JP6142632B2 - Word dictionary registration computer program, speech synthesizer, and word dictionary registration registration method - Google Patents

Description

  The present invention relates to a speech synthesizer that synthesizes a speech signal from text data, a word dictionary registration method and a word dictionary registration computer program used in the speech synthesizer, for example.

  In recent years, speech synthesis technology for automatically synthesizing speech has been developed. Since speech synthesis technology has the advantage that it can create desired speech in a short time, such speech synthesis technology has been adopted in applications that have used pre-recorded speech by professional narrators. There are also things. In particular, in an application in which information provided at a short time interval is updated, such as a guidance broadcast in a commercial facility, a highway radio, a highway telephone, or a weather forecast broadcast, the speech synthesis technology having the above-described advantages is useful.

  In order to generate a speech signal to be synthesized, text data mixed with kanji characters is input to the speech synthesizer via, for example, a keyboard. Then, the speech synthesizer uses a word dictionary in which kanji kana notation representing a word in kanji and kana and a phonetic character string representing pronunciation of the word are registered with respect to the text data, and morphological analysis or Performs language processing such as receiving analysis. The speech synthesizer then generates a phonetic character string of the text data and an intermediate notation in which the phonetic character string is added with a prosodic symbol representing the prosody such as accent position, accent strength, or inflection magnitude. . Then, the speech synthesizer generates a synthesized speech signal based on the intermediate notation.

  Since words used in daily life change day by day, it is virtually impossible to register all words in the word dictionary in advance. Therefore, when the input text data reaches the end of the input text data, the unknown words extracted so far are collectively transmitted to the user, and the unknown word and the information about the unknown word input by the user are registered in the word dictionary. It has been proposed (see, for example, Patent Document 1).

However, there are multiple ways of reading kanji, and the way of reading kanji included in the kanji kana notation of the word differs depending on the word. In some cases, the reading of kanji included in the kanji kana notation of the word to be registered may be different from the reading when the kanji is used alone. Therefore, with the above technique, there is a possibility that the phonetic character string of the word to be registered is not correctly generated.
Also, refer to the derivation rule when there is no notation corresponding to the reading of the official address data of the input character string in the character information storage means in which the information on the reading for each character and the corresponding notation information is registered. A technique for deriving a written address has been proposed (see, for example, Patent Document 2).

JP 7-244491 A JP 2006-4069 A

  However, since the technique disclosed in Patent Document 2 does not generate a phonetic character string for speech synthesis, the word reading itself is the input reading itself. For this reason, when a word reading kana is different from an appropriate phonetic character string, the appropriate phonetic character string is not registered in the word dictionary.

  Accordingly, an object of one aspect of the present specification is to provide a word dictionary registration computer program capable of registering an appropriate phonetic character string of a word in a word dictionary used in a speech synthesizer.

  According to one embodiment, a word is registered in a word dictionary used to generate a synthesized speech signal from text data, in which a kanji kana notation of the word and a phonetic character string representing the pronunciation of the word are registered. A computer program for registering a word dictionary is provided. This computer program reads kanji corresponding to kana that does not match the kana registered in the single kanji dictionary of any kanji included in the kanji notation of the word among the kana of the word to be registered in the word dictionary. A single kanji mismatch that associates all kana characters in the reading kana of a word with one of the kanji characters in the kana notation of the word by specifying based on the kanji corresponding to the kana before or after the kana If it is determined that the sequence of vowels that can be made longer and included in the reading kana of the word based on the correspondence between the reading kana and kanji of the word to be registered corresponds to the same kanji By converting the sequence of vowels to a long sound, a phonetic character string of the word to be registered is generated, and the phonetic character string of the word to be registered is added to the word dictionary along with the kanji kana notation of the word. Having instructions thereon to perform the registering in the computer.

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 word dictionary registration computer program disclosed in this specification can register an appropriate phonetic character string of a word in a word dictionary used in a speech synthesizer.

It is a figure which shows an example of the relationship between the kanji kana notation of a word, and a reading kana, and the phonetic character string registered into a word dictionary. It is a schematic block diagram of the speech synthesizer by one Embodiment. It is a functional block diagram of the process part which the speech synthesizer by one Embodiment has. It is a figure which shows an example of a single Chinese character dictionary. It is a figure which shows an example of matching with the reading kana of the word using dynamic programming, and the reading of each kanji. It is a figure which shows another example of matching with the reading kana of the word using dynamic programming, and the reading of each kanji. It is an operation | movement flowchart of a long sound conversion process. It is an operation | movement flowchart of a word registration process.

Hereinafter, a speech synthesizer according to an embodiment will be described with reference to the drawings.
First, with reference to FIG. 1, the relationship between the kanji kana notation and the reading kana of a word and the phonetic character string registered in the word dictionary will be described. In FIG. 1, a table 100 represents a relationship between kanji kana notation in which a word is written in kanji and kana and reading kana, and table 110 shows kanji kana notation of each word and pronunciation of the word shown in table 100. The relationship of the phonetic character string to represent is represented. In this specification, in order to make the distinction between the reading kana and the phonetic character string easy to understand, in this specification, the reading kana is expressed by hiragana and the phonetic character string is expressed by katakana. In the following, kanji kana notation is simply referred to as notation.

  The phonetic character string of the word registered in the word dictionary represents the actual pronunciation of the word in the synthesized speech. Therefore, even for the same word, the reading kana and the phonetic character string may be different. In particular, in the phonetic character string, a specific vowel sequence is converted into a long sound. For example, in FIG. 1, the reading kana of the word “tofu” is “tofu”, whereas in the phonogram string, the vowel sequence “ou” is converted into a long sound and becomes “torf”. In addition, while the reading kana of the word “movie” is “Eiga”, in the phonetic character string, the sequence of “vowels” of the vowels is converted into long sounds and becomes “Aega”. In general, there are seven types of vowels that can be made longer: "Oh", "Good", "Uu", "Ee", "Oo", "Oo", and "Ei".

However, even if a sequence of these vowels is included in a word, if the sequence is lengthened, it may not be lengthened because of unnatural pronunciation. For example, in the example of FIG. 1, the word “Setouchi” includes a sequence of vowels “O” that can be made longer, but the sequence is not made longer and the phonetic character string is the same as the reading kana. It becomes "Setouchi". Similarly, the word “keystone” also includes a sequence of vowels that can be made longer, “Ei”, but that sequence is not made longer, and the phonetic string is “Kanameishi”, which is the same as the reading kana. Become.
Whether or not a sequence of vowels that can be made longer is actually made longer depends on whether the sequence of vowels is included in one kanji reading or straddles a plurality of kanji readings. That is, if the sequence of vowels that can be made longer, such as the word “Setouchi” in the above example, straddles the Chinese character “To” and the Chinese character “Uchi”, the vowel sequence Is not made longer. On the other hand, when a sequence of vowels that can be made longer like the word “tofu” corresponds to one kanji “tou”, the sequence of vowels is made longer.
Therefore, when registering a word in the word dictionary, in order to determine whether or not to actually lengthen a sequence of vowels that can be lengthened, it is necessary to determine which kanji each mora included in the reading kana corresponds to. It is important to investigate.

  Here, in order to register words, a single kanji dictionary representing the relationship between the notation for each kanji and the reading kana is used. For example, in the single kanji dictionary, “Mame” and “Tou” are associated as kana for the kanji “bean”. Therefore, when registering the word “tofu” in the word dictionary, referring to the single kanji dictionary for the kanji “tofu” included in the word notation, “tou” of the reading kana “tofu” of the word “tofu” Are associated with each other, it can be seen that the reading kana “to” corresponds to the kanji “bean”. In this way, by referring to the single kanji dictionary and checking the correspondence with the reading kana for each kanji included in the word notation, it is possible to know which kanji reading each mora is included in.

  However, some words may contain kanji that is read differently from the reading of the kanji itself, or may contain kana characters that are not included in any kanji in the word. For example, the reading name of the word “Wakaura”, which is the place name, is “Wakaura”, and the reading name includes the kana “NO”, which is not the reading of any kanji in the word “Wakaura”. In addition, the word “Ataka” representing the place name or the type of ship may be read as “Atake”. However, no matter how you combine the readings of the kanji characters “an” and “home” contained in the word “Ataka”, it does not become “Atake”.

  In this way, if a kana reading of a word cannot be expressed by a combination of readings of each kanji included in the notation of the word, even if the single kanji dictionary is referred to, each mora of the word reads which kanji. It is difficult to specify whether it is included. As a result, it is difficult to automatically convert such a word from a reading kana to a phonogram string. For example, the word “Wakaura” in the above example includes “o”, a sequence of vowels that can be made longer, in the “no” part of the reading kana. However, since it is unclear which kanji is included in the reading of the kana character "no", the speech synthesizer erroneously lengthens the vowel sequence to make the phonetic character string "wakanora" There is a fear.

  Therefore, when registering a word in the word dictionary, the speech synthesizer according to the present embodiment selects a kana that does not match any kanji reading kana included in the word notation among kana included in the reading kana of the word. The corresponding kanji is identified from the kanji corresponding to the preceding and succeeding kana. The speech synthesizer then determines whether or not the vowel sequence that can be made longer is actually made longer based on the correspondence between the reading kana and the kanji.

  FIG. 2 is a schematic configuration diagram of a speech synthesizer according to one embodiment. In the present embodiment, the speech synthesizer 1 includes an input unit 2, a storage unit 3, a processing unit 4, and an output unit 5.

The input unit 2 obtains text data that is an original text of synthesized speech and is a kanji-kana mixed text. Further, the input unit 2 acquires the notation and reading kana of the word to be registered in the word dictionary. For this purpose, the input unit 2 includes, for example, a keyboard. The input unit 2 includes a pointing device such as a mouse and a display that displays characters to be input instructed by the pointing device. Alternatively, the input unit 2 may have a touch panel display.
Furthermore, the input unit 2 may acquire text data or word notation and reading kana from another device connected to the speech synthesizer 1 via a communication network. In this case, the input unit 2 includes an interface circuit for connecting the speech synthesizer 1 to a communication network.
The input unit 2 then passes the input text data or word notation and reading kana to the processing unit 4.

The storage unit 3 includes, for example, at least one of a semiconductor memory circuit, a magnetic storage device, and an optical storage device. And the memory | storage part 3 memorize | stores the various data used for the various computer programs used by the process part 4, a speech synthesis process, or a word registration process.
The storage unit 3 stores, for example, a prosodic model and a speech waveform dictionary as data used for speech synthesis processing. Furthermore, the memory | storage part 3 memorize | stores the word dictionary where the notation of the word and the phonetic character string were registered about various words assumed to appear in text data as data used for language processing. Note that the word dictionary may further register part-of-speech information, a utilization form, and the like of each registered word. Furthermore, the memory | storage part 3 memorize | stores the single kanji dictionary showing the correspondence of the notation for every kanji and the kana which is referred when registering a word to a word dictionary.

The output unit 5 outputs the synthesized voice signal received from the processing unit 4 to the speaker 6. For this purpose, the output unit 5 includes, for example, an audio interface circuit for connecting the speaker 6 to the speech synthesizer 1.
The output unit 5 may output the synthesized speech signal to another device connected to the speech synthesizer 1 via the communication network. In this case, the output unit 5 includes an interface circuit for connecting to the speech synthesizer 1 to the communication network. In addition, when the input part 2 also acquires text data via a communication network, the input part 2 and the output part 5 may be integrated.

The processing unit 4 includes one or a plurality of processors, a memory circuit, and a peripheral circuit. Then, the processing unit 4 creates a synthesized speech signal based on the input text data.
FIG. 3 is a functional block diagram of the processing unit 4. The processing unit 4 includes a language processing unit 10, a speech synthesis unit 11, and a dictionary registration unit 12.
Each of these units included in the processing unit 4 is, for example, a functional module realized by a computer program that operates on a processor included in the processing unit 4. Or these each part which the process part 4 has may be mounted in the speech synthesizer 1 as one integrated circuit which implement | achieves the function of each part.

  The language processing unit 10 generates a phonetic character string corresponding to the input text data that is a kanji kana mixed sentence, and further generates an intermediate notation based on the phonetic character string. Here, the intermediate notation is obtained by adding a prosodic symbol representing a prosody to a phonetic character string. The prosodic symbols include, for example, symbols that represent “accent position”, “accent strength”, “pitch pitch”, “inflection magnitude”, “speech speed”, “volume level”, and “separation”. Therefore, the intermediate notation excluding the prosodic symbols matches the phonetic character string.

  The language processing unit 10 reads a word dictionary stored in the storage unit 3 in order to generate an intermediate notation from the input text data. The language processing unit 10 performs, for example, the morphological analysis and dependency analysis on the text data using the word dictionary, and the order and reading of each word appearing in the text data, the position of the accent and the delimiter Determine the position.

The language processing unit 10 can use, for example, a method using dynamic programming as the morphological analysis. The language processing unit 10 can use a syntax analysis technique such as a prefetch LR parser or an LL method, for example, as dependency analysis. Then, the language processing unit 10 creates an intermediate notation according to the order of each word, reading, accent position, and break position.
The language processing unit 10 temporarily stores the generated intermediate notation in the storage unit 3.

  The speech synthesizer 11 creates a synthesized speech signal based on the intermediate notation of the input text data.

  The speech synthesizer 11 generates a target prosody for generating a synthesized speech signal based on the intermediate notation. For this purpose, the speech synthesis unit 11 reads a plurality of prosodic models from the storage unit 3. This prosodic model represents a position in which the voice is raised and a position in which the voice is lowered in time order. Then, the speech synthesizer 11 selects a prosodic model that most closely matches the position of the accent indicated by the intermediate notation among the plural prosodic models. Then, the speech synthesizer 11 creates a target prosody by setting a position where the voice becomes high or low, a position where the voice becomes low, a voice inflection, a pitch, and the like according to the selected prosodic model and synthesis parameters. . The target prosody includes a phoneme length and a pitch frequency for each phoneme as a unit for determining a speech waveform. Note that the phoneme can be, for example, one vowel or one consonant.

The speech synthesizer 11 creates a synthesized speech signal by, for example, a phoneme connection method or a corpus-based method according to the generated target prosody.
For example, for each phoneme, the speech synthesizer 11 selects a speech waveform closest to the phoneme length and pitch frequency of the target prosody from a plurality of speech waveforms registered in the speech waveform dictionary by pattern matching, for example. For this purpose, the speech synthesis unit 11 reads a speech waveform dictionary from the storage unit 3. The speech waveform dictionary records a plurality of speech waveforms and an identification number of each speech waveform. The voice waveform is, for example, a waveform signal extracted in units of phonemes from voice signals obtained by recording various voices in which one or more narrators read various texts.
Furthermore, the speech synthesizer 11 connects the selected speech waveform and the waveform pattern of the corresponding phoneme indicated in the target prosody so that the speech waveform selected for each phoneme can be connected along the target prosody. The deviation amount may be calculated as waveform conversion information.
The speech synthesizer 11 creates waveform generation information including the identification number of the speech waveform selected for each phoneme. The waveform generation information may further include waveform conversion information.

The speech synthesizer 11 reads a speech waveform signal corresponding to the speech waveform identification number of each phoneme included in the waveform generation information from the storage unit 3. Then, the speech synthesizer 11 creates a synthesized speech signal by connecting each speech waveform signal continuously. When the waveform conversion information is included in the waveform generation information, the speech synthesizer 11 continuously connects the speech waveform signals by correcting each speech waveform signal according to the waveform conversion information obtained for the corresponding phoneme. By doing so, a synthesized speech signal is created.
The voice synthesizer 11 outputs the synthesized voice signal to the output unit 5.

  For example, the dictionary registration unit 12 determines whether or not to lengthen the arrangement of vowels that can be made longer in the reading of the word based on the notation and the reading kana of the word to be registered input via the input unit 2. Is determined, and a phonetic character string of the word is generated. The dictionary registration unit 12 registers the notation of the word together with the phonetic character string in the word dictionary. For this purpose, the dictionary registration unit 12 includes a single Chinese character division determination unit 21, a single Chinese character mismatch processing unit 22, a long sound conversion unit 23, and a registration unit 24.

  The single kanji division determination unit 21 determines whether the reading kana of the word to be registered is a combination of the reading kana registered in the single kanji dictionary of the kanji included in the notation of the word. For this purpose, the single-kanji division determination unit 21 compares the reading kana of the kanji registered in the single-kanji dictionary with the reading kana of the word in order from the first kanji included in the notation of the word to be registered. Then, the single kanji division determination unit 21 associates each kana included in the reading kana of the word with the kanji corresponding to the kana reading kana. If all kana included in the reading of a word is associated with one of the kanji included in the word notation, the reading kana of the word to be registered is the single kanji of the kanji included in the word notation It is a combination of reading kana registered in the dictionary. In this case, since it is clear which kanji each mora included in the word reading corresponds to, the long sound conversion unit 23 can immediately determine whether or not the arrangement of vowels that can be made long can actually be made long. Therefore, the single kanji division determination unit 21 notifies the long sound conversion unit 23 of the correspondence between each kana and kanji included in the reading kana of the word.

  On the other hand, if any one of the kana included in the reading of the word is not associated with any of the kanji included in the notation of the word, the kanji included in the notation of the word in the reading of the word A reading kana different from the reading kana registered in the single kanji dictionary is used. In this case, the long sound conversion unit 23 cannot determine whether or not the arrangement of vowels that can be lengthened as they are can be actually lengthened. Therefore, the single-kanji division determination unit 21 notifies the single-kanji mismatch processing unit 22 of the notation of the registration target word and the reading kana.

  FIG. 4 is a diagram illustrating an example of a single kanji dictionary. In the single kanji dictionary 400, for each kanji, one or more general readings of the kanji are registered. For example, for the Chinese character “An”, “An” and “Yasu” are registered as readings. As for the kanji “stone”, “Ishi” and “seki” are registered as readings.

For example, when “Setouchi” is input as a word to be registered, the single-kanji division determination unit 21 refers to the single-kanji dictionary 400, thereby reading the first kanji “Set” reading kana “se” and the word reading. It can be seen that the first kana “se” of the kana “Setouchi” matches. Therefore, the single kanji division determination unit 21 associates the first kana “se” in the reading kana of the word with the first kanji “se” of the word.
Next, the single Chinese character division determination unit 21 refers to the single Chinese character dictionary 400 for the next Chinese character “door” in the word “Setouchi”. Then, the single kanji division determination unit 21 determines that the kanji readings “to” and “ko” “to” are not associated with kanji in the word reading kana “setouchi”. It can be seen that it matches with “to” at the beginning of. Therefore, the single kanji division determination unit 21 associates the second kana “to” of the word reading kana with the second kanji “door” of the word. Similarly, the single kanji division determination unit 21 associates the kana “Uchi” in the reading “Kana” of the word “Setouchi” with the last Kanji “in” of the word. Thus, for the word “Setouchi”, all kana characters included in the reading kana are associated with any kanji included in the notation of the word. For this reason, the single kanji division determination unit 21 notifies the long sound conversion unit 23 of the correspondence between each kana included in the word reading kana “Setouchi” and the kanji characters “se”, “door”, “in”.

  On the other hand, it is assumed that “Wakaura” is input as a word to be registered. In this case, the single-kanji division determination unit 21 refers to the single-kanji dictionary 400 to read the first and second kanji characters “Japanese” and “song” reading kana “wa” “ka” and the word reading kana “wakana”. It can be seen that kana "wa" and "ka" of "Ura" match each other. Therefore, the single kanji division determination unit 21 associates the kana “wa” and “ka” in the reading of the word with the kanji “sum” and “song” included in the word notation. However, of the reading kana of the word, the first kana “no” that is not associated does not match the kana “ura” kana that is not associated in the word notation. For this reason, the single kanji division determination unit 21 cannot associate all kana characters included in the word “Wakaura” with kanji characters included in the notation of the word only by using kanji reading kana registered in the single kanji dictionary. . Therefore, in this example, the single kanji division determination unit 21 notifies the single kanji mismatch processing unit 22 of the word notation “Wakaura” and the reading kana “Wakaura”.

  Further, it is assumed that a word whose notation is “Ataka 1-chome” and whose reading is “Atake Ichichome” is input as a registration target. In this case, the readings of the kanji characters “A” and “Kaku” included in the word notation do not match “Atake” included in the reading of the word. Therefore, the single kanji division determination unit 21 associates all kana characters included in the word “Ataka 1-chome” with kanji characters included in the word notation only by the kana reading kana registered in the single kanji dictionary. I can't. Therefore, also in this example, the single kanji division determination unit 21 notifies the single kanji mismatch processing unit 22 of the word notation “Ataka 1-chome” and the reading kana “Atake Ichichome”.

  The single kanji mismatch processing unit 22 corresponds to the kana corresponding to the kana before or after the kana corresponding to the kana that does not match the kana of the kanji registered in the single kanji dictionary among the reading kana of the word to be registered. It is specified based on the kanji included in the word notation. Therefore, in this embodiment, the single kanji mismatch processing unit 22 uses dynamic programming.

  FIG. 5 is a diagram showing an example of correspondence between word reading kana and dynamic kanji reading using dynamic programming. In this example, the reading pseudonyms “Atake Ichichome” of the registration target word “Ataka 1-chome” are arranged in order from the left in the horizontal direction. The kana readings of each kanji included in the notation of the word are arranged in the vertical direction so that the first kanji is positioned at the bottom. In both the vertical and horizontal directions, a kana is formed for each combination of a reading kana of a word and a mora of a kanji reading kana included in the word notation. In each cell, if the vertical mora and the horizontal mora of the cell match, '0' is written. On the other hand, if they do not match, '1' is written as a penalty value. .

The single kanji mismatch processing unit 22 calculates, for each route from the lower left cell to the upper right cell, the sum of the values of each cell through which the route passes as a penalty value for the route. Then, the single kanji mismatch processing unit 22 identifies the route having the minimum penalty value. In other words, the single kanji mismatch processing unit 22 minimizes the difference between the reading kana of the word and the arrangement of the reading kana registered in the single kanji dictionary of each kanji according to the order of the kanji included in the word notation. The kana is associated with kanji that does not correspond to the reading kana registered in the single kanji dictionary of each kanji. For example, in the example shown in FIG. 5, the penalty value of the route 500 indicated by the arrow is the minimum.
Then, the single kanji mismatch processing unit 22 associates each kana included in the word reading kana with the kanji included in the word notation in accordance with the path having the minimum penalty value. In the example of FIG. 5, according to the route 500, the kana characters “I”, “Cho”, and “Me” are the kanji characters “Ichi”, “Ding”, “ ". And for “Atake”, which is the part that does not match the kanji reading kana, the first kana “a” of the kanji “an” and “home” before the associated kanji “1 chome” is the kanji “an” And the kana “take” is associated with the Chinese character “home”.

  FIG. 6 is a diagram illustrating another example of correspondence between the reading of a word using dynamic programming and the reading of each kanji. In this example, for the word “Wakaura” to be registered, the penalty value of the path 600 indicated by the solid line arrow and the penalty value of the path 601 indicated by the dotted line arrow are equal and minimum. The path 600 and the path 601 differ depending on whether the kana “no” corresponds to the kanji “song” or the kanji “ura”. In this way, when a kana different from the kana readings of two consecutive kanji characters included in the word notation is inserted, the penalty value of the two paths is the minimum. Become. In such a case, in terms of whether or not to make the vowel sequence longer, matching the inserted kana with the previous kanji matches the actual pronunciation of the word. Therefore, in this embodiment, the single kanji mismatch processing unit 22 associates the kana inserted between the readings of two kanji with the kanji corresponding to the front kana. That is, the single kanji mismatch processing unit 22 selects the lowermost path among a plurality of paths having the same penalty value. Therefore, in the example shown in FIG. 6, the single kanji mismatch processing unit 22 selects the path 600, and reads the kana words “wa”, “kan”, “ura”, and the kanji characters “Japanese”, “song”, respectively. Correspond to "Ura".

In addition, when there are a plurality of reading kana characters in the kanji included in the notation of the target word, the single kanji mismatch processing unit 22 may search for the above path in parallel for each reading kana. For example, in the example of FIG. 5, the kanji “A” has a reading Kana “Yasu” in addition to a reading Kana “An”. Accordingly, the single kanji mismatch processing unit 22 provides a square corresponding to the word reading pseudonym “Atake Ichichome” for “Yasu” in parallel with “An”, and the penalty value of the route passing through those squares. Can also be calculated.
Further, the arrangement of reading characters is not limited to the above example. For example, the single kanji mismatch processing unit 22 may arrange the reading kana of each kanji included in the word notation in order from the top, and calculate the penalty value of the route from the upper left cell to the lower right cell. .

  According to the modified example, instead of using dynamic programming, the single kanji mismatch processing unit 22 detects a part of the reading kana of the word that matches the kanji reading kana in order from the first kanji of the registered word. To do. Further, the single kanji mismatch processing unit 22 compares the kanji reading kana and the word reading kana in order from the last kanji of the word, and reads the kanji included in the registered notation of the word reading kana. The part that matches the kana is detected. Thus, the single kanji mismatch processing unit 22 identifies kana characters that do not match any kanji reading kana among the kana readings of the words. Then, the single kanji mismatch processing unit 22 associates the kana that does not match with the kanji associated with the kana immediately before.

  For example, as in the above example, it is assumed that the notation of the word to be registered is “Wakaura” and the reading kana is “Wakaura”. In this case, by comparing the kana readings of the kanji and the kana readings of the words in order from the first kanji of the word, the matching kana is identified, so that the kanji “wa” “song” and kana “wa” “ka” It is associated. In addition, the kanji “Ura” and the kana “Ura” are associated with each other by comparing the kana readings of the kanji and the kana readings of the words in order from the last kanji of the word. Finally, of the word reading kana, only the kana “no” remains without being associated with any kanji. Therefore, the single kanji mismatch processing unit 22 associates the kana “no” with the kanji “song” to which the immediately preceding kana “ka” is associated.

According to this modification, the single kanji mismatch processing unit 22 can reduce the amount of calculation for associating the reading kana of a word with a kanji rather than using dynamic matching.
Alternatively, the single kanji mismatch processing unit 22 may first perform a process of associating a kanji reading kana and a word reading kana sequentially from the beginning and the tail of the word, respectively, as in the above modification. When a plurality of kana characters are not associated with kanji, the single kanji mismatch processing unit 22 may associate the reading kana and kanji by dynamic matching. Thereby, the single kanji mismatch processing unit 22 reduces the amount of calculation when only one kana in the reading kana does not match the reading of the kanji like “Wakaura”, while reducing the calculation amount to “Ataka 1-chome”. Even when a plurality of kana characters do not coincide with kanji readings, kana and kanji characters can be appropriately associated.

  The single kanji mismatch processing unit 22 notifies the long sound conversion unit 23 of the correspondence between each kana and kanji included in the reading kana of the word.

  The long sound conversion unit 23 extracts a sequence of vowels that can be made into long sound included in the reading of the word. Then, the long sound conversion unit 23 determines whether or not the sequence of the vowels is associated with the same kanji based on the correspondence between the reading of the word and the kanji. When the vowel sequence is associated with the same Chinese character, the long sound conversion unit 23 generates a phonetic character string of the word by making the vowel sequence longer.

  FIG. 7 is a diagram showing an operation flowchart of the long sound conversion processing. The long sound conversion unit 23 obtains the total number N of mora included in the word reading kana (step S101). The long sound conversion unit 23 sets the order i from the head of the noted mora to 1 (step S102). When the total number N of mora is 1, since there is no kana to be made into a long sound, the long sound conversion unit 23 may immediately end the long sound conversion process.

  The long sound conversion unit 23 determines whether or not the i-th mora and the (i + 1) -th mora of the word reading pseudonyms include a sequence of vowels that can be made long (step S103). When the i-th mora and the (i + 1) -th mora include a sequence of vowels that can be made longer (step S103-Yes), the long-tone conversion unit 23 uses the i-th mora and the (i + 1) -th mora. Are determined to correspond to the same kanji (step S104). When the i-th mora and the (i + 1) -th mora are associated with the same kanji (step S104-Yes), the long sound conversion unit 23 sets the kana corresponding to the (i + 1) -th mora. Conversion into a long sound symbol (step S105).

  After step S105 or when the i-th mora and the (i + 1) -th mora correspond to different kanji characters in step S104 (step S104-No), the long sound conversion unit 23 selects (i + 1) Is smaller than the total number N of mora (step S106). Similarly, when the i-th mora and the (i + 1) -th mora do not include a sequence of vowels that can be made longer in step S103 (step S103-No), the long sound conversion unit 23 (i + 1) ) Is smaller than the total number N of mora (step S106).

  If (i + 1) is smaller than the total number N of mora (step S106—Yes), the long sound conversion unit 23 increments i by 1 (step S107), and repeats the processing after step S103. On the other hand, if (i + 1) is equal to or greater than the total number N of mora (No at Step S106), the long sound conversion unit 23 ends the long sound conversion process.

  The long sound conversion unit 23 passes the character string obtained after the long sound conversion process for the word reading kana together with the word notation to the registration unit 24.

  The registration unit 24 registers the obtained character string as a phonetic character string in the word dictionary together with the word notation.

  FIG. 8 is an operation flowchart of word registration processing executed by the dictionary registration unit 12 of the speech synthesizer 1. The dictionary registration unit 12 executes word registration processing according to the following operation flowchart for each word to be registered.

  The single-kanji division determination unit 21 of the dictionary registration unit 12 refers to the single-kanji dictionary and associates the kana and kanji of the registration target word (step S201). Then, the single kanji division determination unit 21 determines whether or not all kana included in the reading kana are associated with any kanji in the word (step S202).

  If any of the kana included in the reading kana is not associated with any kanji in the word (step S202-No), the single kanji division determination unit 21 stores the notation of the word and the reading kana in the dictionary registration unit 12. The single kanji mismatch processing unit 22 is notified. Then, the single kanji mismatch processing unit 22 identifies kanji corresponding to kana characters that do not match the kana readings of the kanji registered in the dictionary, based on the kanji characters corresponding to the preceding and following kana characters. Step S203). Then, the single kanji mismatch processing unit 22 notifies the long sound conversion unit 23 of the dictionary registration unit 12 of the correspondence between the reading kana and the kanji.

  On the other hand, in step S202, when all the kana characters included in the reading kana are associated with any kanji in the word (step S202-Yes), the single kanji division determination unit 21 associates the reading kana with the kanji. The relationship is notified to the long sound conversion unit 23.

When the correspondence between the reading kana and the kanji is notified, the long sound conversion unit 23 performs a long sound conversion process on the reading kana of the word to be registered based on the correspondence (step S204). Then, the long sound conversion unit 23 passes the character string obtained after the long sound conversion process for the word reading kana to the registration unit 24 of the dictionary registration unit 12 together with the word notation.
The registration unit 24 registers the character string obtained by the long sound conversion process as a phonetic character string in the word dictionary together with the word notation (step S205). Thereafter, the dictionary registration unit 12 ends the dictionary registration process.

  As described above, when registering a word in the word dictionary, the speech synthesizer automatically generates a phonogram string from the reading kana of the word. At this time, the speech synthesizer converts a kanji character corresponding to a kana character that does not match any kanji reading kana character included in the word notation from among the kana characters included in the reading kana character of the registration target word. Based on the kanji corresponding to the kana. Therefore, this speech synthesizer knows whether or not the sequence of vowels that can be converted into long tones is included in the reading of a word, so that the sequence of vowels is associated with the same kanji. It can be appropriately determined whether or not the sound is actually converted into a long sound.

  According to the modification, the single Chinese character division determination unit 21 may be omitted. In this case, the dictionary registration unit 12 executes the process of the single kanji mismatch processing unit 22 for all the words to be registered, so that each kana included in the reading kana of the word is Correspond to kanji.

  Furthermore, a computer program that causes a computer to realize each function of the processing unit of the speech synthesizer according to each of the above embodiments 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 the 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.

DESCRIPTION OF SYMBOLS 1 Speech synthesizer 2 Input part 3 Storage part 4 Processing part 5 Output part 6 Speaker 10 Language processing part 11 Speech synthesizer part 12 Dictionary registration part 21 Single Chinese character division | segmentation determination part 22 Single Chinese character mismatch processing part 23 Long sound conversion part 24 Registration part

Claims (6)

A word dictionary registration computer for registering a word in a word dictionary in which kanji kana notation of a word and a phonetic character string representing pronunciation of the word are registered, which is used to generate a synthesized speech signal from text data A program,
Among the reading kana of the word to be registered in the word dictionary, the kanji corresponding to the kana that does not match the reading kana registered in the single kanji dictionary of any kanji included in the kanji kana notation of the word is By specifying based on the kanji corresponding to the preceding or succeeding kana, the single kanji mismatch processing is performed to associate all kana included in the reading kana of the word with one of the kanji included in the kana notation of the word And
Based on the correspondence between the reading kana and kanji of the word to be registered, when it is determined that the arrangement of vowels that can be made longer is included in the reading kana of the word and corresponds to the same kanji, the arrangement of the vowels Is converted into a long sound to generate the phonogram string of the word to be registered,
Registering the phonetic character string of the word to be registered in the word dictionary together with the kanji kana notation of the word;
A computer program for registering a word dictionary for causing a computer to execute the operation.   Of the kana included in the reading kana of the word to be registered in the word dictionary, the kana matching the kana registered in the single kanji dictionary of the kanji included in the kanji kana notation of the word is associated with the kanji And causing the computer to execute the single kanji mismatch process when any of the kana included in the reading kana of the word does not correspond to any kanji included in the kanji kana notation of the word. Item 6. A computer program for registering a word dictionary according to Item 1.   Performing the single kanji mismatch processing is registered in the single kanji dictionary of each kanji according to the kanji order included in the kana kana notation of the word and the kana kana notation of the word according to dynamic programming A kana that does not correspond to any kanji included in the kanji kana notation of the word among the reading kana of the word is set as a kana before or after the kana so that the difference from the arrangement of the kana is minimized. The word dictionary registration computer program according to claim 1 or 2, wherein the computer program is associated with a corresponding kanji.   Executing the single kanji mismatch processing detects, in order from the first kanji of the word, a matching part of the reading kana registered in the single kanji dictionary of the kanji and the reading kana of the word, and In order from the last kanji of the word, by detecting a matching part of the reading kana registered in the single kanji dictionary of the kanji and the reading kana of the word, The word according to claim 1, wherein a kana that does not correspond to any kanji included in the kanji kana notation of the word is specified, and the specified kana is associated with a kanji corresponding to a kana before the kana. Computer program for dictionary registration. A speech synthesizer that generates a synthesized speech signal from text data using a word dictionary that registers a kanji kana notation of a word and a phonogram representing a pronunciation of the word,
A storage unit for storing a single kanji dictionary in which kanji notation and reading kana are registered for each kanji;
An input unit for acquiring kana kana notation and reading kana of a word to be registered in the word dictionary;
Among the reading kana of the word to be registered, the kanji corresponding to the kana that does not match the reading kana registered in the single kanji dictionary of any kanji included in the kanji kana notation of the word is displayed before the kana or A single kanji mismatch processing unit that associates all kana included in the reading kana of the word with any of the kanji included in the kana kana notation of the word by specifying the kana corresponding to the later kana,
Based on the correspondence between the reading kana and kanji of the word to be registered, when it is determined that the arrangement of vowels that can be made longer is included in the reading kana of the word and corresponds to the same kanji, the arrangement of the vowels A long sound converting unit that generates the phonetic character string of the word to be registered by converting the sound into a long sound;
A registration unit for registering the phonetic character string of the word to be registered in the word dictionary together with the kanji kana notation of the word;
A speech synthesizer. A method of registering a word in a word dictionary in which kanji kana notation of a word and a phonetic character string representing the pronunciation of the word are used to generate a synthesized speech signal from text data,
Obtain kanji kana notation and kana for the word to be registered in the word dictionary,
Among the reading kana of the word to be registered, the processor selects kanji corresponding to the kana that does not match the reading kana registered in the single kanji dictionary of any kanji included in the kanji kana notation of the word. By specifying based on the kanji corresponding to the preceding or succeeding kana, all kana included in the reading kana of the word is associated with any kanji included in the kana kana notation of the word,
If the processor determines that the sequence of vowels that can be made longer in the reading kana of the word corresponds to the same kanji based on the correspondence between the reading kana and kanji of the word to be registered, By converting the sequence of vowels to a long sound, the phonogram string of the word to be registered is generated,
The processor registers the phonetic character string of the word to be registered in the word dictionary together with the kanji kana notation of the word;
Word dictionary registration method including

Images