JP5445244B2 - Speech synthesis apparatus, speech synthesis method, and speech synthesis program - Google Patents

Description

  The present invention relates to a method for using a user dictionary used during speech synthesis.

  In the speech synthesis technology for inputting text and generating the reading speech, it is necessary to register many words in a language dictionary in which the reading for the word notation is registered in order to increase the correct reading rate of the text. However, as the number of registrations increases, the number of misreads with the same notation increases, causing a reading error.

  Therefore, a technique is disclosed that uses a user dictionary that is independently managed by a user as needed, based on a basic dictionary that collects words that frequently appear in Japanese. A method for selectively using specialized dictionaries that collect technical terms for each field is also disclosed. In addition, a method using another person's user dictionary is also disclosed (see, for example, Patent Document 1, Patent Document 2, Patent Document 3, Patent Document 4, and Patent Document 5).

Japanese Patent No. 3894479 JP 2007-206975 A Japanese Patent No. 3220096 JP-A-11-327871 Japanese Patent Laid-Open No. 2007-80019

  However, these methods have the following problems. First, when using a user dictionary, it is necessary to register a certain number of words by yourself, and the burden of registration work is large for the user.

  Also, when preparing specialized dictionaries by field in advance, it is practically impossible to prepare specialized dictionaries that cannot handle new words or that cover all the vast Japanese vocabulary. There is a problem. Furthermore, when a specialized dictionary is used, there is a problem that a specialized dictionary in an appropriate field cannot be selected unless the way of classifying the field is common between the user and the creator of the specialized dictionary.

  In the method of using another person's user dictionary, a plurality of other person's user dictionaries are examined for a certain notation unknown word, and a reading with a high registration rate is taken into the basic dictionary as a reading for the unknown word. However, in this method, since only one reading can be registered for one notation, it is not possible to cope with the same notation.

  Also, by comparing the degree of similarity between your user dictionary and another person's user dictionary, you can incorporate words that are included in another person's dictionary with a high degree of similarity but not in your own dictionary into your own dictionary. The user dictionaries are individually managed by individuals, and there is a problem that the probability that similar words are registered is low. For example, suppose that Mr. A is interested in politics, economy, and baseball, synthesizes those sentences well, and the words related to them are registered in the user. In addition, Mr. B is interested in performing arts, weather, and baseball, and registers words related to them as a user. Here, looking at the similarity between the user dictionary registered words of Mr. A and Mr. B, only the baseball part is similar, and the parts related to entertainment and weather are not similar. Therefore, the ratio of data that is not similar increases, and Mr. A and Mr. B cannot share the user dictionary. Also, if both A and B have registered only baseball-related words, even if the degree of similarity is high, the number of words that can be taken into A's dictionary because the degree of similarity is high, and the sharing effect Can not get much. Thus, the probability of sharing is small, and even if it can be shared, the effect is small.

  Therefore, the present invention aims to improve the reading accuracy rate even for unknown words that are not registered in one's user dictionary by effectively using another person's user dictionary. To do.

To achieve the above object, the speech synthesis apparatus disclosed below, the headword, the headword readings, and the user of the user dictionary to store in association with surrounding text having relevance to the entry word, the basic A text input unit that is accessible to the dictionary and receives input of the text to be synthesized , and a plurality of other persons that associate and store the headword, the reading of the headword, and the surrounding text having relevance to the headword unknown word extraction unit that extracts of the interface unit the user dictionary to be referenced, the unknown word from the compositing target text accepts an input by the text input unit not included in the basic dictionary in the user dictionary of the user When, the peripheral text with relevance to the unknown word from the synthesis target text that contains the unknown word which the unknown word extraction unit has extracted extraction Headword extraction and peripheral text extraction unit, through said interface unit refers to the user dictionary of the others, the entry word that matches the user dictionary of the person to the unknown word extracted is the unknown word extraction section that an extraction unit, the degree of coincidence between said peripheral text from said user dictionary and the peripheral text associated with the entry word which the index word extracting unit has extracted from the compositing target text surrounding the text extraction unit has extracted the others a match degree calculating section for calculating, based on said degrees of coincidence the coincidence degree calculation unit has calculated, the reading from the user dictionary of the others of the entry word which the index word extracting unit has extracted the from the compositing target text and a reading determination unit which determines whether the reading of the unknown word to the unknown word extraction portion is extracted.

  According to the above configuration, it is possible to improve the correct reading rate even for unknown words that are not registered in the user dictionary by effectively using the user dictionary of another person.

1 is a block diagram showing an example of the overall configuration of a speech synthesizer according to Embodiment 1 of the present invention. The flowchart which shows an example of operation | movement of the speech synthesizer which concerns on Embodiment 1 of this invention. Diagram showing examples of various data The block diagram which shows an example of a structure of the speech synthesizer concerning Embodiment 2 of this invention. The block diagram which shows an example of the whole structure of the speech synthesizer which concerns on Embodiment 3 of this invention.

[Embodiment 1]
FIG. 1 is a block diagram showing the overall configuration of a speech synthesis apparatus 100 according to Embodiment 1 of the present invention. In FIG. 1, the speech synthesizer 100 includes a basic dictionary 110, a user dictionary 120, a user dictionary interface unit 130, a text input unit 140, a language processing unit 150, a waveform processing unit 160, and a speech output unit 170.

  The basic dictionary 110 is a dictionary that stores basic words necessary for speech synthesis. The basic dictionary may be referred to as a standard dictionary. The user dictionary 120 is a user-specific dictionary in which a user registers words as needed. The other person's user dictionaries 121, 122, and 123 are dictionaries unique to other users in which other users individually register words at any time. The basic dictionary 110 stores, for example, notation of words that frequently appear in Japanese (= headwords), reading of the headwords, accents, and the like.

  The user dictionary 120 stores a notation of a word registered by the user (= headword), reading of the headword, accents, and peripheral text related to the headword. The other person's user dictionaries 121, 122, 123 store notation of words registered by other users (= headwords), readings of the headwords, accents, and peripheral texts related to the headwords. Yes. The basic dictionary 110 may store peripheral text having relevance to the headword.

  The speech synthesizer 100 can exchange user dictionary information between different users via the user dictionary interface unit 130 and the interface units 131 to 133 of another person's user dictionary.

  Text input unit 140 accepts input of text to be synthesized. For example, a configuration in which a user inputs text through a keyboard, a configuration in which electronic input is performed through a drive that reads a medium such as a CD or a flexible disk, a configuration in which text read by an OCR by a scanner is input, or wired or A configuration in which text is electronically received via a wireless network, or a combination thereof may be used.

  The input text is sent to the language processing unit 150. The language processing unit 150 analyzes and outputs the input text reading, accent, and the like. The language processing unit 150 includes a morphological analysis unit 151, an unknown word extraction unit 152, a surrounding text extraction unit 153, a headword extraction unit 154, a feature level setting unit 155, a matching degree calculation unit 156, and a reading determination unit 157.

  The morpheme analysis unit 151 performs morpheme analysis using the basic dictionary 110 and the user dictionary 120 to determine readings, parts of speech, and the like.

  The unknown word extraction unit 152 extracts, as an unknown word, a morpheme sequence that is determined not to be registered in the basic dictionary 110 or the user dictionary 120 in the input text from the result of decomposing the text into morphemes by morphological analysis. .

  The surrounding text extraction unit 153 extracts words having relevance to the unknown word from the input text as the surrounding text from the result of the morphological analysis. Peripheral text extraction unit 153 performs a morphological analysis on a sentence containing an unknown word or a sentence in the vicinity of a sentence containing an unknown word, and has a relevance based on the attribute of each word obtained as a result of the morphological analysis Text can be extracted. The attribute of each word includes the grammatical nature of the word, such as the part of speech of the word. As a method for extracting the surrounding text, only a specific word, for example, only a specific part of speech may be extracted, or a phrase including the specific part of speech may be extracted. Note that the peripheral text to be extracted is not limited to a word, and may be, for example, a sentence including a whole sentence or a plurality of words.

  Note that. The extraction processing in the unknown word extraction unit 152 and the surrounding text extraction unit 153 can be performed without performing morphological analysis. For example, a portion in which kanji, katakana, and alphabets are simply extracted is regarded as surrounding text, and further, text that is not registered in the basic dictionary 110 or the user dictionary 120 in the surrounding text. May be extracted as an unknown word.

  The headword extraction unit 154 refers to the other person's user dictionary via the user dictionary interface unit 130 and the other person's user dictionary interface units 131 to 133, and the unknown word extracted by the unknown word extraction unit 152 from the other person's user dictionary Extract headwords that match.

  The feature level setting unit 155 sets a feature level corresponding to the attribute in the surrounding text associated with the headword extracted by the headword extraction unit 154. That is, the feature level setting unit 155 assigns a feature level corresponding to the attribute to the surrounding text stored in association with the same headword as the unknown word. The feature level can be, for example, a value indicating the degree of relevance of the input text with the context. For example, the feature level setting unit 155 can set the feature level for the surrounding text with reference to an accessible table that stores the feature level associated with the attribute of the surrounding text in advance. It can be considered that the higher the feature level, the more likely it is to suggest the context of the sentence. For example, in the table, if the text is registered in the user dictionary, the feature level is 5; if it is a proper noun, the feature level is 4; if it is a compound noun, the feature level is 3; The feature level may be set to 2 etc. so that the feature level of the surrounding text extracted with reference to the table can be set.

  The coincidence calculation unit 156 calculates the coincidence between the surrounding text stored in association with the same headword as the unknown word and the surrounding text extracted from the input text. For example, although the surrounding text matches, the degree of matching may be obtained by taking the sum of the feature levels. In this example, the addition value that is the sum of the feature levels is used as the “matching degree”, but the present invention is not limited to this. For example, a representative value representing the degree of coincidence such as the integrated value, average value, maximum value, minimum value, etc. of the feature level can be used.

  The feature level setting unit 155 is not essential and may be arbitrary. At this time, for example, the degree of coincidence can be calculated using the number of matches between the surrounding text extracted from the input text and the surrounding text extracted from another person's user dictionary.

  Based on the degree of coincidence calculated by the degree of coincidence calculation unit 156, the reading determination unit 157 determines whether or not to read the headword extracted from the other person's user dictionary as the unknown word extracted from the input text. To do.

  For example, the reading determination unit 157 has the highest matching degree between the surrounding text stored in association with the same headword as the unknown word calculated by the matching degree calculation unit 156 and the surrounding text extracted from the input text. The reading of the headword extracted from the other person's user dictionary is determined as the reading of the unknown word. Note that the reading determination unit 157 may further determine accents and other necessary information in addition to reading.

  The waveform processing unit 160 generates synthesized speech data according to the reading and accent information output from the language processing unit 150. Although not shown, the waveform processing unit 160 may have a waveform dictionary for synthesizing speech. For example, the waveform processing unit 160 adjusts the pitch of the voice in the waveform dictionary so as to be a target accent in digital signal processing using, for example, the PSOLA (Pitch Synchronous Over Add) method. Can be connected to generate synthesized speech.

  The audio output unit 170 converts the audio data generated by the waveform processing unit 160 into a format corresponding to various audio formats and outputs the converted data.

  In the above description, the speech synthesizer 100 includes the basic dictionary 110, the user dictionary 120, the user dictionary interface unit 130, the text input unit 140, the language processing unit 150, the waveform processing unit 160, and the speech output unit 170. However, the configuration of the speech synthesizer 100 is not limited to this. For example, the speech synthesizer 100 may be on a server connected to a network. In this case, for example, the text input unit 140 is configured to receive text input from a user terminal connected to the network, and the voice output unit 170 transmits the synthesized voice data to the user terminal via the network. It can be set as the structure to do. In addition, when the server stores the other person's user dictionary and accepts text input from a plurality of users, the user dictionary can be used for the plurality of users to synthesize speech for the input text. It is good also as a structure. Further, the functional units included in the speech synthesizer 100 may be distributed among a plurality of computers.

  Hereinafter, the operation of the speech synthesizer according to the first embodiment of the present invention will be described with reference to the flowchart in FIG. 2 and various data examples in FIG. 3. As shown in FIGS. 3C, 3A, and 3B, words related to the text that is usually read aloud are registered in the user dictionaries of myself, Mr. A, and Mr. B, respectively. Shall.

  First, it is assumed that the composition target text describing the interchange of the expressway shown in FIG. 3D is input to the text input unit 140 (step S201).

  Next, the morpheme analysis unit 151 performs morpheme analysis on the input composition target text (step S202).

  As a result of the morphological analysis, the unknown word detection unit 152 extracts words that are not registered in the basic dictionary or the user dictionary as unknown words (step S203). Here, it is assumed that the IC is extracted as an unknown word.

  Next, the surrounding text extraction unit 153 extracts the surrounding text from the sentence including the unknown word IC or the sentences before and after the unknown word IC. Here, for example, a configuration for extracting nouns as surrounding text will be described. It is assumed that “general road”, “highway”, and “entrance / exit”, which are nouns in the sentence including the IC, are extracted as the surrounding text (step S204).

  Next, the headword extraction unit 154 refers to the other person's user dictionary via the user dictionary interface unit 130 and the other person's user dictionary interface units 131 to 133, and the unknown word extraction unit 152 extracts from the other person's user dictionary. The headword that matches the unknown word and the surrounding text associated with the headword are extracted (step S205). Here, first, when searching for a headword corresponding to IC from Mr. A's user dictionary, IC: Interchange, and its surrounding text are “ETC only”, “highway”, “general road”, “charge” Is extracted. Next, when a headword corresponding to the IC is searched from the user dictionary of Mr. B, “IC” is extracted and “integrated circuit”, “element”, “semiconductor”, and “electronic circuit” are extracted as surrounding text. .

  Next, the feature level setting unit 155 applies the feature level corresponding to the attribute to the peripheral text associated with the same headword as the unknown word extracted by the headword extraction unit 154 and to the peripheral text extracted from the peripheral text 153. Is set (step S206). Here, the word registered in the user dictionary has the highest feature level 4, the proper noun registered in the basic dictionary has the feature level 3, the compound noun has the feature level 2, and the other nouns have the feature level. Is set as 1. According to the above, “highway” already registered in his / her user dictionary is feature level 4, compound word “general road” is feature level 2, and common noun “entrance / exit” is feature level 1. .

  Further, frequency information may be used as a feature level setting method. For example, it is assumed that an adjacent sentence including the unknown word “IC” is as follows. "IC" is the doorway that connects ordinary roads and expressways. In general, ICs are installed on highways every 10 km. The point where the expressway connects to the expressway is called JC. In this case, four “highways” are extracted as the IC surrounding text, and the appearance frequency may be used as the feature level as it is.

  Next, the matching score calculation unit 156 calculates the matching score between the surrounding text stored in association with the same headword as the unknown word and the surrounding text extracted from the input text (step S207). Here, the surrounding text “ETC only” stored in association with the headwords in the user dictionary of “A highway”, “general road”, and “entrance / exit”, which are already extracted, are the extracted surrounding text of the input text. The degree of coincidence with “highway”, “general road”, and “fee” is calculated. The degree of coincidence is calculated by, for example, calculating the sum of the feature levels of the coincidence, and is the sum of “4” for “highway” and “2” for “general road”. Becomes “6”. In the case of Mr. B, similar to Mr. A, the degree of coincidence between the input text already extracted and the surrounding text is calculated. However, since there is no matching surrounding text here, the degree of matching with Mr. B is “0”.

  Next, the reading determination unit 157 has the highest matching degree between the surrounding text stored in association with the same headword as the unknown word calculated by the matching degree calculation unit 156 and the surrounding text extracted from the input text. The reading of the headword extracted from the higher user dictionary of the other person is determined as the reading of the unknown word. Here, since the degree of coincidence with Mr. A is “6” and the degree of coincidence with Mr. B is “0”, the reading “IC: Interchange” of Mr. A is adopted and determined (step S208). . Finally, the reading determination unit 157 adds the reading and the accent by combining the determined reading and the part whose reading has already been found by the morphological analysis (step S209).

  By using the above method, you can search for unknown words used in similar contexts by searching other people's user dictionaries, including not only unknown words but also surrounding text that is likely to be used for topics related to the unknown words. Can be searched from someone else's dictionary. Therefore, it is possible to reduce the problem of differently read words. In addition, the speech synthesizer according to the present embodiment uses the unknown word and the surrounding text included in the text to be synthesized instead of the entire user dictionary, and compares the similarity with the user dictionary by comparing with the other user dictionary. Even if it is low, it is possible to find and use a dictionary in which appropriate readings are registered.

  Furthermore, the present invention improves the reliability of the degree of matching by assigning a feature level to the surrounding text. For example, by assigning a high feature level to a user dictionary word group or proper noun, the surrounding text representing the feature of the context including the unknown word is more emphasized, and the probability that an appropriate user dictionary of another person can be selected is increased. The reading accuracy can be increased.

  That is, it is possible to correctly obtain a reading of a word that is not registered in the user dictionary. Words registered by users and words that are included in unknown words are words that could not be covered by the basic word dictionary in which words that appear frequently in Japanese are registered, so they contain many proper nouns and technical terms. This is a word that represents the characteristics of the content of the text to be synthesized. Therefore, by comparing these words with words registered in the other person's user dictionary, it is possible to estimate other users who have already read out the text having the same content. In addition, since the user dictionary is prepared for reading out text with similar contents, there is a high probability that the correct reading is inevitably registered, and it is possible to avoid the problem of misread words. Further, unlike the case where a conventional field-specific dictionary is used, it is not necessary to estimate a specific field from the text to be synthesized. Furthermore, since the latest user dictionary maintained by the user can be used constantly, new words can be handled.

[Embodiment 2]
The speech synthesizer according to the present embodiment includes an unknown word conversion unit that performs predetermined conversion on the notation of the unknown word extracted by the unknown word extraction unit 152 from the synthesis target text and generates an unknown word that can be regarded as the same as the unknown word. Furthermore, you may provide. The unknown word conversion unit converts the notation of the unknown word extracted by the unknown word extraction unit 152 from the synthesis target text into a notation that can be regarded as the same. For example, a conversion rule is recorded in advance, and the unknown word conversion unit can perform conversion using this conversion rule. The conversion rule may be, for example, referable data that defines the correspondence between the converted data and the converted data, or may be specified by a predetermined conversion program. Further, it may further include a peripheral text conversion unit that performs predetermined conversion on the notation of the peripheral text extracted by the peripheral text extraction unit 153 from the synthesis target text and generates a peripheral text that can be regarded as the same as the peripheral text. The surrounding text conversion unit converts the notation of the surrounding text extracted by the surrounding text extraction unit 153 from the synthesis target text into a notation that can be regarded as the same.

  As a result, even when the unknown words and / or surrounding texts are represented differently, it is possible to cope with cases where the readings can be regarded as the same. For example, this is a configuration example effective when the unknown word notation is “entrance / exit” and the headword registered in the user dictionary of another person is “entrance / exit”.

  The configuration of the coincidence degree calculation unit of the speech synthesizer according to Embodiment 2 of the present invention is the block diagram shown in FIG. In the speech synthesizer according to the second embodiment of the present invention, the matching degree calculation unit 156 includes a headword matching degree determination unit 180 and a surrounding text matching degree determination unit 190 as compared with the configuration of the first embodiment. The coincidence degree determination unit 180 includes a reading expansion unit 181, a specific character deletion unit 182, and a character conversion unit 183, and the surrounding text coincidence degree determination unit 190 includes a reading expansion unit 191, a specific character deletion unit 192, and a character conversion unit. 193.

  The headword coincidence degree determination unit 180 determines the degree of coincidence between an unknown word and a headword registered in another person's user dictionary. The surrounding text matching degree determination unit 190 determines the matching degree between the surrounding text of the synthesis target text and the surrounding text registered in the other person's user dictionary.

  The reading expansion units 181 and 191 generate text obtained by converting unknown words or entry words in a user dictionary and / or kanji in surrounding text into hiragana. The specific character deletion units 182 and 192 generate text in which symbols such as a sending kana, a hyphen, and a period are deleted. The character conversion units 183 and 193 perform conversion between uppercase and lowercase letters, conversion between hiragana and katakana, conversion between Roman numerals and Arabic numerals, and generate converted text.

  If the converted text and the entry word of the other person's user dictionary completely match, even if the notation is different, the unknown word and the entry word of the other person's user dictionary are considered to be the same.

  Next, when the headwords are considered to match, the surrounding text is compared in the same way as the headword. Even in the comparison of surrounding text, even if the notation does not match, when matching by reading expansion, specific character deletion, character conversion, it is regarded as the same surrounding text, and the feature level adding unit 194 adds the feature level And the degree of coincidence is calculated.

  In the above description, the example in which the unknown word of the synthesis target text is converted and the headword of the other person's user dictionary is not converted has been described, but the present invention is not limited to this. You may compare what converted the unknown word of the synthetic | combination object text, and the thing which converted the headword of the other person's user dictionary. Moreover, although the example which compares what converted the surrounding text of a synthetic | combination object text with the thing which has not converted the surrounding text of another user's dictionary was demonstrated, it is not restricted to this. You may compare what converted the surrounding text of the synthetic | combination object text, and what converted the surrounding text of another person's user dictionary.

[Embodiment 3]
The entire configuration of the speech synthesizer according to Embodiment 3 of the present invention is a block diagram shown in FIG.

  The speech synthesizer according to the present embodiment additionally registers a headword of another person's user dictionary and a reading of the headword when the matching degree calculated by the matching degree calculation unit 156 is equal to or greater than a predetermined value in the user dictionary of the user. The user dictionary update unit 200 may further be provided. The user dictionary updating unit 200 further additionally registers peripheral text having a feature level of the peripheral text that is greater than or equal to a predetermined value among the peripheral text having relevance to the additionally registered headword in the user dictionary. Also good.

  As a result, the contents of the other person's user dictionary can be taken into the user's user dictionary. This increases the amount of data in the user dictionary of the user, so that it is not necessary to refer to the user dictionary of another person, and high-speed processing can be performed.

  The speech synthesizer according to the third embodiment of the present invention further includes a user dictionary update unit 200 with respect to the configuration of the first embodiment. The user dictionary update unit 200 includes a validity determination unit 210, a registered peripheral text determination unit 220, And a registration unit 230. The user dictionary updating unit 200 has a function of additionally registering readings of other people's user dictionaries determined by the reading determining unit 157 in its own user dictionary.

  For example, assume that the text shown in FIG. From this, it is assumed that the unknown word is extracted as “IC” and the surrounding text is extracted as “semiconductor”, “circuit”, “integrated circuit”, and “electronic circuit”. The degree-of-match calculation unit 156 calculates the degree of match between the surrounding text of the input text that has already been extracted and the surrounding text of another person's user dictionary. The coincidence calculation method is the same as that in the first embodiment, and the coincidence calculation unit 156 calculates, for example, the sum of the feature levels of the matches, and there is no matching surrounding text. 0 ". In the case of Mr. B, it is the sum of “1” of “semiconductor”, “4” of “integrated circuit”, and “2” of “electronic circuit”, and the degree of coincidence with Mr. B is “7”. . It is assumed that the reading determining unit 157 adopts “Icy” as a reading for the headword of Mr. B.

  Here, the user dictionary update unit 200 determines in the validity determination unit 210 whether or not the degree of coincidence is greater than or equal to a threshold value. When the matching degree is equal to or higher than the threshold, the user dictionary updating unit 200 adds a reading “Icy” of Mr. B's headword to his / her dictionary because there is a high possibility of correct reading appearing in similar text. Decide to register.

  Next, the registered surrounding text determining unit 220 selects and employs a user text of another person and the surrounding text in the user's user dictionary having a feature level of a certain value or more. By adopting the surrounding text that is not included in the text you have entered, you can increase your ability to handle various types of input. The registered peripheral text determination unit 220 may be omitted from the configuration of the user dictionary update unit 200.

  Finally, the registration unit 230 uses an unknown word as a headword, reads another user's dictionary as a reading, and additionally registers the peripheral text selected by the registered peripheral text determination unit 220 as a peripheral text in its own user dictionary. When registering, the feature level may be registered at the same time.

  The configuration described in the above embodiment merely shows a specific example, and does not limit the technical scope of the present invention. Any configuration can be employed within the scope of the effects of the present invention.

  In the embodiment of the present invention, a software program for realizing the above-described embodiment (in the embodiment, a program corresponding to the flowchart shown in FIG. 2) is supplied to the apparatus, and a computer of the apparatus is supplied. This includes the case where it is also achieved by reading and executing the program. Therefore, in order to realize the functional processing described in this embodiment by a computer, the program itself installed in the computer is also an embodiment of the present invention. That is, a program for realizing the functional processing of the present invention is also included in one aspect of the embodiment. A medium recording a program for realizing the functional processing of the present invention is also included in one aspect of the embodiment.

100 speech synthesizer 110 basic dictionary 120 user dictionary 121 other person's user dictionary 122 other person's user dictionary 123 other person's user dictionary 130 user dictionary interface unit 131 user dictionary interface unit 132 user dictionary interface unit 133 user dictionary interface unit 140 text input unit 150 Language processing unit 151 Morphological analysis unit 152 Unknown word extraction unit 153 Peripheral text extraction unit 154 Headword extraction unit 155 Feature level setting unit 156 Matching degree calculation unit 157 Reading determination unit 160 Waveform processing unit 170 Speech output unit 180 Headword matching degree determination Section 181 Reading development section 182 Specific character deletion section 183 Character conversion section 190 Peripheral text matching degree determination section 191 Reading expansion section 192 Specific character deletion section 193 Character conversion section 194 Feature level addition section 200 U The dictionary update unit 210 Validity determination unit 220 Registration peripheral text determination unit 230 Registration unit

Claims (9)

Headword, the headword readings, and the user of the user dictionary to store in association with surrounding text having relevance to the entry word, is accessible to the basic dictionary,
A text input part that accepts input of text to be synthesized;
An interface unit that makes it possible to refer to a plurality of other user's dictionaries that store the headword, the reading of the headword, and the surrounding text that is related to the headword ;
And unknown word extracting unit for extracting an unknown word from the compositing target text accepts an input in the text input section not included in the basic dictionary in the user dictionary of the user,
A peripheral text extraction unit for extracting the surrounding text with relevance to the unknown word from the synthesis target text that contains the unknown word which the unknown word extraction portion is extracted,
Index word extracting unit for extracting the headword via said interface unit refers to the user dictionary of the others, to match the unknown word which the unknown word extraction unit has extracted from the user dictionary of the others,
Matching degree calculating the degree of coincidence between said peripheral text the peripheral text extracting unit from the peripheral text and the compositing target text from the user dictionary associated with the entry word which the index word extracting unit has extracted the others and extracted A calculation unit;
On the basis of the degrees of coincidence the coincidence degree calculation unit has calculated, the reading from the user dictionary of the others of the entry word which the index word extracting unit has extracted, the unknown word extraction unit from said compositing target text extracted the A speech synthesizer comprising: a reading determination unit that determines whether or not to read an unknown word. To the peripheral text the peripheral text extracting unit has extracted, further comprising a feature level setting unit for setting a characteristic level in accordance with the attribute of the peripheral text,
The coincidence degree calculation unit calculates the degree of coincidence with the feature level of the peripheral text the feature level setting unit has set, the speech synthesis apparatus according to claim 1. The coincidence degree calculation unit calculates a sum of the characteristic level of said peripheral text for the entry word which the index word extracting unit from the user dictionary and the extraction of the others as the degree of matching,
The reading determination unit, the value of the sum of the feature level to read the highest the entry word is determined as a reading of the unknown word, the speech synthesis apparatus according to claim 2. The synthesis target from text representation of the unknown word to the unknown word extraction unit has extracted performs predetermined conversion, further comprising an unknown word conversion unit for generating the unknown words which can be regarded as the same as the unknown word,
The entry word extraction unit extracts the headword via said interface unit refers to the user dictionary of the others, consistent with the unknown word generated by the conversion of the unknown word conversion unit from the user dictionary of the person And
Said match degree calculating section, the compositing target text with the surrounding text associated with the entry word which the index word extracting unit has extracted from the user dictionary of the others and the peripheral text the peripheral text extracting unit has extracted The speech synthesizer according to claim 1, wherein the degree of coincidence is calculated. The synthesis target text notation of the peripheral text the peripheral text extracting unit has extracted performs predetermined conversion, further comprising a peripheral text conversion unit for generating the peripheral text which can be regarded as the same as the peripheral text,
It said match degree calculating section, the match with the surrounding text associated with the entry words extracted is the index word extraction unit and the peripheral text generated by the conversion of the peripheral text conversion unit from a user dictionary of the person The speech synthesizer according to claim 1, which also calculates the degree. The degree of coincidence the coincidence degree calculation unit has calculated is not less than a predetermined value, the reading of the entry word and the entry word of the user dictionary of the others, the user dictionary updating unit that additionally registers the user dictionary of the user The speech synthesizer according to claim 1, further comprising: The user dictionary updating unit is further among the peripheral text have relevance to the additional registered headword, additionally registers the surrounding text feature level of the peripheral text is not less than a predetermined value in the user dictionary of the user The speech synthesizer according to claim 6. Headword, the headword readings, and the user of the user dictionary to store in association with surrounding text having relevance to the entry word, is accessible computers and basic dictionary,
A text input step that accepts input of text to be synthesized;
And unknown word extracting step of extracting an unknown word from the compositing target text accepted input in the text input step not included in the basic dictionary in the user dictionary of the user,
And peripheral text extraction step of extracting the surrounding text with relevance to the unknown word from the synthesis target text that contains the unknown word extracted by the unknown word extraction step,
Via the interface unit refers to the user dictionary of others, and headword extracting a headword matching the unknown word extracted by the unknown word extraction step from a user dictionary of the others,
Matching degree calculating the degree of coincidence between said peripheral text from the user dictionary and the peripheral text associated with the entry words extracted by the index word extracting step from the synthesis target text extracted by the neighborhood text extraction step of the others A calculation step;
On the basis of the degree of matching calculated by the matching degree calculating step, the unknown to the reading from the user dictionary of the others headword the index word extracting unit has extracted, extracted with the unknown word extraction step from the synthesis target text A speech synthesis method including a reading determination step for determining whether or not to read a word. Headword, the headword readings, and the user of the user dictionary to store in association with surrounding text having relevance to the entry word, accessible computers and basic dictionary,
A text input step that accepts input of text to be synthesized;
And unknown word extracting step of extracting an unknown word from the compositing target text accepted input in the text input step not included in the basic dictionary in the user dictionary of the user,
And peripheral text extracting a surrounding text with relevance to the unknown word from the synthesis target text that contains the unknown word extracted by the unknown word extraction step,
Via the interface unit refers to the user dictionary of others, and headword extracting the index word that matches the unknown word extracted by the unknown word extraction step from a user dictionary of the others,
Matching degree calculating the degree of coincidence between said peripheral text from the user dictionary and the peripheral text associated with the entry words extracted by the index word extracting step from the synthesis target text extracted by the neighborhood text extraction step of the others A calculation step;
On the basis of the degree of matching calculated by the matching degree calculation step, the reading from the user dictionary of the others of the entry word in which the index word extracting unit has extracted, extracted with the unknown word extraction step from the synthesis target text the A speech synthesis program for executing a reading determination step for determining whether or not to read an unknown word.

Images