  FIG. 6 is a diagram illustrating an example of the positional relationship between the prompt sound candidate, the reading of the corresponding kanji, and the word boundary. FIG. 6 shows three types of combinations 601 to 603 of kanji included in the registration target word / phrase and readings corresponding to the kanji. In combination 601, reading “Gatsuko” is associated with two Chinese characters “School”, and of these, sound “Tsu” is a prompt sound candidate. And this prompting sound candidate is located at the end of the reading of the corresponding kanji “gaku”, and the kanji “gaku” “school” is not a word boundary. Furthermore, the reading “Gatsuko” is a sound reading. Accordingly, the determination unit 14 uses the prompting sound candidate “TSU” as a prompting sound.

  Also, in the combination 602, the reading “Otsuki” is associated with the two Chinese characters “Otsuki”, and among these, the sound “tsu” is a prompt sound candidate. Then, since this prompt sound candidate is located at the head of the reading of the corresponding kanji character “month”, the determination unit 14 determines that the prompt sound candidate “tsu” is not a prompt sound.

  On the other hand, in the combination 603, the reading “Datsui” is associated with the two Chinese characters “Tax Emission”. Of these, the sound “tsu” is a prompt sound candidate. Although this stuttering candidate is located at the end of the reading of the corresponding kanji character “de”, the kanji character “de” and “tax” are word boundaries. Therefore, the determination unit 14 determines that the prompting sound candidate “TSU” is not a prompting sound.

  For example, the determination unit 14 refers to a co-occurrence probability database in order to determine whether a boundary between two Chinese characters corresponds to a word boundary.

FIG. 7 is a diagram illustrating an example of a co-occurrence probability database. In the co-occurrence probability database 700, the top kanji characters are shown in the top columns, and the left kanji column is the back kanji character in the two kanji rows. Is shown. The other columns of the co-occurrence probability database 700 represent the co-occurrence probabilities in which the kanji shown at the left end of the same row as the column appears after the kanji shown at the top of the column. . For example, in this example, the co-occurrence probability that the Chinese character “fungus” appears after the Chinese character “disaster” is 0.75.
Note that the co-occurrence probability database is not limited to the one that indicates the co-occurrence probability of a combination of two Chinese characters, and one or both may indicate the co-occurrence probability of a combination of words including two or more Chinese characters.

  The determination unit 14 determines the co-occurrence probability for the combination of the Chinese character whose sounding sound candidate is located at the end of the reading and the next Chinese character with reference to the co-occurrence probability database. When the co-occurrence probability is less than a predetermined threshold, the determination unit 14 determines that the boundary between the two Chinese characters is a word boundary. On the other hand, when the co-occurrence probability is equal to or higher than the predetermined threshold, It is determined that the boundary between two Chinese characters is not a word boundary. The predetermined threshold is set to 0.6, for example.

  Furthermore, the determination unit 14 may use the beginning and the end of the registration target word as word boundaries.

  Moreover, the determination part 14 may determine the position of a word boundary according to another method. For example, the determination unit 14 may apply morphological analysis to a word / phrase to be registered and divide the word / phrase into morphemes. And the determination part 14 is good also considering the boundary between morphemes as a word boundary.

  Or the determination part 14 is good also as a word boundary between a suffix and the Chinese character immediately before it. Similarly, the determination unit 14 may use a word boundary between the prefix and the Chinese character immediately after that. For example, in a phrase with a prefix such as “first XX” or “live XX” (where ◯ represents an arbitrary Chinese character), the determination unit 14 uses “first” and “active” as prefixes. , “First” or “Life” and the kanji immediately after that may be used as a word boundary. In addition, in a phrase with a suffix such as “XX period” (where “○” represents an arbitrary kanji character), the determination unit 14 determines that “period” and the preceding kanji character It is good also as a word boundary.

  Or the determination part 14 is good also considering the boundary of the Chinese character used as a part of a person's name, and the Chinese character immediately before or immediately after that as a word boundary. The kanji used as part of the personal name is, for example, “child”, “husband”, “male”, “male”, or “beauty”. For example, in a phrase such as “XX child” (where “○” represents an arbitrary kanji character), the determination unit 14 determines that the “child” is a kanji character representing a part of a person's name, so May be used as a word boundary.

  The determination unit 14 corrects the stuttering candidate determined to be stuttering in the reading data to stuttering. Similarly, the determination unit 14 corrects the prompting sound candidate determined to be the prompting sound in the reading data to the prompting sound. For example, when the reading data is text data, the determination unit 14 corrects the stuttering sound and the prompt sound to lowercase notation. The determination unit 14 includes notation data of a word to be registered and reading data (hereinafter referred to as corrected reading data) of the word in which a sound determined to be a stuttering sound and a sounding sound is corrected among the sounding sound candidates and the sounding sound candidates. Is displayed on the display unit 4.

  The registration unit 15 registers the notation data of the word to be registered and the corrected reading data in the dictionary stored in the storage unit 3. When the dictionary in which the word / phrase is to be registered is stored in another device connected to the dictionary registration device 1 via the communication network, the registration unit 15 displays the notation data of the word / phrase to be registered and the corrected reading. Data may be transmitted to another device via the communication unit 2.

  FIG. 8 is an operation flowchart of dictionary registration processing including prompting sound determination processing executed by the processing unit 5 of the dictionary registration device 1. In the operation flowchart shown below, the processing in steps S101 to S108 corresponds to the prompt sound determination processing. Each time the processing unit 5 acquires notation data and reading data of a registration target phrase via the communication unit 2, the processing unit 5 executes a dictionary registration process according to the following operation flowchart.

  The reading candidate detection unit 11 refers to the single kanji dictionary and detects the reading of the kanji as a reading candidate for each kanji included in the word to be registered (step S101). The reading candidate detection unit 11 outputs the reading candidates for each Chinese character included in the word / phrase to be registered and the sound lesson type information of the reading candidates to the associating unit 13.

  The prompting sound candidate extraction unit 12 extracts, as a stuttering candidate, a sound included in a sequence of specific sounds that may include stuttering among sounds that can be a stuttering included in the reading of the word to be registered (step S102). Further, the prompting sound candidate extraction unit 12 extracts a sound included in a specific sound that may include a prompting sound as a prompting sound candidate among the sounds that can be the prompting sound included in the reading of the registration target phrase (step S103). ). The prompt sound candidate extraction unit 12 outputs the extracted stutter sound candidate and the prompt sound candidate to the association unit 13 and the determination unit 14.

  The associating unit 13 compares, for each kanji included in the word to be registered, a candidate for reading the kanji and a reading for the word, and determines a portion corresponding to each kanji in the word reading. Specify (step S104). The associating unit 13 notifies the determining unit 14 of information indicating the correspondence between the reading of the word to be registered and each kanji.

The determination unit 14 determines whether the stuttering candidate is located at a position other than the head of the corresponding kanji reading and the corresponding kanji reading is a sound reading (step S105). If the stuttering candidate is located at a position other than the beginning of the corresponding kanji reading and the corresponding kanji reading is aloud (step S105-Yes), the determination unit 14 determines that the stuttering candidate is a stuttering. Then, the determination unit 14 corrects the stuttering candidate to stuttering in the reading data (step S106).
On the other hand, if the stuttering candidate is located at the head of the corresponding kanji reading or if the corresponding kanji reading is a kanji reading (step S105-No), the determination unit 14 determines that the stuttering candidate is not a stuttering. .

After step S106, or after the determination unit 14 determines that the stuttering candidate is not a stuttering, the determination unit 14 is positioned at the end of the corresponding kanji reading, does not touch the word boundary, and Then, it is determined whether or not the corresponding kanji reading is a sound reading (step S107). If the prompt sound candidate is located at the end of the corresponding kanji reading, does not touch the word boundary, and the corresponding kanji reading is a sound reading (step S107—Yes), the determination unit 14 determines that the prompt sound candidate is It is determined that the sound is prompt. Then, the determination unit 14 corrects the prompting sound candidate to the prompting sound in the reading data (step S108).
On the other hand, if the prompt sound candidate is located at a position other than the end of the corresponding kanji reading, touches a word boundary, or if the corresponding kanji reading is a knot reading (step S107-No), the determination unit 14 It is determined that the sound prompt candidate is not a sound prompt.

After step S108, or after the determination unit 14 determines that the prompting sound candidate is not a prompting sound, the registration unit 15 registers the notation data of the word to be registered and the corrected reading data in the dictionary (step S109). Thereafter, the processing unit 5 ends the dictionary registration process.
Note that the processing unit 5 may interchange the order of the processing in step S101 and the processing in steps S102 and S103. Further, the processing unit 5 may exchange the order of the processes in steps S105 and S106 and the processes in steps S107 and S108.

  As described above, this dictionary registration device is adapted for the position of the corresponding kanji reading at which the specific sound that may contain stuttering or sounding sounds and the included sounding or sounding sound candidates can be stuttering or sounding sounds. And whether or not the corresponding kanji reading is learned. Thereby, this dictionary registration apparatus can determine appropriately whether a stuttering candidate and a prompting sound candidate are a stuttering sound or a sounding sound, respectively. Therefore, this dictionary registration device automatically corrects the notation of stuttering and sounding sounds and automatically reads the wording and notation data without correcting the notation of stuttering and sounding sounds in the reading data manually. Since data can be registered in the dictionary, man-hours for dictionary creation can be reduced.

  Note that the processing unit 5 may obtain the co-occurrence probability for the combination of the Chinese character corresponding to the prompt sound candidate and the Chinese character immediately thereafter based on the search corpus.

  FIG. 9 is a functional block diagram of the processing unit 5 of the dictionary registration device 1 according to a modification. In this modification, the processing unit 5 includes a reading candidate detection unit 11, a prompt sound candidate extraction unit 12, an association unit 13, a determination unit 14, a registration unit 15, and a co-occurrence probability calculation unit 16. . The processing unit 5 according to this modification is different from the processing unit 5 shown in FIG. 2 in that it has a co-occurrence probability calculation unit 16. Therefore, in the following, the co-occurrence probability calculation unit 16 and its related parts will be described.

  The determination unit 14 notifies the co-occurrence probability calculation unit 16 of the combination of the kanji and the next kanji as the combination of kanjis for which the co-occurrence probability is to be calculated when the prompt sound candidate is located at the end of the corresponding kanji reading. To do.

  The co-occurrence probability calculating unit 16 obtains a search corpus including a combination of kanji for which the co-occurrence probability is to be calculated. The co-occurrence probability calculating unit 16 uses, for example, a kanji-kana mixed sentence included in a Web page published on various Web sites connected to the dictionary registration device 1 via a communication network as a search corpus. Text data can be used. Alternatively, for example, text data of a kanji mixed text included in a product manual, minutes, e-mail, etc. can be used as a search corpus. For example, the co-occurrence probability calculation unit 16 reads these search corpora from the storage unit 3. Alternatively, the co-occurrence probability calculation unit 16 acquires a search corpus from the other device connected to the dictionary registration device 1 via the communication unit 2 via the communication network.

For example, the co-occurrence probability calculation unit 16 counts the number of combinations of kanji for which the co-occurrence probability is to be calculated and the number of each kanji included in the combination included in the text data included in the search corpus. . Then, the co-occurrence probability calculating unit 16 calculates the co-occurrence probability CoP according to the following equation.
CoP = 2N ₁₂ / (N ₁ + N ₂ )
Here, N ₁ represents the number of leading Chinese characters included in the combination of kanji for which the co-occurrence probability is to be calculated, and N ₂ represents the number of subsequent kanji characters included in the combination of kanji for which the co-occurrence probability is to be calculated. Represents. N ₁₂ represents the number of combinations of the Chinese characters.
For example, the combination of kanji for which the co-occurrence probability is to be calculated is “sterilization”, and three combinations “sterilization” are included in the search corpus, and each of the kanji “disaster” and “fungus” is four. Suppose that they are included. In this case, the co-occurrence probability Cop is 2 * 3 / (4 + 4) = 0.75.
In addition, the co-occurrence probability calculating unit 16 may calculate the co-occurrence probability by using the number of web search service searches as another method for calculating the co-occurrence probability. When using the Web search service, the co-occurrence probability calculating unit 16 calculates the co-occurrence probability CoP using the following equation.
CoP = (number of searches for word combinations) / (number of searches for word 1 and word 2 and conditions)
For example, if the search results when searching for “sterilization” using the Web search service and searching for “sterile and germs” are 1,270,000 and 1,750,000, respectively, the co-occurrence probability CoP is 1,270,000 / 1,750,000 = 0.72.

The co-occurrence probability calculation unit 16 notifies the determination unit 14 of the calculated co-occurrence probability. And the determination part 14 should just determine whether a prompt sound candidate touches the word boundary based on the notified co-occurrence probability.
In addition, the co-occurrence probability calculating unit 16 may add the combination of kanji and the co-occurrence probability that are the calculation targets to the co-occurrence probability database.

  According to this modified example, the dictionary registration device 1 does not know the co-occurrence probability for the combination of kanji characters related to the prompt sound candidate obtained from the word to be registered, but the co-occurrence probability for the combination. Can be calculated. Therefore, the dictionary registration apparatus 1 can determine whether or not the prompt sound candidate is in contact with the word boundary without having a large-scale co-occurrence probability database that represents the co-occurrence probabilities for a large number of combinations of Chinese characters in advance. It is possible to appropriately determine whether the prompt sound candidate is a prompt sound.

According to another modification, the storage unit 3 may store an exception list that uses the prompting sound candidate as the prompting sound for the combination of the Chinese character corresponding to the prompting sound candidate and the preceding and following Chinese characters. Then, when the combination of the kanji corresponding to the prompt sound candidate and the preceding and following kanji is registered in the exception list, the determination unit 14 determines whether the prompt sound candidate is in contact with the word boundary and the kanji corresponding to the prompt sound candidate. Regardless of the reading lesson type, it may be a prompt sound candidate and a prompt sound.
For example, a proper noun, such as a name of a person such as “Kappei” or a place name such as “Tottori”, may contain a sounding sound even if the kanji is a kanji reading. . Therefore, by registering such an exceptional combination of kanji in advance as an exception list, the determination unit 14 can determine whether or not the prompt sound candidate is a prompt sound more accurately.

  According to still another modification, the dictionary registration device may acquire a database including notation data of a character string including a plurality of words and reading data of the character string via the communication unit 2, for example. . In this case, the processing unit 5 decomposes the character string for each word / phrase by applying, for example, morphological analysis to the character string represented in the notation data. And the process part 5 may perform the dictionary registration process by said embodiment or modification with respect to the phrase which is not registered into a dictionary among the acquired phrases. Thus, the dictionary registration device can automatically register a plurality of words / phrases in a dictionary from one database.

  Furthermore, the dictionary registration device according to the above-described embodiment or modification may read notation data and reading data of a word to be registered from a recording medium such as a magnetic recording medium or an optical recording medium in which those data are recorded. In this case, the dictionary registration device may have an access device for such a recording medium as the acquisition unit.

  Furthermore, a computer program that causes a computer to realize each function of the processing unit of the dictionary registration device according to the above-described embodiment or modification is recorded on a computer-readable medium, for example, a magnetic recording medium, an optical recording medium, or a semiconductor memory. It may be provided in a customized form.

  All examples and specific terms listed herein are intended for instructional purposes to help the reader understand the concepts contributed by the inventor to the present invention and the promotion of the technology. It should be construed that it is not limited to the construction of any example herein, such specific examples and conditions, with respect to showing the superiority and inferiority of the present invention. Although embodiments of the present invention have been described in detail, it should be understood that various changes, substitutions and modifications can be made thereto without departing from the spirit and scope of the present invention.

The following supplementary notes are further disclosed regarding the embodiment described above and its modifications.
(Appendix 1)
An acquisition unit for acquiring notation data representing a notation of a word including at least one kanji and reading data representing the reading of the word without distinguishing stuttering and prompt sounds and direct sounds;
A storage unit for storing a single kanji dictionary in which each reading of a plurality of kanji characters and a phonetic type indicating whether the reading is a phonetic reading or a kanji reading is registered;
A candidate reading detection unit that detects each candidate for reading of the at least one kanji included in the phrase represented in the notation data with reference to the single kanji dictionary;
A prompting sound candidate that extracts a sound that may be a prompting sound or a stuttering sound included in the reading of the word represented in the reading data as a prompting sound candidate. An extractor;
For each of the at least one kanji included in the phrase, a candidate that most closely matches the reading of the phrase among the candidate readings of the kanji is identified, and the reading portion of the phrase corresponding to the identified candidate Is an association part for reading the kanji,
According to the position of the prompt sound candidate in the reading of the first Chinese character corresponding to the prompt sound candidate among the at least one Chinese character included in the word and the learning type of the reading of the first Chinese character A determination unit that determines whether the prompt sound candidate is either a stutter sound or a prompt sound;
A prompt sound determination device having
(Appendix 2)
The determination unit is a sound that the prompting sound candidate can be a prompting sound, is located at the end of the reading of the first kanji and does not touch a boundary between words used independently of each other, The prompt sound determination device according to appendix 1, wherein when the reading of the first Chinese character is a sound reading, the prompt sound candidate is a prompt sound.
(Appendix 3)
When the co-occurrence probability that the first kanji and the second kanji following the first kanji are used in combination is less than a predetermined threshold, the determination unit determines that the first kanji and the first kanji The prompt sound determination apparatus according to appendix 2, wherein a boundary between two Chinese characters is determined to be a boundary between the words.
(Appendix 4)
The determination unit, when the prompting sound candidate is a sound that can be a stuttering and is located at a position other than the beginning of the reading of the first kanji, and the reading of the first kanji is a reading aloud, 4. The prompt sound determination apparatus according to any one of appendices 1 to 3, wherein the prompt sound candidate is a stutter.
(Appendix 5)
The association unit replaces a sound that can be a prompt sound among the prompt sound candidates with a sound that may be replaced with a prompt sound, and for each of the at least one Chinese character included in the phrase, The prompting sound determination device according to any one of appendices 1 to 4, wherein a candidate that most closely matches the reading of the word is identified among the reading candidates of the above.
(Appendix 6)
The associating unit regards the stuttering sound and the prompt sound included in the candidate for reading the kanji as matching the sound obtained by directing the stuttering sound and the prompt sound, and for each of the at least one kanji character included in the phrase, The prompting sound determination device according to any one of appendices 1 to 5, wherein a candidate that most closely matches the reading of the word is identified among candidates for reading the kanji.
(Appendix 7)
And a registration unit for registering the notation data and the reading data obtained by correcting the prompting sound candidate determined to be a prompting sound or a stuttering sound into a notation representing a prompting sound or a stuttering in a dictionary representing a relationship between the notation of the phrase and the reading. The prompt sound determination device according to any one of appendices 1 to 6.
(Appendix 8)
The phrase represented in the notation data representing the notation of a word including at least one kanji, with reference to a single kanji dictionary in which each kanji reading and a phonetic type indicating whether the reading is a kanji reading or a kanji reading is registered A candidate for reading each of the at least one kanji included in
A series of sounds that may contain a sound or a stutter among the sounds that can be a sound or a sound included in the reading of the word represented in the reading data expressed without distinguishing between a sound and a sound. To be extracted as a prompt sound candidate,
Extracting sounds included in the reading of the word represented in the reading data, which may be a prompting sound or a stuttering sound, that may contain a prompting sound or a stuttering sound as a prompting sound candidate,
For each of the at least one kanji included in the phrase, a candidate that most closely matches the reading of the phrase among the candidate readings of the kanji is identified, and the reading portion of the phrase corresponding to the identified candidate As the reading of the kanji
According to the position of the prompt sound candidate in the reading of the first Chinese character corresponding to the prompt sound candidate among the at least one Chinese character included in the word and the learning type of the reading of the first Chinese character Determining whether the prompting sound candidate is either a stuttering sound or a prompting sound;
A prompt sound determination method including the above.
(Appendix 9)
The phrase represented in the notation data representing the notation of a word including at least one kanji, with reference to a single kanji dictionary in which each kanji reading and a phonetic type indicating whether the reading is a kanji reading or a kanji reading is registered A candidate for reading each of the at least one kanji included in
A series of sounds that may contain a sound or a stutter among the sounds that can be a sound or a sound included in the reading of the word represented in the reading data expressed without distinguishing between a sound and a sound. To be extracted as a prompt sound candidate,
For each of the at least one kanji included in the phrase, a candidate that most closely matches the reading of the phrase among the candidate readings of the kanji is identified, and the reading portion of the phrase corresponding to the identified candidate As the reading of the kanji
According to the position of the prompt sound candidate in the reading of the first Chinese character corresponding to the prompt sound candidate among the at least one Chinese character included in the word and the learning type of the reading of the first Chinese character Determining whether the prompting sound candidate is either a stuttering sound or a prompting sound;
A computer program for determining a prompting sound for causing a computer to execute the above.

1 Dictionary registration device (encouragement sound judgment device)
2 communication unit 3 storage unit 4 display unit 5 processing unit 11 reading candidate detection unit 12 prompt sound candidate extraction unit 13 association unit 14 determination unit 15 registration unit 16 co-occurrence probability calculation unit

Claims (7)

An acquisition unit for acquiring notation data representing a notation of a word including at least one kanji and reading data representing the reading of the word without distinguishing stuttering and prompt sounds and direct sounds;
A storage unit for storing a single kanji dictionary in which each reading of a plurality of kanji characters and a phonetic type indicating whether the reading is a phonetic reading or a kanji reading is registered;
A candidate reading detection unit that detects each candidate for reading of the at least one kanji included in the phrase represented in the notation data with reference to the single kanji dictionary;
A prompting sound candidate that extracts a sound that may be a prompting sound or a stuttering sound included in the reading of the word represented in the reading data as a prompting sound candidate. An extractor;
For each of the at least one kanji included in the phrase, a candidate that most closely matches the reading of the phrase among the candidate readings of the kanji is identified, and the reading portion of the phrase corresponding to the identified candidate Is an association part for reading the kanji,
According to the position of the prompt sound candidate in the reading of the first Chinese character corresponding to the prompt sound candidate among the at least one Chinese character included in the word and the learning type of the reading of the first Chinese character A determination unit that determines whether the prompt sound candidate is either a stutter sound or a prompt sound;
I have a,
The determination unit determines a boundary between words that are used independently of each other for a combination of the first kanji and the kanji following the first kanji when the prompting sound candidate is a sound that can be a sound. By applying a predetermined method of detecting, it is determined whether or not the boundary between the first kanji and the kanji following the first kanji is the boundary, and the reading in the reading of the first kanji Depending on the position of the prompt sound candidate, whether the prompt sound candidate touches the boundary, and whether or not the prompt sound candidate is the prompt sound according to the reading type of the reading of the first kanji Prompt sound determination device. The determination unit, the screw geminate consonant candidate is a sound that can be a double consonant, located at the end of the reading of the first Chinese character, and not in contact with the boundary, and the first Chinese character The prompting sound determination device according to claim 1, wherein when the reading is a sound reading, the prompting sound candidate is a prompting sound.   The determination unit, when the prompting sound candidate is a sound that can be a stuttering and is located at a position other than the beginning of the reading of the first kanji, and the reading of the first kanji is a reading aloud, The prompt sound determination apparatus according to claim 1, wherein the prompt sound candidate is a stutter.   The association unit replaces a sound that can be a prompt sound among the prompt sound candidates with a sound that may be replaced with a prompt sound, and for each of the at least one Chinese character included in the phrase, The prompt sound determination apparatus according to any one of claims 1 to 3, wherein a candidate that most closely matches the reading of the word is specified among the reading candidates.   The associating unit regards the stuttering sound and the prompt sound included in the candidate for reading the kanji as matching the sound obtained by directing the stuttering sound and the prompt sound, and for each of the at least one kanji character included in the phrase, The prompting sound determination apparatus according to any one of claims 1 to 4, wherein a candidate that most closely matches the reading of the word is identified among candidates for reading the kanji. The computer refers to a single kanji dictionary in which each kanji character reading and a phonetic type indicating whether the reading is sound reading or kanji reading is registered, and is represented in notation data representing the notation of a word including at least one kanji. Detecting candidates for reading each of the at least one kanji included in the word,
There is a possibility that a sound or a stuttering sound is included among the sounds that can be a sound or a sound that is included in the reading of the word or phrase that is expressed by the computer without distinguishing between the sound and the sound and the sound. Extract sounds included in the sequence as prompting sound candidates,
The computer extracts sounds included in the reading of the word represented in the reading data, which may be a prompting sound or a stuttering sound, which may contain a prompting sound or a stuttering sound, as a prompting sound candidate And
The computer specifies, for each of the at least one kanji included in the phrase, a candidate that most closely matches the reading of the phrase among candidates for reading the kanji, and the phrase corresponding to the identified candidate The part of reading is the reading of the kanji,
The computer has the position of the prompting sound candidate in the reading of the first Chinese character corresponding to the prompting sound candidate of the at least one Chinese character included in the phrase, and the phonetic of the reading of the first Chinese character According to the type, it is determined whether the prompting sound candidate is either a stuttering sound or a prompting sound,
Look at including it,
The determination of whether the prompt sound candidate is either a stutter sound or a sound is a combination of the first Chinese character and the Chinese character that follows the first Chinese character when the prompt sound candidate is a sound that can be a prompt sound. On the other hand, by applying a predetermined method for detecting a boundary between words used independently of each other, whether or not the boundary between the first kanji and the kanji following the first kanji is the boundary. Depending on the position of the prompting sound candidate in the reading of the first Chinese character, whether the prompting sound candidate is in contact with the boundary, and the learning type of the reading of the first Chinese character Determine whether the prompt sound candidate is a prompt sound
Including a screw double consonants determination method that. The phrase represented in the notation data representing the notation of a word including at least one kanji, with reference to a single kanji dictionary in which each kanji reading and a phonetic type indicating whether the reading is a kanji reading or a kanji reading is registered A candidate for reading each of the at least one kanji included in
A series of sounds that may contain a sound or a stutter among the sounds that can be a sound or a sound included in the reading of the word represented in the reading data expressed without distinguishing between a sound and a sound. To be extracted as a prompt sound candidate,
For each of the at least one kanji included in the phrase, a candidate that most closely matches the reading of the phrase among the candidate readings of the kanji is identified, and the reading portion of the phrase corresponding to the identified candidate As the reading of the kanji
According to the position of the prompt sound candidate in the reading of the first Chinese character corresponding to the prompt sound candidate among the at least one Chinese character included in the word and the learning type of the reading of the first Chinese character Determining whether the prompting sound candidate is either a stuttering sound or a prompting sound;
Let the computer do
The determination of whether the prompt sound candidate is either a stutter sound or a sound is a combination of the first Chinese character and the Chinese character that follows the first Chinese character when the prompt sound candidate is a sound that can be a prompt sound. On the other hand, by applying a predetermined method for detecting a boundary between words used independently of each other, whether or not the boundary between the first kanji and the kanji following the first kanji is the boundary. Depending on the position of the prompting sound candidate in the reading of the first Chinese character, whether the prompting sound candidate is in contact with the boundary, and the learning type of the reading of the first Chinese character Determine whether the prompt sound candidate is a prompt sound
Screw assimilated sound judgment computer program includes.

Images