JP3589972B2 - Speech synthesizer - Google Patents

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to word notation processing in text analysis of speech synthesis.
[0002]
[Prior art]
When a sentence is classified by a transmission medium, it is divided into a character language expressed by characters and a speech language by voice. When sentences are classified by style, there are sentences (sentence style, written words) and spoken languages (spoken style, spoken words). Sentences are mainly used in written languages, and colloquials are mainly used in spoken languages.
[0003]
In a character language, a sentence is usually used, but in an e-mail or the like, a spoken language is often used. In such a case, the prosody is often represented by characters. Document [1] classifies the method of emphasizing words. When the inventor applied the above method to the collected example sentences, it was found that many notations could be explained as emphasis.
[0004]
FIG. 28 shows the notation used in each case added to the emphasis classification of the document [1] by the inventor.
[0005]
FIG. 26 is a configuration diagram of a conventional speech synthesizer. When a Japanese sentence is input from the sentence input unit 001, a linguistic analysis is performed by the linguistic analysis unit 002. First, the vocabulary analysis unit 003 deletes and replaces characters and the like unnecessary for language analysis, and divides the input into sentences. Next, the morphological analysis unit 004 performs a morphological analysis to divide the input sentence into words, and selects a word string determined to be most appropriate as a division of the input sentence from the extracted words. The retrieval of words from the word dictionary 006 is performed by the dictionary search processing unit 005.
[0006]
FIG. 27 shows a part of a word dictionary search result of an input sentence. As shown in the figure, at each character position of the input sentence, a word having the character at the head and having a partial character string of the input sentence as a word heading is extracted from the word dictionary. For words that are not in the dictionary, the unknown word processing unit 007 estimates unknown words and creates words.
[0007]
After the morphological analysis is completed, the syntax analysis unit 008 performs syntax analysis. The accent combination and rendaku processing unit 009 performs the accent combination processing of the accent obtained from the word dictionary with the preceding and following words using the result of the morphological analysis and the syntax analysis, and becomes a basic unit when reading out. Create accent phrases. The prosody processing unit 010 sets a phrase command and a pause command for a row of accent phrases. The synthesis parameter unit 011 generates a time series of parameters necessary for controlling speech synthesis based on the analysis result of the language analysis unit. The audio waveform generation unit 012 performs D / A conversion on the control parameter to generate an audio waveform. The audio output unit 013 outputs the audio waveform as audio.
[0008]
As an algorithm of processing in the morphological analysis unit 004 and the dictionary search processing unit 005, there is an algorithm shown in Document [2].
[0009]
Literature [1]: Lecture: Japanese and Japanese language education 2, Japanese speech and phonology (above), Meiji Shoin, pp. 316-342
Reference [2]: Iwanami Lecture @ Science of Language 3, Words and Dictionaries, Iwanami Shoten, pp. 55-73
[0010]
[Problems to be solved by the invention]
As shown in 1 of FIG. 28, when the “pronunciation method by pronunciation” is performed and the notation of the emphasized word is created, a plurality of words having different pronunciations can be created. Furthermore, since the notation of these words is not unique, the words created by the above-mentioned emphasis method are very large. For example, for "Wow," words such as "Wow,""Wow,""Wow,""Wow,""Wow,""Wow," can be made. .
[0011]
Hereinafter, in the present invention, as a term for explanation, such as "wow", emphasizing a certain word, the word resulting from the notation change is "emphasized", "wow", especially Words used without emphasis will be referred to as "neutral".
[0012]
As described above, the number of emphasized forms is extremely large for one neutral form, and it is virtually impossible to register all emphasized forms in the word dictionary. Therefore, usually, only the neutral form and the frequently emphasized form are registered in the word dictionary. As a result, with respect to sentences in which the emphasis forms are frequently used, particularly, sentences such as colloquial texts, there is a disadvantage that the analysis accuracy of the input sentence is reduced and speech synthesis becomes unnatural.
[0013]
Also, when words with emphasized forms, such as "soooooooo," are registered in the word dictionary, speech is synthesized using neutral prosody because the emphasized forms are not considered, and the synthesized sound is unnatural. There was a disadvantage that it was.
[0014]
Further, 2. of FIG. As described in [1] and in the literature [1], there is also an “emphasized word” in which the neutral form already has the meaning of emphasis, and even for such a word, a natural synthesized sound could not be obtained.
[0015]
In the present invention, it is possible to provide a speech synthesis device that suppresses an increase in the storage capacity of a memory that stores a word dictionary, controls the prosody flexibly by using emphasized forms and emphasized words, and is more user-friendly. Aim.
[0016]
[Means for Solving the Problems]
A speech synthesis device according to the present invention includes a sentence input unit for inputting a sentence including an emphasized word with an emphasis, a language analysis unit for performing a language analysis of the input sentence, and a result of a language analysis performed by the language analysis unit. Prosody processing means for setting prosody information, synthesis parameter generation means for generating speech synthesis parameters based on the result of language analysis in which the prosody information is set by the prosody processing means, and generating a speech waveform based on the speech synthesis parameters. A speech synthesizer comprising: a speech waveform generating means for generating; and a speech output means for outputting a speech waveform as speech, wherein the language analyzing means deletes and replaces characters unnecessary for language analysis from the input text. A vocabulary analysis unit that divides the document in which unnecessary characters are deleted and replaced into sentences, and a connection between words extracted from the word dictionary by applying various notation neutralization rules By verifying the possibility, the morphological analysis means for converting the notation of the emphasized form in the sentence divided by the lexical analysis means into the notation of the neutral form and dividing the sentence into words, and according to the application status of the notation neutralization rule And applying a reading correction rule to the reading of the word divided by the morphological analysis means to correct the reading to an emphasized form, and applying a prosody modification rule to the prosody of the word divided by the morphological analysis means. And an emphasis notation processing means for modifying the prosody emphasized to output the result to the prosody processing main means.
[0020]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an embodiment (embodiment) of a speech synthesis device of the present invention will be described in detail with reference to the drawings.
[0021]
<First embodiment>
<Structure>
FIG. 1 is a configuration diagram showing a first embodiment of the present invention. 101 is a sentence input unit for inputting a Japanese sentence, 102 is a language analysis unit for performing linguistic analysis of the input sentence, and 103 is a unit that deletes and replaces characters and the like unnecessary for language analysis and converts the input into a sentence. A vocabulary analysis unit 104 for dividing the input sentence into words, a morphological analysis unit for dividing the input sentence into words, a dictionary search processing unit 105 for searching a word dictionary, and a word dictionary 106 for storing linguistic information used in the language analysis unit is there.
[0022]
Reference numeral 107 denotes a notation neutralization rule for changing a search character string in a dictionary search. The notation neutralization rule is a rule for changing the notation of the emphasized form to the notation of the neutral form and matching the notation of a word registered in the dictionary. FIG. 2 is a classification of characters related to the emphasis notation used in the notation neutralization rule. FIG. 3 shows a notation neutralization rule used in the dictionary search processing unit 105.
[0023]
An unknown word processing unit 108 estimates an unknown word and creates a word candidate when the word in the input document is not in the word dictionary.
[0024]
109, referring to the application result of the notation neutralization rule 107 for the word of the morphological analysis result, corrects the reading of the extracted word so as to be closer to the reading of the emphasized type in the input sentence, contrary to the search. , And an emphasis notation processing unit that corrects the prosody. Reference numeral 110 denotes a reading correction rule storage unit that stores reading correction rules used in the emphasis notation processing unit. FIG. 6 shows the reading correction rule.
[0025]
A prosody modification rule storage unit 111 stores prosody modification rules used in the emphasis notation processing unit. FIG. 8 shows a prosody modification rule. In the prosody modification rule, the "accent emphasis rule" is a rule for strengthening the accent when an emphasized form is used in the input sentence. The "speed delay rule" is a rule in which the emphasized form is used. This is a rule that causes the utterance speed to be slowed down and the effect of emphasis to be produced when the utterance is present.
[0026]
Reference numeral 112 denotes a syntactic analysis unit that performs syntactic analysis using the result of the morphological analysis, 113 denotes an accent combination, rendaku processing, etc., an accent / rendation processing unit, and 114 denotes a prosody such as a phrase or a pause output from the language analysis unit. A prosody processing unit 115 for setting information, a synthesis parameter generation unit 115 for generating a voice synthesis control parameter, a voice waveform generation unit 116 for D / A converting the control parameter to generate a voice waveform, and 117 An audio output unit that outputs a waveform as audio.
[0027]
<Operation>
When a Japanese sentence is input from the sentence input unit 101, a linguistic analysis is performed by the linguistic analysis unit 102. First, the vocabulary analysis unit 103 deletes and replaces characters unnecessary for language analysis, and divides the input into sentences. Next, the morphological analysis unit 104 performs a morphological analysis to divide the input sentence into words. Regardless of the morphological analysis algorithm, at each character position of the input sentence, a process of starting with the character position and heading for a partial character string of the input sentence as a heading is performed to collectively extract the words from the word dictionary 106. The morphological analysis unit 104 verifies the possibility of connection between the words extracted from the word dictionary 106, determines the most appropriate word string as word division of the input sentence, and outputs it. The words extracted from the word dictionary 106 are provided with various information required for speech synthesis in addition to information necessary for morphological analysis, such as headings, parts of speech, and connection information.
[0028]
The retrieval of words from the word dictionary 106 is performed by the dictionary search processing unit 105. In the present invention, in addition to performing a dictionary search as shown in FIG. 27 using a substring of an input character string as a search character string to be used in a word dictionary search, a character classification shown in FIG. A search is performed using a character string obtained by replacing characters according to the conversion rule 107. FIG. 3 shows a notation neutralization rule used in the dictionary search unit.
[0029]
FIG. 4 is a flowchart of a dictionary search. Search () is a procedure for obtaining a word that is a partial character string of the input sentence as shown in FIG. 27. For example, it is assumed that S = “mail”. N = 9 and i = 4. In step S1, D (S (4)) = φ. In step S2, when Search (S (4)) is executed with S (4) = “ma- ぁ”, D (S (4)) = ｛ma (inflection, ...), ma (adverb) , ...) get｝. In step S3, the chain of long letter equivalents of S (4), the chain of consonant sounds, and the chain of sound repelling are combined into one, and the letters corresponding to long sounds are replaced with the standard notation, so that S (4) = “Ma ぁ”. . In step S4, S2 = S (4) = “Ma-A”. In step S5, a combination without the long-sound deletion rule, no prompting rule, and no sound-repelling rule is selected. In step S6, it is determined whether or not there is a combination that has not been applied. If there is a combination, then in step S7, when the rule is applied to S2, S2 = “much better”. When Search (S2) is executed in step S8, D (S2) = {ma (adverb, ...), ma (adverb, ...)}. In step S9, D (S (4)) = D (S (4)) {D (S2) = {ma (intransitive verb, ...), ma (adverb, ...)}, and the process returns to step S4. If there is no unapplied combination in step S6, the process ends.
[0030]
In step S4, S2 = S (4) = “Ma-A”. In step S5, a combination without the long sound deletion rule, the prompt sound deletion rule, and the sound repelling rule is selected. In this case, if the rule selected in step S7 is applied to S2, then S2 = "". In step S8, when Search (S2) is executed, D (S2) = {Masu (ordinary noun, ...), Masu (auxiliary verb, ...), Ma (inspirational verb, ...), Ma (adverb, ...)} Become. At step S9, D (S (4)) = D (S (4)) ∪D (S2) = ｛(ordinary noun, ...), mas (auxiliary verb, ...), ma (inflection verb, ...), s Ma (adverb, ...)｝.
[0031]
Hereinafter, steps S4 to S9 are executed by changing the combination of rules, but D (S (4)) does not change, the combination of rules disappears, the determination result in step S6 becomes NO, and the process ends. .
[0032]
As a result, D (S (4)) = {mass (ordinary noun,...), Mas (auxiliary verb,.
[0033]
The unknown word correction unit 108 creates some unknown word candidates and adds them to the dictionary search result because the beginning of the partial character string of the input sentence used in the dictionary search may be an unknown word.
[0034]
Whether or not "mas (auxiliary verb)" extracted from the dictionary is obtained as a part of the final morphological analysis result depends on the morphological analysis algorithm, but in this example, a set of words included in the dictionary lookup result Is the same as the dictionary lookup result of the sentence "mailed." Without the emphasized form. As long as "mailed" can be correctly analyzed, the correct result is also obtained for "mailed." can get.
[0035]
FIG. 5 shows an example of a morphological analysis result according to the present invention. The word extracted from the dictionary stores information on the applied notation neutralization rule and the character position to which the rule is applied, since the word is taken out of the dictionary because it is required by the emphasis notation processing unit 109. Also, the word heading before the writing neutralization rule is applied is stored. If multiple rules are applied, the multiple rules are stored.
[0036]
After the morphological analysis is completed, the emphasis notation processing unit 109 first corrects the reading of the morphological analysis result using the reading correction rules stored in the reading correction rule storage unit 110. FIG. 6 shows the reading correction rule. FIG. 7 is an application example of the reading correction rule.
[0037]
Next, the emphasis notation processing unit 109 modifies the prosody according to the prosody modification rule stored in the prosody modification rule storage unit 111. FIG. 9 is an application example of the prosody modification rule. In FIG. 9, first, a reading correction rule is applied to a morphological analysis result obtained by analyzing a sentence “Posted mail. The reading of "mas" is changed depending on the application status of the notation neutralization rule included in the morphological analysis result. Next, the prosody modification rules are applied. Based on the application status of the notation neutralization rule included in the morphological analysis result, the accent enhancement rule is applied to “mas”. Furthermore, in the input sentence, "~~" is used as the notation of the prolonged part of "masu", so the speed delay rule is applied. As a result, as shown in FIG. 9, the accent of "mas" and the delay speed are changed.
[0038]
Using the output of the morphological analysis, the syntax analysis unit 112 performs syntax analysis. The accent combining and rendaku processing unit 113 uses the results of the morphological analysis and the syntax analysis to perform accent combining processing on the accent obtained from the word dictionary 106 with the preceding and succeeding words, and Create an accent phrase The prosody processing unit 114 sets a phrase command and a pause command for the string of accent phrases.
[0039]
The synthesis parameter generation unit 115 generates a time series of parameters necessary for controlling speech synthesis based on the analysis result of the language analysis unit 102. The audio waveform generation unit 116 performs D / A conversion of the control parameter to generate an audio waveform. The audio output unit 117 outputs an audio waveform as audio.
[0040]
<Second embodiment>
<Structure>
FIG. 10 is a configuration diagram of the second embodiment. In the first embodiment, instead of the word dictionary 106, a word dictionary 201 to which emphasis information is added is used, and an emphasis word processing unit 202 and an emphasis word processing rule storage unit 203 are provided. The points are different.
[0041]
FIG. 11 is an example of emphasis information added to the word dictionary. The emphasized word information is information indicating that a word is a vocabulary that is always emphasized. If E1, it is an emphasized word. The reading change flag is a flag indicating whether or not the word has been changed in reading due to emphasis. If it is ON, it has been changed, and if it is OFF, it has not been changed. The reading change information is information describing how to change the reading when the word is emphasized. The adjective "Wow" indicates that the reading can be changed to "Sgoi", "Sugoi", or "Sungoi". The readings generated in the reading change information can be used for the change of the word headings as they are. Although the word dictionary normally registers the stem portion, the emphasis word processing rule in FIG. 12 uses an adjective including the end of the word for explanation.
[0042]
<Operation>
The same operation as in the first embodiment is performed up to the emphasis notation processing unit 109. However, since the word dictionary 201 used in the morphological analysis contains the emphasis information, the result of the morphological analysis also includes the emphasis information extracted from the word dictionary. The emphasized word processing unit 202 processes the emphasized word using the emphasized word processing rule of the emphasized word processing rule storage unit 203.
[0043]
FIG. 13 shows an example of the emphasized word processing of a sentence including the emphasized word. It is assumed that “pretty” is registered in the word dictionary as an emphasized word. The processing result of the emphasis notation processing unit 109 has a normal accent. The emphasized word processing unit 202 refers to the emphasized word processing rules. In this case, “emphasis” is applied to “pretty”, and the “pretty” accent becomes stronger.
[0044]
FIG. 14 shows an example of the emphasized word processing when the user specifies emphasis. Assume that it is specified that a character string between “<emphasis>” and “</ emphasis>” is emphasized. Similarly to FIG. 13, it is assumed that “quite” is registered as an emphasized word in the word dictionary. The processing result of the emphasis notation processing unit 109 has a normal accent. The emphasized word processing unit 202 refers to the emphasized word processing rules. In this case, the “emphasized word accent enhancement rule” is applied to “pretty”, the “pretty” accent becomes strong, and the prosody becomes “ka & nari”. Further, the “emphasis word reading change rule” is applied. Since "pretty" has A4 as the reading change information, the first beat is prolonged, and the prosody is "Ka &nari".
[0045]
FIG. 15 shows the non-emphasized word processing when the user designates “emphasized”. Assume that it is specified that a character string between “<emphasis>” and “</ emphasis>” is emphasized. Unlike the examples in FIGS. 13 and 14, it is assumed that there is no emphasized word in the text. The processing result of the emphasis notation processing unit is given a normal accent. The emphasized word processing unit refers to the emphasized word processing rules. In this case, the processing for the three words “this”, “weird”, and “by chance” will be described. The emphasized word information extracted from the word dictionary is “E0” indicating that all three words are non-emphasized words. Therefore, the “non-emphasized word emphasis rule” is applied to the three words. As a result, at the same time that the accent is strengthened, the pronunciation is corrected, and as shown in the emphasized word processing result of FIG. 15, the pronunciations "Koruna", "He &&Nanna", and "Hyotshite" and the prosody (accent) are obtained, respectively. Can be
[0046]
<Third embodiment>
<Structure>
FIG. 16 is a configuration diagram showing a third embodiment of the present invention. The difference from the second embodiment is that modules 301 to 305 are provided. Reference numeral 301 denotes an emphasis information extraction unit that extracts words in which spelling neutralization is performed from the result of the morphological analysis, 302 denotes an emphasis information storage unit that stores the extracted emphasis information, and 303 denotes an emphasis information stored in the 302. The emphasis information management unit 304, which is a user interface for adding a word to the dictionary, retrieves the emphasis information from the emphasis information storage unit 302, and checks the emphasis information for the user to confirm whether the information can be added to the dictionary. The unit 305 is an emphasis information registration unit that registers the emphasis information confirmed to be registered in the dictionary in the word dictionary.
[0047]
<Operation>
The sentence input from the sentence input unit 101 is divided into sentences by the vocabulary analysis unit 103 and morphologically analyzed by the morphological analysis unit 104. The morphological analysis result is, for example, as shown in FIG. In the modules 109 and thereafter, the same operation as in the second embodiment is performed, and speech synthesis is performed.
[0048]
The emphasis information extraction unit 301 scans each word of the morphological analysis result, checks whether the notation neutralization rule is applied, and stores only the information of the word to which the rule is applied in the emphasis information storage unit 302. I do. FIG. 17 is an example of the extraction emphasis information. The processing up to this point is performed for all sentences input from the sentence input unit.
[0049]
After the above processing is completed for all the sentences in the sentence, the user operates the emphasis information management unit 303. The emphasis information management unit 303 operates the emphasis information confirmation unit 304. The emphasis information confirmation unit 304 extracts the morphological analysis result from the emphasis information storage unit 302 word by word. Hereinafter, a description will be given assuming that a result as shown in FIG. 17 is obtained in the emphasis information storage unit 302.
[0050]
First, "Masu (auxiliary verb, ma & s, rule for deleting long-equivalent characters, immediately after the first character,"") is extracted. Display word information on the user's screen. FIG. 18 is an example of a user display screen. The neutral form is a form in which the character string of the input sentence is changed according to the notation neutralization rule of the present apparatus so that it can be matched with the registered word in the word dictionary. The emphasized form is the form that actually appears in the character string of the input sentence. It is.
[0051]
If the notation neutralization rule is applied incorrectly, an erroneous neutral form may be created, and the user is prompted for confirmation. In addition to the neutral form and the emphasized form, the user is allowed to display the original text and the part of speech subjected to the morphological analysis to make the judgment accurate. In the above example, if the user determines that "Ma- ~ s" is an emphasized form of "masu", a designation to add such emphasis information to the auxiliary verb "mas" is made. Is done.
[0052]
When the user's confirmation is input from the emphasized information confirming unit 304, the emphasized information managing unit 303 calls the emphasized information registering unit 305 and passes the morphological analysis result. The emphasis information registration unit 305 refers to the application result of the notation neutralization rule of the passed morphological analysis result. In this case, since the rule for deleting long-equivalent characters immediately after the first character is applied to the change from "mama" to "masu", "mama" is changed to "mama" The change is “prolonging the initial beat”, and information A4 (“prolonging the initial beat”) shown in the reading change information of the emphasis information is added to “mas” in the word dictionary. . As a result, the emphasis information of “mas” is as shown in FIG.
[0053]
Next, from the morphological analysis result shown in FIG. 17, “Shikari (verb, shikari, sounding deletion rule, immediately after the first character,“ firm ”)” is extracted. On the user's screen, “neutral form: shikari”, “emphasized form: firm”, “part of speech: verb”, and an input sentence are displayed. In this case, "sturdy" is not an emphasized form of the verb "shikari". If the user determines that the extraction emphasis information is inappropriate, the extraction emphasis information confirmation unit does not instruct registration in the word dictionary, so the extraction emphasis information confirmation unit reads the next morphological analysis result without calling the extraction emphasis information registration unit. Is extracted from the extraction emphasis information storage unit, and the above processing is repeated.
[0054]
When all the morphological analysis results in the extraction emphasis information storage unit have been processed, the operation ends.
[0055]
<Fourth embodiment>
<Structure>
FIG. 20 is a configuration diagram showing a fourth embodiment of the present invention. The difference from the second embodiment lies in that an emphasis notation creation unit 401 and an emphasis notation output unit 402 are provided. FIG. 21 shows rules for creating emphasis notation.
[0056]
<Operation>
The same operations as in the second embodiment are performed from the sentence input unit 101 to the voice output unit 117. On the other hand, when the processing of the emphasized word processing unit 202 ends, the analysis result is sent to the emphasized notation creation unit.
[0057]
FIG. 22 is an example of information sent from the emphasized word processing unit 202 to the emphasized notation creation unit 401. For each word, part-of-speech, pronunciation (including accent position and accent strength), emphasized word information (E0: non-emphasized word, E1: emphasized word) indicating whether the word is an emphasized word, and the notation has already been changed by emphasis. Information indicating whether reading is performed (ON: reading corrected, OFF: reading uncorrected), and details of correction (such as "initial beat lengthening").
[0058]
The emphasis notation creation unit 401 corrects a word heading using the emphasis notation creation rule. In the example of FIG. 22, since the reading correction flags of “one”, “why”, “this”, and “da” are ON, the “emphasis word notation creation rule” of FIG. 21 is applied. As a result, the word "integration" is obtained by converting the pronunciation into katakana, and the words "why", "this", and "da" are converted into hiragana. Further, as for the notation of "da", "a" is used instead of a long sound. On the other hand, since the reading correction flag is OFF for “koto”, “ni”, “n”, “ta”, and “n”, the “non-emphasized word notation creation rule” in FIG. Is used as a word heading as it is. As a result, a result as shown in FIG. 23 is obtained. The result obtained by the emphasis notation creating unit is sent to the emphasis notation output unit 402.
[0059]
The emphasis notation output unit 402 creates a sentence by listing the word headings of the word string, and outputs the created sentence at the same time as the voice output unit 117 outputs the voice, and displays it to the user together with the input sentence. I do. FIG. 24 is an example of display to the user.
[0060]
FIG. 25 is an example of display to the user when emphasis is specified for a part of a sentence. Only the specified portion is emphasized and synthesized, and only that portion is displayed in an emphasized form.
[0061]
Note that the present invention is not limited to the above-described embodiment, and various modifications are possible based on the gist of the present invention. For example, in addition to the character type used in the first embodiment, a rule such as a reading point used in the sentence "i, ka, na, i" or a kana type notation such as "beautiful" is used. be able to. In the first embodiment, only the accent is corrected as the prosody, but the pitch can be corrected according to the type of the input character.
[0062]
【The invention's effect】
As described in detail above, according to the first aspect of the present invention, a sentence inputting means for inputting a sentence including an emphasized word and a language analyzing means for performing a linguistic analysis of the input sentence Prosody processing means for setting prosody information to the language analysis result by the language analysis means, synthesis parameter generation means for generating speech synthesis parameters, speech waveform generation means for generating a speech waveform based on the synthesis parameters, A voice synthesizing device comprising: a voice output unit configured to output a waveform as voice, wherein the language analyzing unit deletes and replaces characters unnecessary for language analysis from the input sentence and divides the input sentence into sentences. Using the analysis means and a notation neutralization rule that is a rule for converting the highlighted word into a neutral word, storing the sentence divided by the vocabulary analysis means, and storing only the neutral word Morphological analysis means for dividing the words into words by referring to the word dictionary, and correcting the reading of the emphasized words converted into neutral words by referring to the reading correction rules for the words divided by the morphological analysis means And emphasis notation processing means for referring to the prosody modification rule to modify the prosody of the word to a prosody, so that (1) a word whose notation has been changed to represent emphasis Thus, there is an effect that speech synthesis can be performed without registering a new word in the word dictionary. Also, (2) there is an effect that natural speech synthesis can be performed with emphasized prosody for a word whose notation has been changed to indicate emphasis.
[0063]
According to the second aspect of the present invention, in the voice synthesizing apparatus according to the first aspect, the word dictionary in the linguistic analysis unit includes a word to which emphasis information is added to determine whether the word is an emphasis word. A lexical analysis unit, wherein the linguistic analysis unit further includes, when the morphological analysis unit determines that the word is an emphasized word, corrects the prosody of the emphasized word by referring to an emphasis processing rule for the emphasized word. Since the word processing means is provided, there is an effect that (1) by emphasizing the prosody of the emphasized word, it is possible to obtain a speech synthesis sound having a natural prosody. Also, (2) there is an effect that by emphasizing the emphasized prosody, a synthesized speech with a natural prosody can be obtained.
[0064]
According to the fourth aspect of the present invention, in the speech synthesizer according to the second aspect, further, a word extracted from all sentences in the input sentence, to which a notation neutralization rule is applied. And extracting a word from the emphasis information storage means and displaying information related to the word by an operation of the user, and determining that the word is an emphasis word. Since this configuration is provided with an emphasis information management unit that adds only the word to the word and adds the word to the word dictionary and registers the word in the word dictionary, it is possible to easily obtain dictionary information on a change when the word is emphasized. effective.
[0065]
According to a fifth aspect of the present invention, in the voice synthesizing apparatus according to the second aspect, the emphasis notation creation rule is further applied to each word information output from the emphasized word processing unit. , A sentence reflecting the prosody of the voice in the notation is created, and the emphasis notation creating means for displaying the sentence is displayed together with the input sentence. .
[Brief description of the drawings]
FIG. 1 is a configuration diagram of an apparatus showing a first embodiment.
FIG. 2 is a classification diagram of characters used in the notation of an emphasis type.
FIG. 3 is a notation neutralization rule used in a dictionary search unit.
FIG. 4 is a flowchart of a dictionary search.
FIG. 5 is an example of a morphological analysis result.
FIG. 6 is an example of a reading correction rule.
FIG. 7 is an application example of a reading correction rule.
FIG. 8 is an example of a prosody modification rule.
FIG. 9 is an application example of a prosody modification rule.
FIG. 10 is a configuration diagram of an apparatus showing a second embodiment.
FIG. 11 is an example of emphasis information.
FIG. 12 is an example of an emphasis word processing rule.
FIG. 13 is an example of emphasized word processing (emphasized word / user not emphasized).
FIG. 14 is an example of emphasized word processing (emphasized word / user's emphasis designation).
FIG. 15 is an example of emphasized word processing (non-emphasized word / user emphasized word designation).
FIG. 16 is a configuration diagram of an apparatus showing a third embodiment.
FIG. 17 is an example of extraction emphasis information.
FIG. 18 is a display example of a user display screen.
FIG. 19 is an example of adding reading change information to a word dictionary.
FIG. 20 is a configuration diagram of an apparatus showing a fourth embodiment.
FIG. 21 is an example of an emphasis notation creation rule.
FIG. 22 is an example of information sent to an emphasis notation creation rule unit.
FIG. 23 is an example of an application result of the emphasis notation creation rule.
FIG. 24 is an example of display to the user (emphasis on the entire sentence).
FIG. 25 is a display example (partial emphasis) to a user.
FIG. 26 is a configuration diagram of a conventional device.
FIG. 27 is an example of a word dictionary search.
FIG. 28 is a diagram showing a notation of emphasis.
[Explanation of symbols]
101 sentence input section
102 Language analysis unit
103 Vocabulary analysis unit
104 Morphological analyzer
105 Dictionary search processing unit
106 Word Dictionary
107 Notation Neutralization Rule
108 Unknown Word Processor
109 Emphasis notation processing unit
110 Reading Correction Rule Storage Unit
111 Prosody modification rule storage
112 syntax analyzer
113 Accent Combination / Rendaku Processing Unit
114 Prosody processing unit
115 Synthesis Parameter Generation Unit
116 sound waveform generator
117 Audio output unit
201 Word dictionary (with emphasis information)
202 Emphasis word processing unit
203 Keyword processing rule storage
301 Emphasis information extraction unit
302 Emphasis information storage
303 Emphasis information management unit
304 Emphasis information confirmation section
305 Emphasis information registration section
401 Emphasis notation creation unit
402 Emphasis notation output unit

Claims (5)

Sentence inputting means for inputting a sentence including the highlighted word with emphasis, language analyzing means for performing language analysis of the input sentence, and setting prosodic information for a result of the language analysis by the language analyzing means. Prosody processing means, a synthesis parameter generation means for generating a speech synthesis parameter based on the result of the language analysis in which the prosody information is set by the prosody processing means, and a speech waveform based on the speech synthesis parameter A voice synthesizing apparatus, comprising: a voice waveform generating unit that performs voice output; and a voice output unit that outputs the voice waveform as voice.
The language analysis means,
Vocabulary analysis means for deleting and replacing characters unnecessary for the linguistic analysis from the input sentence, and for dividing the document in which the unnecessary characters have been deleted and replaced into sentences.
By applying various notation neutralization rules and verifying the possibility of connection between words extracted from the word dictionary , the notation of the emphasized form in the sentence divided by the vocabulary analysis means is converted to the notation of the neutral form Morphological analysis means for dividing the sentence into words and
Depending on the application status of the notation neutralization conventions, as well as modifications to the reading of the enhancement type by applying the read modified rules for pronunciation of the word is divided by the morphological analysis means, divided by the morphological analysis means Correct the prosody stressed by applying the prosody modification rule for prosody of the words were, speech synthesis apparatus characterized by comprising a enhancement representation processing means for outputting to the prosodic processing unit. The speech synthesizer according to claim 1,
The word dictionary in the language analysis means is a word dictionary to which emphasis information as to whether the word is an emphasis word is added,
The language analysis unit further includes, when the morphological analysis unit determines that the word is an emphasized word, refers to an emphasis processing rule for the emphasized word to correct the prosody of the emphasized word. A speech synthesizer comprising: The speech synthesizer according to claim 2,
The speech synthesizer according to claim 1, wherein the emphasis word processing means performs the emphasis process only on a character string specified by a user. The speech synthesizer according to claim 2, further comprising:
For words extracted from all the sentences in the input sentence, emphasis information extraction means for extracting only words to which the notation neutralization rule is applied and storing them in emphasis information storage means,
The user reads out the word from the emphasis information storage means to display information about the word, and gives the emphasis information of the word only to the word determined to be the emphasis word by the user to thereby emphasize the word. A voice synthesizing device comprising: an emphasis information management unit that registers the word to which information is added in the word dictionary. The speech synthesizer according to claim 2, further comprising:
The word processing apparatus further includes a highlighted notation creating unit that creates a sentence reflecting the prosody of the voice in the notation by applying a highlighted notation creating rule to each word information output from the emphasized word processing unit and displays the sentence together with the input sentence. A speech synthesizer characterized by the following.

Images