JP2008146019A - System for creating dictionary for speech synthesis, semiconductor integrated circuit device, and method for manufacturing semiconductor integrated circuit device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To create a subset speech dictionary with which a speech can be synthesized with good pronunciation quality for a prescribed target utterance target document using a sufficient necessary amount of data. <P>SOLUTION: The system for creasing a dictionary for speech synthesis includes: a first speech synthesis dictionary memory means 182 that stores dictionary data composing a first dictionary for speech synthesis; a second speech synthesis dictionary creating means 120 that analyzes an utterance target document, checks frequency of occurrence of each word composing the utterance target sentence, determines words to be stored in a second dictionary for speech synthesis on the basis of the frequency of occurrence, and creates the second dictionary for speech synthesis using the dictionary data stored in the first dictionary for speech synthesis corresponding to the determined words to be stored; and a speech synthesis means 130 that creates synthesized speech corresponding to the utterance target document using the second dictionary for speech synthesis. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a speech synthesis dictionary creation system, a semiconductor integrated circuit device, and a method for manufacturing a semiconductor integrated circuit device.

  The TTS speech synthesis LSI that synthesizes speech from text data, which is a collection of character data, has a parametric method that synthesizes sound by modeling the utterance process of the human body, and phoneme data consisting of speech data of a real person. In addition, a concati-native system that combines them as needed and synthesizes them by partially transforming the joints, and further develops speech synthesis from language-based analysis to form synthesized speech. There are many methods such as a corpus-based method.

  Regardless of the method, before converting from text to sound, there is a conversion dictionary (database) that translates the textual representation of the notation marked with SHIFT-JIS code into how to pronounce it. It is essential.

Further, in the concati native method and the corpus-based method, a dictionary (database) from “reading” to “phoneme” is also required.
JP 2003-208191 A

  In a single-chip TTS-LSI with limited on-chip resources (ROM capacity, etc.), if the dictionary file for speech synthesis that can be mounted is limited to a relatively small vocabulary, the vocabulary that can be handled is limited, so sufficient pronunciation Quality may not be obtained.

  In the case of a small-capacity system, the “notation → reading” data dictionary corresponding to a sufficient vocabulary does not have a “phoneme” dictionary that covers many cases effective for improving voice quality. If the vocabulary that was leaked from the dictionary was included, there was a situation where the sound quality deteriorated at the corresponding part or reading was impossible.

  The present invention has been made in view of the technical problems as described above, and the object of the present invention is to synthesize speech with good pronunciation quality with a necessary and sufficient amount of data for a predetermined utterance target sentence. To create a simple subset speech dictionary.

(1) The present invention
The second speech synthesis dictionary having a smaller amount of data than the first speech synthesis dictionary from the first speech synthesis dictionary, which is a set of dictionary data necessary for generating synthesized speech corresponding to the utterance target sentence. A speech synthesis dictionary creation system for creating
A first speech synthesis dictionary storage means for storing dictionary data constituting the first speech synthesis dictionary;
Analyzing the utterance target sentence, examining the appearance frequency of each word constituting the utterance target sentence, determining a storage word in the second speech synthesis dictionary based on the appearance frequency, and corresponding to the determined storage word Second speech synthesis dictionary creation means for creating a second speech synthesis dictionary using dictionary data stored in the first speech synthesis dictionary;
Speech synthesis means for generating synthesized speech corresponding to the utterance target sentence using the second speech synthesis dictionary.

  The first speech synthesis dictionary is a full-set dictionary (large-capacity dictionary) having dictionary data of a scale capable of generating synthesized speech for an arbitrary utterance target sentence, and the second speech synthesis dictionary is a specific dictionary. It is a subset dictionary (small-capacity dictionary) having data of a scale capable of generating synthesized speech for an utterance target sentence.

  The first speech synthesis dictionary includes, for example, a vocabulary dictionary (“notation → reading” data dictionary), a phoneme dictionary (a dictionary covering many cases effective for improving voice quality), and the like. These dictionary data are stored in the first speech synthesis dictionary storage means and function as a dictionary database. Note that the type of dictionary is determined according to the speech synthesis method. For example, both the vocabulary dictionary and the phoneme dictionary may be included, or only the vocabulary dictionary may be included.

  The vocabulary dictionary is a dictionary for performing front-end processing in text-to-speech processing, and is a dictionary in which symbolic linguistic representation (for example, reading data corresponding to text notation) corresponding to text notation is stored. .

  In front-end processing, the numbers and abbreviations in the text are converted into expressions for reading (called text normalization, preprocessing, tokenization, etc.), and each word is converted into a phonetic symbol, Processing to divide into prosodic units such as idioms, clauses and sentences (processing to assign phonetic symbols to words as text-to-phoneme (TTP) conversion or grapheme-to-phoneme (GTP)) conversion The symbolic language expression is created and output by combining phonetic symbols and prosodic information.

  In the text normalization step, conversion processing is performed so that synonyms, numbers, abbreviations, etc. included in the text can be uttered. Many TTS (text-to-speech) systems do not analyze the meaning of the input text, but use different heuristics, such as examining the words before and after, and using statistical appearance frequencies. Distinguish between synonyms.

  The phoneme dictionary is a dictionary that stores waveform information of an actual sound (phoneme) corresponding to a symbolic language expression that is output from the front end. The main technologies for generating speech waveforms at the back end include concatenative synthesis and formant synthesis. Linked synthesis is basically a method of combining recorded audio fragments.

  The speech synthesis means performs front-end processing and back-end processing based on the vocabulary information and sound information stored in the first speech synthesis dictionary, and generates synthesized speech corresponding to the received utterance target sentence.

  For example, the second speech synthesis dictionary creating means may preferentially determine words with high appearance frequency as stored words. For example, a specific ratio (for example, 80%) of the storage capacity determined to be assignable to the second speech conversion dictionary in advance may be assigned in order from the vocabulary with the highest appearance frequency. At that time, if the appearance frequency is not a certain number of times (for example, twice), the allocation may be stopped even if the above-mentioned ratio is not reached. Since the appearance frequency generally has a “long tail” type distribution, it can be expected that many parts of the target sentence are covered in this way.

  Since the speech synthesis means generates the synthesized speech corresponding to the utterance target sentence using the second speech synthesis dictionary, the user can confirm the result of speech synthesis of the utterance target sentence.

  According to the present invention, a specific speech target sentence is analyzed, dictionary data necessary and sufficient for speech synthesis of the specific speech target sentence is extracted from the first speech synthesis dictionary, and the first speech is extracted. A second speech synthesis dictionary having a smaller amount of data than the synthesis dictionary can be generated.

  Therefore, even if a speech dictionary file that can be mounted on a single-chip TTS-LSI with limited on-chip resources (ROM capacity, etc.) is limited to a relatively small vocabulary, speech synthesis can be performed with high accuracy for a specific utterance target sentence. A subset dictionary (second speech synthesis dictionary) can be generated.

  In the present invention, the data amount of the vocabulary dictionary can be reduced by selecting and extracting the vocabulary stored in the second speech synthesis dictionary. By reducing the data amount of the vocabulary dictionary, the data amount of the corresponding phoneme dictionary is also reduced as a result, so that the data amount of both the vocabulary dictionary and the phoneme dictionary of the second speech synthesis dictionary can be reduced.

(2) The speech synthesis dictionary creation system of the present invention includes:
An utterance target that changes an utterance target sentence that replaces an unstored word that is not stored in the second speech synthesis dictionary with words stored in the second speech synthesis dictionary among words constituting the utterance target sentence. Text change means;
It is characterized by including.

  The replacement here may be, for example, a case where an unstored word is replaced with its synonym (synonym stored in the second dictionary for speech synthesis), or an unstored word is replaced with its kana notation (a dictionary for kana notation is It may be replaced with the one stored in the second speech synthesis dictionary.

  According to the present invention, it is possible to improve the accuracy of speech synthesis of an utterance target sentence without increasing the number of words stored in the second speech synthesis dictionary.

  Note that the speech synthesis means uses the second speech synthesis dictionary to generate synthesized speech corresponding to the utterance target sentence of the changed word, so that the user can confirm the result of speech synthesis of the utterance target sentence after the change. it can.

(3) A speech synthesis dictionary creation system according to the present invention includes:
The utterance target sentence changing means is:
A speech synthesis dictionary characterized by recording a change history regarding replacement of words constituting an utterance target sentence is created.

  The change history includes the changed word and the original language information of the utterance target sentence corresponding to the changed word. Accordingly, when a predetermined word / phrase is changed a plurality of times, at least the original word (the word / phrase included in the utterance target sentence given first) and the finally changed word information are included.

  The change history may be generated separately from the utterance target sentence, or may be generated in a format in which a comment of the change history is inserted into the utterance target sentence.

(4) The speech synthesis dictionary creation system according to the present invention includes:
The utterance target sentence changing means is:
A synonym for analyzing whether or not there is a synonym with the stored word in the second speech synthesis dictionary for the unstored word, and replacing the unstored word in the utterance target sentence with a synonym if there is a synonym Synonym replacement processing means for performing word replacement processing is included.

  For example, when the first word and the second word included in the utterance target sentence are synonyms and can be replaced, the first word is a storage word of the second speech synthesis dictionary, and the second word When the word is not a stored word of the second speech synthesis dictionary, in the present invention, it is possible to perform an utterance target sentence changing process for replacing the second word in the utterance target sentence with the first word.

  For example, synonyms of unstored words may be searched using a synonym dictionary in which synonyms are defined. For example, the synonym dictionary is searched for synonyms for each unstored word in the utterance target sentence, the second speech synthesis dictionary is searched, and the synonym obtained as a result of the search is stored in the second speech synthesis dictionary. It is checked whether or not it is a word, and if it is a stored word, a replacement process for replacing the unstored word of the utterance target sentence with the stored word may be performed.

  According to the present invention, it is possible to increase the accuracy of speech synthesis of an utterance target sentence without changing the meaning content of the utterance target sentence and without increasing the number of words stored in the second speech synthesis dictionary.

  The speech synthesis means uses the second dictionary for speech synthesis to generate synthesized speech corresponding to the utterance target sentence after replacement with the synonym, so that the user can synthesize speech of the utterance target sentence after replacement with the synonym. The result can be confirmed.

(5) A speech synthesis dictionary creation system according to the present invention includes:
The utterance target sentence changing means is:
It includes a kana replacement processing means for performing a kana replacement process for replacing the unstored word with a kana notation representing the reading of the word.

  Here, it is assumed that the second dictionary for speech synthesis includes dictionary data for performing speech synthesis corresponding to kana notation.

  According to the present invention, speech synthesis can be performed on a specific utterance target sentence by replacing special words and phrases having a low appearance frequency with kana notation (although some inflection and accent may be unnatural). A second speech synthesis dictionary can be created.

(6) The speech synthesis dictionary creation system according to the present invention includes:
Editing process for accepting an evaluation input for an utterance target sentence synthesized by speech using the second speech synthesis dictionary and determining or changing the second speech synthesis dictionary or utterance target sentence according to the contents of the evaluation input Means,
It is characterized by including.

  For example, the evaluation input may be returned as either OK or NG.

  In this way, the user can listen to and confirm the synthesized speech of the utterance target sentence generated using the second speech synthesis dictionary that is being created, while confirming the second speech synthesis dictionary or utterance target sentence. Can be confirmed or changed. Therefore, the second speech synthesis dictionary editing process can be performed while confirming the result in real time, so that a user-friendly speech synthesis dictionary creation system can be provided for the user.

(7) The speech synthesis dictionary creation system of the present invention includes:
The editing processing means includes
Accepts user-specified input for stored words in the second dictionary for speech synthesis,
The second speech synthesis dictionary creation means includes:
A storage word is determined based on the user's designated input.

  For example, after determining the storage word according to the appearance frequency of each word constituting the utterance target sentence, for the words to be included in the remaining capacity, the specification input from the user is accepted and determined according to the specification input. Also good.

  In this way, it is possible to make an adjustment that directly reflects the user's intention regarding the contents of the words stored in the second speech synthesis dictionary. Accordingly, it is possible to edit the second speech synthesis dictionary that closely corresponds to the individual needs of individual users.

(8) The present invention
A non-volatile storage unit that stores dictionary data that constitutes the second dictionary for speech synthesis generated by the dictionary for speech synthesis according to any one of the above,
And a synthesized speech data generation processing unit that generates synthesized speech data corresponding to a predetermined utterance target sentence using dictionary data stored in the non-volatile storage unit.

(9) The present invention
A method of manufacturing a semiconductor integrated circuit device for speech synthesis, including a non-volatile storage unit,
The semiconductor integrated circuit device analyzes the speech target sentence scheduled for speech synthesis, examines the appearance frequency of each word constituting the speech target sentence, and based on the appearance frequency, stores the words stored in the second speech synthesis dictionary. Determining and generating a second speech synthesis dictionary using dictionary data stored in the first speech synthesis dictionary corresponding to the determined stored word;
Generating synthesized speech corresponding to the utterance target sentence using the second dictionary for speech synthesis;
Writing the dictionary data constituting the generated second speech synthesis dictionary into the nonvolatile storage unit of the semiconductor integrated circuit device;
It is characterized by including.

  DESCRIPTION OF EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. The embodiments described below do not unduly limit the contents of the present invention described in the claims. Also, not all of the configurations described below are essential constituent requirements of the present invention.

  FIG. 1 is a diagram for explaining a semiconductor integrated circuit device having a built-in speech synthesis dictionary created by the speech synthesis dictionary creation system and the speech synthesis dictionary creation system of the present embodiment.

  Reference numeral 100 denotes a speech synthesis dictionary creation system according to the present embodiment, which is a large-capacity dictionary (first speech synthesis dictionary that is a set of dictionary data necessary for generating synthesized speech corresponding to an utterance target sentence 101. ) A speech synthesis dictionary creation system that creates a small-capacity dictionary (second speech synthesis dictionary) 184 with a smaller amount of data than a large-capacity dictionary (first speech synthesis dictionary) 182 from This can be realized by installing a large-capacity dictionary 182 for speech synthesis corresponding to TTS, a subset dictionary creation software 122 for speech synthesis, and speech synthesis software 132 on a computer.

  The speech synthesis large-capacity dictionary 182 functions as a first speech synthesis dictionary storage unit in which dictionary data constituting the first speech synthesis dictionary is stored.

  The speech synthesis subset dictionary creation software 102 analyzes the utterance target sentence, checks the appearance frequency of each word constituting the utterance target sentence, and based on the appearance frequency, the small-capacity dictionary (second speech synthesis dictionary) 184 And a small-capacity dictionary (for second speech synthesis) using dictionary data stored in a large-capacity dictionary (first speech synthesis dictionary) 182 corresponding to the determined storage word. (Dictionary) 184 functions as second speech synthesis dictionary creation means.

  Further, the speech synthesis subset dictionary creation software 122 generates a small-capacity dictionary for unstored words that are not stored in the small-capacity dictionary (second speech synthesis dictionary) 184 among the words constituting the utterance target sentence. (Second speech synthesis dictionary) The utterance target sentence changing means for changing the utterance target sentence to be replaced with the stored word of 184 may be used.

  The speech synthesis subset dictionary creation software 122 accepts an evaluation input for the speech-sentence text synthesized using the small-capacity dictionary (second speech synthesis dictionary) 184, and the second is generated according to the content of the evaluation input. The voice synthesis dictionary or the speech processing sentence may be functioned as editing processing means for confirming or changing the sentence.

  The speech synthesis software 132 functions as speech synthesis means for generating synthesized speech corresponding to the utterance target sentence using the small-capacity dictionary (second speech synthesis dictionary) 184. In practice, it is also possible to generate synthesized speech corresponding to the utterance target sentence using the large-capacity dictionary (second speech synthesis dictionary) 182.

  The speech synthesis dictionary creation system 100 according to the present embodiment determines a stored word based on an utterance target sentence and extracts dictionary data corresponding to the stored word from a large-capacity dictionary (first speech synthesis dictionary) 182. It is stored in the small capacity dictionary (second speech synthesis dictionary) 184.

  Then, the dictionary data of the small-capacity dictionary is written into the ROM (nonvolatile storage unit) of the TTS-LSI (an example of a semiconductor integrated circuit device) 10 to create a small-capacity dictionary.

  A TTS-LSI (an example of a semiconductor integrated circuit device) 10 is a semiconductor integrated circuit device that includes a small-capacity dictionary 30 and a speech synthesis system 20 and generates synthesized speech data corresponding to a predetermined utterance target sentence. The small-capacity dictionary 30 functions as a non-volatile storage unit in which dictionary data constituting the speech synthesis dictionary is stored. The speech synthesis system 20 functions as a synthesized speech data generation processing unit that generates synthesized speech data corresponding to a predetermined utterance target sentence using dictionary data stored in the nonvolatile storage unit.

  In the present embodiment, it can be mounted as in a specific use, for example, when there is a specific use for a vocabulary to be read or a TTS-LSI (an example of an integrated circuit device) 10 in which a text to be read is known in advance. The voice dictionary file is relatively limited to small vocabulary.

  A small-capacity dictionary (subset dictionary) 30 for TTS-LSI (an example of an integrated circuit device) 10 includes a large-capacity dictionary (full set dictionary) 182 on the personal computer 100 and a TTS-LSI (an example of an integrated circuit device). The dictionary data constituting the small-capacity dictionary (second speech synthesis dictionary) created by extracting the dictionary data corresponding to the vocabulary necessary for the predetermined utterance text to be synthesized at 10 is written.

  In this way, a dictionary corresponding to a specific application of the TTS-LSI (an example of an integrated circuit device) 10 can be created, so that a sufficient performance can be ensured with a small-capacity dictionary. Further, when the utterance target sentence is known in advance, a dictionary limited to only the vocabulary of the utterance target sentence is created, so that waste of resources can be eliminated, and TTS-LSI (an example of an integrated circuit device) 10 can be optimized.

  FIG. 2 is an example of a functional block diagram of the speech synthesis dictionary creation system of the present embodiment. Note that the speech synthesis dictionary creation system 100 of the present embodiment does not need to include all the components (each unit) in FIG. 2, and may have a configuration in which some of them are omitted.

  The operation unit 160 is for inputting a user operation or the like as data, and the function can be realized by hardware such as an operation button, an operation lever, a touch panel, or a microphone.

  The storage unit 170 serves as a work area for the processing unit 110, the communication unit 196, and the like, and its function can be realized by hardware such as a RAM.

  The information storage medium 180 (a computer-readable medium) stores programs, data, and the like, and functions as an optical disk (CD, DVD, etc.), a magneto-optical disk (MO), a magnetic disk, a hard disk, and a magnetic disk. It can be realized by hardware such as a tape or a memory (ROM).

  The information storage medium 180 stores a program for causing the computer to function as each unit of the present embodiment and auxiliary data (additional data), and also stores large-capacity dictionary data for speech synthesis, and the first speech synthesis dictionary. It functions as the storage unit 182. Note that the information storage medium 180 may also store dictionary data of the second speech synthesis word extracted from the first speech synthesis dictionary.

  The processing unit 100 performs various processes of the present embodiment based on a program (data) stored in the information storage medium 180, data read from the information storage medium 180, and the like. That is, the information storage medium 180 stores a program for causing a computer to function as each unit of the present embodiment (a program for causing a computer to execute processing of each unit).

  The display unit 190 outputs an image generated according to the present embodiment, and functions thereof are a CRT display, an LCD (liquid crystal display), an OELD (organic EL display), a PDP (plasma display panel), and a touch panel display. It can be realized by hardware such as.

  The sound output unit 192 outputs the synthesized speech generated by the present embodiment, and the function can be realized by hardware such as a speaker or headphones.

  The communication unit 196 performs various controls for communicating with the outside (for example, a host device or another terminal), and functions thereof include hardware such as various processors or communication ASICs, It can be realized by a program.

  Note that a program (data) for causing a computer to function as each unit of the present embodiment is transferred from the information storage medium included in the host device (server device) to the information storage medium 180 (or storage unit 170) via the network and communication unit 196. You may make it deliver. Use of the information storage medium of such a host device (server device or the like) can also be included in the scope of the present invention.

  The processing unit 110 (processor) performs various processes using the storage unit 170 as a work area based on operation data, a program, and the like from the operation unit 160. The functions of the processing unit 110 can be realized by hardware such as various processors (CPU, DSP, etc.), ASIC (gate array, etc.), and programs.

  The processing unit 110 includes a second speech synthesis dictionary creation unit 120, a synthesized speech data generation processing unit 130, an utterance target sentence change processing unit 140, and a dictionary editing processing unit 150.

  The second speech synthesis dictionary creation unit 120 analyzes the utterance target sentence, examines the appearance frequency of each word constituting the utterance target sentence, and based on the appearance frequency, stores the words stored in the second speech synthesis dictionary. The second speech synthesis dictionary is generated using the dictionary data stored in the first speech synthesis dictionary corresponding to the determined stored word.

  The synthesized speech data generation processing unit 130 generates synthesized speech data corresponding to the utterance target sentence using the second speech synthesis dictionary.

  The utterance target sentence change processing unit 140 replaces unstored words that are not stored in the second speech synthesis dictionary among the words constituting the utterance target sentences with the storage words of the second speech synthesis dictionary. Change the text to be spoken.

  The utterance target sentence change processing unit 140 includes a change history recording processing unit 142, a synonym replacement processing unit 144, and a kana replacement processing unit 146.

  The change history recording processing unit 142 performs a process of recording a change history related to replacement of words constituting the utterance target sentence.

  The synonym replacement processing unit 144 analyzes whether an unstored word has a synonym with a stored word in the second speech synthesis dictionary, and if there is a synonym, the unstored utterance target sentence is stored. Perform synonym replacement processing to replace words with synonyms.

  The kana replacement processing unit 146 performs kana replacement processing for replacing an unstored word with a kana notation representing the reading of the word.

  The dictionary editing processing unit 150 receives an evaluation input with respect to the utterance target sentence synthesized by using the second speech synthesis dictionary, and determines the second speech synthesis dictionary or the utterance target sentence according to the contents of the evaluation input. Alternatively, change processing is performed.

  Further, the dictionary editing processing unit 150 accepts a user's designation input for a storage word in the second speech synthesis dictionary, and the second speech synthesis dictionary creation unit 120 determines a storage word based on the user's designation input. You may do it.

  Next, the operation of the present invention will be described using a specific example.

  FIG. 3 is a flowchart for explaining the flow of processing of the present embodiment.

  First, the utterance target sentence is profiled (step S10). For example, utterance target sentences are broken down into vocabularies, and the appearance frequency of each vocabulary is tabulated.

  Next, frequent word dictionary extraction (primary extraction) is performed (step S20). For example, based on the profiling data, a specific ratio (for example, 80%) of the storage capacity determined to be assignable to the dictionary in advance is assigned in order from the vocabulary with the highest appearance frequency. At this time, if the appearance frequency is not a certain number of times (for example, twice), the allocation is stopped even if the above ratio is not reached. Since the appearance frequency generally has a “long tail” type distribution, it can be expected that many parts of the target sentence are covered by the subset dictionary at this stage.

  Next, an utterance trial of the utterance target sentence is performed using the subset dictionary after the primary extraction, and the user is confirmed (step S30).

  Then, a confirmation input from the user (for example, OK or NG) is received. If OK, the process ends (confirms the contents of the subset dictionary with the contents after the primary extraction). If NG, the subsequent processes are performed ( Step S40).

  Next, a low-appearance vocabulary replacement process is performed (step S40). Use the “Synonym” dictionary to check whether vocabulary replacement is possible for the vocabulary leaked during the primary extraction process. A case where the vocabulary is replaced with an already assigned vocabulary and a case where a plurality of vocabularies can be combined into one by the replacement are examined, and the utterance target sentence is changed by the replacement (step S50).

  Next, an utterance trial of the utterance target sentence after the change is performed using the subset dictionary after the primary extraction, and the user is confirmed (step S60). The confirmation here may be, for example, confirmation in a format in which the changed portion is displayed and output on the screen as text or the like, but even in that case, it is preferable to confirm the sound after the change because there is no mistake.

  Whether the results can be replaced or not can be shown to the user and added to the dictionary after judgment, and those that are replaced can be replaced with priority. is there. At this time, since it is not necessary to add a dictionary to those already assigned, the vocabulary of the target sentence is replaced. In addition, when sorting in the order of frequency and adding to the subset dictionary within the range of the remaining percentage that has already been assigned, search for whether there is a vocabulary that can be replaced. The utterance target sentence may be replaced with a newly added vocabulary.

  Then, a confirmation input from the user (for example, OK or NG) is received. If OK, the process ends (confirms the contents of the subset dictionary with the contents after the primary extraction). If NG, the subsequent processes are performed ( Step S70).

  Next, a process of recording the change of the utterance target sentence as a change history is performed (step S80).

  FIG. 4 is a diagram for explaining an example of a change history recording process at the time of replacement.

  For example, as shown in FIG. 4, the change history of the utterance target sentence may be left in a format in which comments 220, 230, and 240 are inserted into the utterance target sentence 200 itself. For example, the comment may be distinguished from the utterance target sentence by surrounding it with brackets (222 and 226, 232 and 238, 232 and 236 in FIG. 4) to indicate that it is a comment.

  Here, 210 is a replacement word (a part of the utterance target sentence). Comments 220 and 240 indicate that the portion between these comments is the replacement word before and after the replacement. Reference numeral 230 denotes a comment indicating that the original word (the phrase included in the original utterance target sentence) corresponding to the replacement word is “performance”.

  Next, it is confirmed whether or not manual editing is performed for the user. If so, manual dictionary editing processing is performed (steps S90 and S100). The vocabulary that has not been extracted from the utterance target text may be sorted in the order of frequency, and then added to the subset dictionary within the range of the remaining ratio that has already been assigned, from the highest frequency.

  Next, for words that cannot be dealt with by the above processing, registration as words is abandoned and converted to “single pronunciation” by inserting ruby into the target sentence (step S110).

  FIG. 5 is a diagram for explaining an example of a change history recording process at the time of ruby swing (kana replacement process).

  For example, if the vocabulary “quantum theory” cannot be registered, it is converted into a ruby (either katakana or hiragana) as shown in 310 of FIG. At that time, text TAGing may be performed as shown in FIG. 5 to indicate that the corresponding part is ruby, and that the original vocabulary is “quantum theory” although it is not pronounced.

  That is, as shown in FIG. 5, comments 320, 330, and 340 are inserted into the utterance target sentence 300 itself. Here, 310 is the kana after kana conversion (a part of the utterance target sentence). Comments 320 and 340 indicate that the portion between these comments is a kana conversion word before and after the kana conversion word. Reference numeral 330 denotes a comment indicating that the original word (a phrase included in the original utterance target sentence) corresponding to the kana conversion word is “quantum theory”.

  Since the subset dictionary (second speech synthesis dictionary) includes speech synthesis data for kana notation, words represented by kana characters can be pronounced. However, since it is only recognized as a kana character, it is difficult to produce the inflection and accent peculiar to the word, and the pronunciation is similar to a stick reading.

  Therefore, an utterance trial of the utterance target sentence after the change is performed using the subset dictionary and confirmed with the user (step S120).

  Then, a confirmation input (for example, OK or NG) from the user is accepted, and if OK, the process ends (confirms the contents of the subset dictionary with the contents after the primary extraction). If NG, the process returns to step S100 and thereafter. Processing is performed (step S130).

  In the above embodiment, the extraction of the vocabulary dictionary of the subset dictionary has been described as an example. According to this method, by narrowing down the vocabulary, the phonemes can be narrowed down to only those corresponding to the extracted vocabulary, and as a result, the subset phoneme dictionary can also be reduced.

  However, if there is a problem with the subset phoneme dictionary size, the primary extraction may be retried by changing the ratio.

  FIG. 6 is a diagram for explaining the configuration of a single chip TTS-LSI (semiconductor integrated circuit device) on which a subset dictionary is mounted.

  The single chip TTS-LSI 10 includes a subset dictionary 30. The subset dictionary 30 functions as a nonvolatile storage unit in which dictionary data constituting the second speech synthesis dictionary generated by the speech synthesis dictionary creation system of the present embodiment is stored. The subset dictionary 30 includes a vocabulary dictionary 32 and a phoneme dictionary 34 and can be realized by a ROM, a flash EEPROM, or the like.

  The vocabulary dictionary 32 is a dictionary for performing front-end processing in text-to-speech processing, and is a dictionary in which symbolic linguistic representation (for example, reading data corresponding to text notation) corresponding to text notation is stored. is there.

  In front-end processing, the numbers and abbreviations in the text are converted into expressions for reading (called text normalization, preprocessing, tokenization, etc.), and each word is converted into a phonetic symbol, Processing to divide into prosodic units such as idioms, clauses and sentences (processing to assign phonetic symbols to words as text-to-phoneme (TTP) conversion or grapheme-to-phoneme (GTP)) conversion The symbolic language expression is created by combining phonetic symbols and prosodic information, and used as the output of the front end.

  The phoneme dictionary 34 is a dictionary that stores waveform information of an actual sound (phoneme) corresponding to a symbolic language expression that is an output of the front end.

  The subset dictionary 30 stores data of the second speech synthesis dictionary created by the speech synthesis dictionary creation system. For example, you may comprise by the phoneme dictionary which consists of the vocabulary dictionary produced | generated by the procedure demonstrated in FIG. 3, and the phoneme dictionary data required for the said vocabulary dictionary.

  The single chip TTS-LSI 10 includes a host I / F 50. The host I / F 50 is an interface block for exchanging commands and data with the host computer. The host I / F 50 includes a TTS command / data buffer 52 in which an utterance target sentence (text data) instructed by the host is stored. The utterance target sentence is input to the synthesized voice data generation processing unit 20.

  The single chip TTS-LSI 10 includes a synthesized voice data generation processing unit 20. The synthesized speech data generation processing unit 20 functions as a synthesized speech generation unit that generates synthesized speech data corresponding to a predetermined utterance target sentence using dictionary data (subset dictionary) stored in the nonvolatile storage unit 30. The synthesized speech data generation processing unit 20 includes a notation → sound notation conversion block 22, a phoneme selection unit 24, a pronunciation block 26, and a filter processing unit 28. The function of each part may be realized by providing a dedicated circuit, or may be realized by the CPU executing a program for realizing the function of each part. The function of the synthesized speech data generation processing unit 20 is equivalent to the function of the synthesized speech data generation processing unit 130 of the speech synthesis dictionary creation system in FIG.

  The notation-to-speech notation conversion block 22 searches the vocabulary dictionary 32 and sends the utterance target sentence received as a symbolic language expression 23 to the phoneme selection unit.

  The phoneme selection unit 24 receives the encoded language expression 23 of the utterance target sentence, searches the phoneme dictionary 34, and passes a set of phonemes corresponding to the encoded language expression 23 to the pronunciation block 26.

  The pronunciation block 26 generates a synthesized speech waveform 27 based on the set of phonemes.

  The filter processing unit 28 uses the filter to change the sound quality of the synthesized speech waveform or change to the voice of another character.

  The single chip TTS-LSI 10 includes a speaker I / F 40. The synthesized speech waveform filtered by the filter processing unit 28 is output to an external speaker via the amplifier 42 of the speaker I / F 40.

  The single-chip TTS-LSI 10 according to the present embodiment has only a small-capacity subset dictionary and generates highly accurate synthesized speech data for a predetermined utterance target sentence corresponding to a device in which the single-chip TTS-LSI 10 is incorporated. can do.

  FIG. 7 is a flowchart for explaining the method for manufacturing the semiconductor integrated circuit device of the present embodiment. The semiconductor integrated circuit device of the present embodiment is a semiconductor integrated circuit device including a synthesized speech data generation processing unit and a non-volatile storage unit storing dictionary data used for speech synthesis processing, and is manufactured through the following steps.

  First, an utterance target sentence scheduled to be uttered by a semiconductor integrated circuit device is analyzed, the appearance frequency of each word constituting the utterance target sentence is examined, and a word stored in the second speech synthesis dictionary is determined based on the appearance frequency. A second speech synthesis dictionary is generated using dictionary data stored in the first speech synthesis dictionary corresponding to the determined stored word (step S10).

  Next, synthesized speech corresponding to the utterance target sentence is generated using the second speech synthesis dictionary (step S20). An evaluation input from the user is accepted for the synthesized speech generated here. If OK, the content of the second speech synthesis dictionary is confirmed, and if it is NG, editing of the second speech synthesis dictionary is continued. May be.

  Next, the generated dictionary data constituting the second dictionary for speech synthesis is written into the nonvolatile storage unit of the semiconductor integrated circuit device (step S30). For example, as a mask ROM, dictionary data constituting the second dictionary for speech synthesis may be written in the nonvolatile storage unit when the semiconductor integrated circuit device is manufactured.

  In addition, this invention is not limited to this embodiment, A various deformation | transformation implementation is possible within the range of the summary of this invention.

  It can also be applied to TTS systems for languages other than Japanese.

1 is a diagram for explaining a speech synthesis dictionary creation system and a semiconductor integrated circuit device according to an embodiment. FIG. An example of a functional block diagram of the dictionary system for speech synthesis of this embodiment. The flowchart for demonstrating the flow of the process of this Embodiment. The figure for demonstrating an example of the change log recording process at the time of replacement. The figure for demonstrating an example of the change log recording process at the time of ruby swing (kana replacement process). The figure for demonstrating the structure of the single chip TTS-LSI (semiconductor integrated circuit device) by which a subset dictionary is mounted. 9 is a flowchart for explaining a manufacturing method of the semiconductor integrated circuit device of the present embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Semiconductor integrated circuit device (TTS-LSI), 20 synthetic speech data generation processing part (speech synthesis system), 22 notation-to-speech notation conversion block, 24 phoneme selection part, 26 pronunciation block, 28 filter processing part, 30 small quantity dictionary (Subset dictionary), 32 vocabulary dictionary, 34 phoneme dictionary, 40 speaker I / F, 50 host I / F, 100 speech synthesis dictionary creation system, 110 processing unit, 120 second speech synthesis dictionary creation unit, 122 subset Dictionary creation software, 130 synthesized speech data generation processing unit, 132 speech synthesis software, 140 utterance target sentence changing unit, 142 change history recording processing unit, 144 synonym replacement processing unit, 146 kana replacement processing unit, 150 dictionary editing processing , 182 First speech synthesis dictionary storage unit (large-capacity dictionary), 18 4 large-capacity dictionary

Claims (9)

The second speech synthesis dictionary having a smaller amount of data than the first speech synthesis dictionary from the first speech synthesis dictionary, which is a set of dictionary data necessary for generating synthesized speech corresponding to the utterance target sentence. A speech synthesis dictionary creation system for creating
A first speech synthesis dictionary storage means for storing dictionary data constituting the first speech synthesis dictionary;
Analyzing the utterance target sentence, examining the appearance frequency of each word constituting the utterance target sentence, determining a storage word in the second speech synthesis dictionary based on the appearance frequency, and corresponding to the determined storage word Second speech synthesis dictionary creation means for creating a second speech synthesis dictionary using dictionary data stored in the first speech synthesis dictionary;
A speech synthesis dictionary creating system comprising speech synthesis means for generating synthesized speech corresponding to an utterance target sentence using a second speech synthesis dictionary. In claim 1,
An utterance target that changes an utterance target sentence that replaces an unstored word that is not stored in the second speech synthesis dictionary with words stored in the second speech synthesis dictionary among words constituting the utterance target sentence. Text change means;
A dictionary creation system for speech synthesis, characterized by comprising: In claim 2,
The utterance target sentence changing means is:
A speech synthesis dictionary creation system, characterized by creating a speech synthesis dictionary characterized by recording a change history related to replacement of words constituting an utterance target sentence. In any one of Claims 2 thru | or 3.
The utterance target sentence changing means is:
A synonym for analyzing whether or not there is a synonym with the stored word in the second speech synthesis dictionary for the unstored word, and replacing the unstored word in the utterance target sentence with a synonym if there is a synonym A speech synthesis dictionary creation system comprising synonym replacement processing means for performing word replacement processing. In any of claims 2 to 4,
The utterance target sentence changing means is:
A speech synthesis dictionary creation system comprising kana replacement processing means for performing a kana replacement process for replacing the unstored word with a kana notation representing the reading of the word. In any one of Claims 1 thru | or 5,
Editing process for accepting an evaluation input for an utterance target sentence synthesized by speech using the second speech synthesis dictionary and determining or changing the second speech synthesis dictionary or utterance target sentence according to the contents of the evaluation input Means,
A dictionary creation system for speech synthesis, characterized by comprising: In any one of Claims 1 thru | or 6.
The editing processing means includes
Accepts user-specified input for stored words in the second dictionary for speech synthesis,
The second speech synthesis dictionary creation means includes:
A dictionary system for speech synthesis, wherein a storage word is determined based on a user's designated input. A non-volatile storage unit storing dictionary data constituting a second speech synthesis dictionary generated by the speech synthesis dictionary creation system according to claim 1;
A synthesized voice data generation processing unit that generates synthesized voice data corresponding to a predetermined utterance target sentence using dictionary data stored in the nonvolatile storage unit;
A semiconductor integrated circuit device comprising: A method of manufacturing a semiconductor integrated circuit device for speech synthesis, including a non-volatile storage unit,
The semiconductor integrated circuit device analyzes the speech target sentence scheduled for speech synthesis, examines the appearance frequency of each word constituting the speech target sentence, and based on the appearance frequency, stores the words stored in the second speech synthesis dictionary. Determining and generating a second speech synthesis dictionary using dictionary data stored in the first speech synthesis dictionary corresponding to the determined stored word;
Generating synthesized speech corresponding to the utterance target sentence using the second dictionary for speech synthesis;
Writing the dictionary data constituting the generated second speech synthesis dictionary into the nonvolatile storage unit of the semiconductor integrated circuit device;
A method for manufacturing a semiconductor integrated circuit device, comprising:

Images