JP3681111B2 - Speech synthesis apparatus, speech synthesis method, and speech synthesis program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice synthesizer, in which individual and dialectal accents are generated from a language dictionary, based on the standard language. SOLUTION: The synthesizer has a text processing means 2, which outputs phoneme string information and rhythm information from the text data, a language dictionary 1 which is referred to by the means 2, a rhythm conversion means 4 which converts the rhythm information, based on a conversion rule and a ruled voice synthesizing means 3 which synthesizes voice waveforms from the phoneme string information and converted rhythm information. The means 4 has an accent conversion rule storage section 401, an accent conversion section 402, a phoneme string conversion rule storage section 403 and a phoneme string conversion section 404. The section 402 converts accent information into accent type information, that is different from input information, in accordance with the conversion rule stored in the section 401. The section 404 converts phoneme string information into a phoneme string that differs from the input information, in accordance with the conversion rule stored in the section 403.

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a speech synthesizer, a speech synthesis method, and a speech synthesis program for inputting text data consisting of a character string or a symbol string and converting it into speech data.
[0002]
[Prior art]
In speech synthesis technology, not only accurately convey information, but also customize according to user's preference, switch voice according to the situation, or let multiple characters talk in entertainment such as games Therefore, it is required to synthesize a plurality of individual voices.
[0003]
The main factors that determine speech are: voice quality (thick, thin, husky, etc.), overall (average) pitch, vocalization speed, pause (between phrases (length)) ), Accents (words, phrases, etc., pitches of sounds within short units), phrase pitch components (pitches of the entire sentence (phrase component strength), generally higher at first, gradually lowering), and the like. Of these, parts other than voice quality are generally called prosody. Among prosody, accent has a great influence on voice personality along with voice quality.
[0004]
In the conventional method of synthesizing speech (JP-A-1-3214496416 and JP-A-9-244777), as shown in FIG. 7, the text processing means 2 is used to analyze a sentence in a language dictionary 1 referred to. In addition to the data of the morpheme information storage unit 101 and the standard word accent storage unit 102 for providing standard accents, a dialect accent storage unit 103 is provided and accent data is selected from the dialect accent storage unit 103 to synthesize dialect speech. Was.
[0005]
[Problems to be solved by the invention]
However, the above-described conventional speech synthesis method has not been realized due to difficulty in actually creating an accent dictionary and development cost. The reason will be described below.
[0006]
In a conventional Japanese speech synthesizer, a language dictionary that determines an accent for each word is generally created based on a published standard word accent dictionary (for example, NHK “Japanese Pronunciation Accent Dictionary”). Synthesize speech with standard word accents. In order to utter correctly with accents other than standard words such as dialects, it is necessary to create an accent dictionary with tens of thousands of words.
[0007]
However, even with dialects, the actual dialects differ depending on the region and speaker, and it is very difficult to identify and organize dialects into one. For this reason, to create an accurate dialect accent dictionary, even if it is limited to one region such as Osaka dialect, it takes enormous time and cost, and is difficult to realize. Even if a dictionary can be created, adding dialect accents to a large number of words leads to an increase in memory, and further increasing the number of dialects also leads to an increase in memory, which is difficult to realize. Furthermore, it is considered that it is one of the reasons why it is difficult to evaluate the effect even if the cost is created.
[0008]
For example, in the “Osaka / Tokyo Accent Speech Dictionary: Miyoko Sugitan, 1995”, even if the speaker is from the same city of Osaka, the accent on each word is different. It shows that there is only such a tendency. On the contrary, even if the dialect accent of a certain region is accurately reproduced, since the evaluation differs depending on the listener, very few users may rate it as correct.
[0009]
Furthermore, even if only some of the words are examined for dialect accents and added to the dictionary, if other words are spoken with standard word accents, the dialect style changes rapidly during the utterance. There is a risk that the user will be confused and that the meaning of the sentence will not be understood.
[0010]
Specific accent-shaped shapes will be described later, but generally speaking, dialect accents cannot be reproduced by using only the 0-type to n-type accent types in the conventional standard word accent. Need no new accent type. According to the “Osaka / Tokyo Accent Dictionary” mentioned above as an example, the Osaka dial accent is different from the standard words 0 type to n type, from high type 0 type to n type and low type 0 type to n type. It cannot be expressed without using the accent type. On the other hand, since these accent types are common to Kansai dialects, if these accent types are used, a certain degree of Kansai dialect can be expressed even if the accent type is different from the dialect of a specific region.
[0011]
The present invention has been made to solve such problems of the prior art, and the dialect accent dictionary is incomplete, or there is no dialect accent dictionary, and it seems to be a dialect even from a language dictionary based on standard words. It is an object of the present invention to provide a speech synthesizer, a speech synthesis method, and a speech synthesis program that can generate unique accents.
[0012]
[Means for Solving the Problems]
The speech synthesizer of the present invention A language dictionary storing morpheme information for each word and accent information of standard words for each word; Input text data Refer to the language dictionary Analyze For each phrase, Phoneme sequence information and Contains an accent type that indicates the location of the accent core of a phrase in a standard word Prosodic information When A text processing means for outputting the prosody, a prosody conversion means for converting the prosodic information input from the text processing means into different prosody information based on a conversion rule, and synthesizing a speech waveform from the phoneme string information and the converted prosodic information A regular speech synthesis means And the prosody conversion means includes an accent conversion rule storage unit and an accent conversion unit. The accent conversion rule storage unit includes a high-frequency characteristic characteristic of each dialect from the accent type in the standard word. A conversion rule for conversion to an accent type or a low-rise accent type is stored, and the accent conversion unit selects the conversion rule stored in the accent conversion rule storage unit based on an inputted dialect selection parameter Then, based on the conversion rule, for all the clauses in the input text data, from the accent type in the standard word, the high-rise accent type or low-rise accent characteristic of the dialect corresponding to the dialect selection parameter Convert to type And the above-mentioned object is achieved.
[0014]
The prosody conversion means further includes a phoneme sequence conversion rule storage unit and a phoneme sequence conversion unit, and the phoneme sequence conversion unit stores the phoneme sequence conversion rule storage unit based on the input rule selection parameter. Selected conversion rule, and according to the selected conversion rule, Input from the text processing means Convert phoneme sequence information to phoneme sequence different from the input phoneme sequence It may be characterized by .
[0016]
The prosody conversion means includes the The output of the text processing means Bypass the accent conversion unit and the phoneme string conversion unit To output Has a bypass switch, The language dictionary includes Some words, dialects Accent type is stored in advance , The accent type of the dialect is stored for words included in the input text data in case of, The text processing means is Of dialects maintained in the language dictionary The prosody conversion unit may select and output an accent type, and the output of the text processing unit may bypass the accent conversion unit and the phoneme sequence conversion unit by the bypass switch. .
[0017]
The speech synthesis method of the present invention The speech synthesizer A speech synthesis method using the prosody conversion means. By the accent conversion unit, Based on dialect selection parameters entered The conversion rule stored in the accent conversion rule storage unit is selected, and from the accent type in the standard word to the dialect selection parameter for all the clauses in the input text data based on the conversion rule It is characterized in that it is converted to a high-rise accent type or low-rise accent type characteristic of the corresponding dialect .
[0018]
The prosody conversion means includes a bypass switch that outputs the text processing means by bypassing the accent conversion unit and the phoneme string conversion unit, and the language dictionary includes: Some words, dialects Accent type is stored in advance , The accent type of the dialect is stored for words included in the input text data in case of, The text processing means is Of dialects maintained in the language dictionary The prosody conversion unit may select and output an accent type, and the output of the text processing unit may bypass the accent conversion unit and the phoneme sequence conversion unit by the bypass switch. .
[0019]
The speech synthesis program of the present invention A method for executing a speech synthesis method using the speech synthesizer A speech synthesis program comprising the prosody conversion means. By the accent conversion unit, Based on dialect selection parameters entered The conversion rule stored in the accent conversion rule storage unit is selected, and from the accent type in the standard word to the dialect selection parameter for all the clauses in the input text data based on the conversion rule Perform conversion to the high or low accent type that is characteristic of the corresponding dialect That Characterize .
[0020]
The prosody conversion means includes a bypass switch that outputs the text processing means by bypassing the accent conversion unit and the phoneme string conversion unit, and the language dictionary includes: Some words, dialects Accent type is stored in advance , The accent type of the dialect is stored for words included in the input text data in case of, The text processing means is Of dialects maintained in the language dictionary The prosody conversion unit may select and output an accent type, and the output of the text processing unit may bypass the accent conversion unit and the phoneme sequence conversion unit by the bypass switch. .
[0021]
The operation of the present invention will be described below.
[0022]
In the present invention, an accent type (or phoneme string) analyzed based on a standard word can be converted based on a conversion rule, so that it is not necessary to create a dialect accent dictionary, and thus a conventional development cost is required. Speech synthesis with different dialect accents, which is too difficult to realize, is possible. Furthermore, since there is almost no increase in memory, it is possible to perform various speech synthesis such as a plurality of dialects by one apparatus.
[0023]
Further, in the accent conversion unit, according to the conversion rules, from the integer type accents of standard words 0 type to n type, the high accent type 0 type characteristic of Kansai dialect to the n type or the low accent type 0 type By converting from to n-type, even if the text processing means outputs prosodic information based on the language dictionary of standard words, it is possible to synthesize speech with Kansai dialect accents.
[0024]
Furthermore, when dialect prosody (accent) is registered for some words in a language dictionary based on standard words, speech synthesis is performed using the exact prosody registered in the dictionary for those words. The words that have not been investigated are converted into local dialect-like accents (or phoneme strings) using conversion rules. Thereby, it becomes possible to approach a more natural dialect, and it is possible to avoid the confusion of the voice listener due to the mixture of the dialect accent and the standard word accent.
[0025]
The algorithm for synthesizing speech is complicated and is generally realized by software using a general-purpose microcomputer or DSP. According to the speech synthesis program of the present invention, it is possible to add a dialect-like speech synthesis function in a system using a microcomputer such as a personal computer or a system using a DSP.
[0026]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments of the present invention will be described below with reference to the drawings.
[0027]
(Embodiment 1)
FIG. 1 is a diagram for explaining the configuration of a speech synthesizer according to an embodiment of the present invention. This speech synthesizer comprises a language dictionary 1, text processing means 2, recorded speech synthesis means 3, and prosody conversion means 4.
[0028]
The language dictionary 1 stores morpheme information for each word and accent information for each word, and is generally realized by a read-only memory.
[0029]
The text processing means 2 detects words using the language dictionary 1 from the input text data, determines the sentence structure from morphemes such as verbs and nouns, determines the length of the pose based on the sentence structure, Determines the strength of the phrase component that controls the overall inflection. Also, the accent type of each phrase is determined from the accent information of the independent words included in the phrase. Then, a phoneme symbol string, accent type, phrase component strength, pause length, and voice volume information are output for each phrase. Generally speaking, the loudness of a voice is often fixed within a sentence, but when viewed microscopically, a speech segment (basically, this is combined to synthesize speech) is recorded. The change in the loudness of the voice is output as it is. In prosodic control, the volume of this voice can be increased or decreased for each sentence. In addition, the utterance speed can be controlled by statistically examining the time length of each phoneme and determining how fast or slow the time length is. Furthermore, the pitch of the voice can be controlled by adding the phrase component obtained from the sentence structure and the accent component determined by each word. It Overall It is possible to adjust the strength of the inflection by adjusting the pitch of the voice by raising or lowering the pitch and by adjusting the strength of the phrase component or the accent component. It is often the case that the user can decide how much adjustment is made. This text processing means 2 is generally realized by software using a CPU or DSP.
[0030]
The regular speech synthesizer 3 synthesizes a speech waveform based on the input information. Specifically, the accent generation unit 302 refers to the accent shape storage unit 301 and sets the pitch of each phoneme in the input phoneme symbol string of each phrase. Next, the phrase generation unit 304 refers to the phrase shape storage unit 303, and for each phoneme of the phoneme symbol string spanning a plurality of clauses, converts the pitch shape representing the inflection with the strength of the phrase component. Adjust and add. The phrase shape will be described later. Next, the waveform generation unit 306 refers to the waveform dictionary 305 for each phoneme and synthesizes a speech waveform corresponding to the pitch of the sound generated by the accent generation unit 302 and the phrase generation unit 304. This regular speech synthesizer 3 is generally realized by software using a CPU or DSP.
[0031]
Various methods can be used as a method of waveform generation. Here, a waveform superposition method is used in which zero-phase unit waveforms are superimposed at intervals (pitch) expressed by the reciprocal of the frequency of the voice to be synthesized. By using it, it is possible to synthesize relatively high-quality speech.
[0032]
FIG. 2 shows an example of the phrase shape. The phrase shape is a model of the shape of the voice pitch of the whole sentence when a human utters a sentence, and has a shape that increases at the beginning of each phrase and gently descends toward the end of the sentence. In addition, there is a case where a “he” -shaped curve generally referred to as a Fujisaki model is used. FIG. 2 shows a phrase shape that descends linearly on the logarithmic frequency. In this example, the pitch of the voice at the beginning of the phrase is given as the value of the phrase component, and the descending speed is adjusted so that the steady height is reached at the end of the phrase. Therefore, as the phrase becomes longer, the descending speed becomes slower.
[0033]
The above is the same function as a normal speech synthesizer. In this embodiment, the prosody conversion means 4 is further provided to change the prosody of the synthesized speech. The prosody generally refers to accents, phrase component strength, average voice pitch, utterance speed, pause length, and the like. For example, dialect-like speech synthesis can be performed by converting the accent type. It is also possible to perform operations such as synthesizing exaggerated speech in a speech style by gradually increasing the accent strength, phrase strength, and voice volume toward the end of the sentence. Furthermore, by converting not only prosody but also phonemes, it is possible to convert dialect and age specific words. This prosody conversion means 4 is generally realized by software using a CPU or DSP.
[0034]
In the following, the prosody conversion means 4 will be described in more detail with reference to FIG. The prosody conversion means 4 includes an accent conversion rule storage unit 401 and an accent conversion unit 402. The accent conversion unit 402 selects an accent conversion rule from the conversion rules stored in the accent conversion rule storage unit 401 based on the dialect selection parameter. The dialect selection parameter can be embedded in the input text using a special symbol as a control command, or can be input by the user by some method (for example, providing a dialect selection parameter input unit). Normally, software can be used to set a male voice, female voice, voice pitch, loudness, speed, etc. using a setting screen, and a dialect can be set on the same screen. Alternatively, a setting device can be provided separately. Then, the input phoneme string and accent type are converted into different accent types according to the selected accent conversion rule. Specific accent type conversion will be described later.
[0035]
Furthermore, a phoneme string conversion rule storage unit 403 and a phoneme string conversion unit 404 may be provided inside the prosody conversion means 4. In this case, similar to the accent, a phoneme string conversion rule is selected from the conversion rules stored in the phoneme string conversion rule storage unit 403 based on the rule selection parameter. Similarly to the dialect selection parameter, the rule selection parameter can be embedded in the input text using a special symbol as a control command, or can be input by a user (for example, providing a dialect selection parameter input unit). it can. Then, the input phoneme string and accent type are converted into different phoneme strings according to the selected phoneme string conversion rule.
[0036]
First, a specific example of the phoneme string conversion rule will be described. For example, as an example of rules for converting to childish voice,
(1) Convert “shi” to “chi”
(2) Convert “su” and “tsu” to “chu”
(3) Convert “ra” to “da”
(4) Convert “zo” to “do”
Etc.
[0037]
In addition to this, the rules that replace “hi” and “shi” in some dialects, the rules that “shi” in other dialects are pronounced “su”, etc. are memorized. Choice Various conversions can be made by selecting using parameters. How to choose these transformation rules, so do not change for each clause can be realized by setting the rule selection parameters to apply in advance which conversion rules prior to speech synthesis.
[0038]
Next, a specific conversion method to Japanese dialect accent conversion will be described. FIG. 3 is a diagram showing a specific accent type used in the present embodiment. FIG. 3A shows the standard word accent shape used in conventional speech synthesis for each accent type and number of mora. Here, mora is a unit that represents the beat of a voice, and corresponds to kana characters in Japanese. In general, an accent is expressed in two levels, high and low, for each mora. Type 0 is also referred to as a flat plate type, and starts with a low tone “L” except for 1 mora, followed by a high tone “H” after the second mora. In addition, what is shown by () has shown the pitch of the adjunct word when an adjunct follows a word. For example, “I” is 3 mora and 0 type in the standard language, so “LHH”. If the accompanying word “ha” is followed by “I am”, it is “LHHH”. Type 1 has a high first mora, and the following mora is lined with bass. In the two or more types, the accent type number indicates the position of the accent nucleus, the second mora to the accent nucleus is high, and the back from the mora next to the accent nucleus is low. For example, “Haneashi” becomes “LLH” because it is 3 types with 3 mora, and it is the same as 0 type by itself. However, when the attached word “ha” is followed by “Haneashiha”, “LHHL” It becomes.
[0039]
FIG. 3B and FIG. 3C are diagrams showing examples of a high rise accent type and a low rise accent type that are characteristic of the Kansai dialect, respectively. In the high-pitched accent type shown in FIG. 3B, the beginning mora always starts with a high tone “H”. Further, when the accent type is 1 and 1 mora, the treble is lowered from the high sound within the mora, and the duration is increased. This is represented as “D” in FIG. In the low rise accent type of FIG. 3C, in the case of the 0 type, only the last mora has a high tone. Type 1 is the same as the standard type and high-rise accent type except for 1 mora. Further, when the accent type is 1 type and 1 mora, the sound rises from low to high within the mora, and the duration is long. This is represented by “U} in FIG. 3C. In the case of type 2 or higher, it starts with a low tone, and only the position of the accent nucleus becomes a high tone, and thereafter it becomes a low tone.
[0040]
In the present embodiment, these accent shapes are stored in the accent shape storage unit 301 in the regular speech synthesizer 3, and the accent type is selected from these in the accent conversion unit 402 in the prosody conversion unit 4. Conventionally, an accent type can be specified by an integer value from 0 type to n type, but in this embodiment, an integer value different from this is used to specify a standard type, a high rise type, or a low rise type. Specifies the accent type of type. Therefore, the phoneme symbol string, accent type, phrase component strength, pose length, voice volume, and accent type information are output from the prosody conversion means 4 for each phrase.
[0041]
When an accent type other than Kansai dialect is realized, if it cannot be expressed in the above-described type, it is possible to further expand the accent type and add an accent type. Further, by selecting an accent conversion rule based on a dialect selection parameter, it is possible to synthesize voice by switching a plurality of dialect accent types with a single device. Dialect selection parameters can be set using a setting screen. For example, an integer, 0 is a standard language, 1 is a Kansai dialect, 2 is a Kagoshima dialect, and the like.
[0042]
Next, a specific example of an accent conversion rule for synthesizing Kansai dialect speech will be shown. For example,
(1) 2-mora clauses convert 0 type to 1 type and 1 type to 0 type
(2) If the first vowel differs from the second vowel in a type 1 phrase of 3 mora or more, convert it to type 2
(3) Make the type 2 accent low
(4) Other than (3) above, if (accent type ≧ number of mora)
(5) Other than the above (4), if it is 2 mora, convert it to low-rise type
(6) Other than the above (5), if the number of mora is an odd number, it is converted to a low rise type, and if it is an even number, it is converted to a high rise type.
Such a set of rules is stored in the accent conversion rule storage unit 401 as a Kansai dialect accent conversion rule. In this conversion rule (3), the type 2 accent has the same shape regardless of whether it is a low-rise type or a standard type.
[0043]
An example of conversion using this rule is shown using an example sentence. For example,
"Play with video games and computers"
Assume that the example sentence is input to the text processing means 2. By text processing means, "Play video games, play on PCs, play games, etc."
Suppose that it is delimited by the phrase Here, the phrase break is represented by a midpoint (•). At this time, the standard word accent type of each phrase is
"4 ・ 0 ・ 1 ・ 0 ・ 0"
It becomes. If this accent is expressed by the pitch of the sound (H: treble, L: bass),
"LHHHLLL / LHHHH / HLLL / LH / LHH"
become that way. According to the above conversion rules, the accent type of each phrase is
"Low rise 4, low rise 0, high rise 1, high rise 1, low rise 3"
Is converted as follows. If this accent is expressed in terms of pitch,
"LLLLHLLL / LLLLH / HLLL / HL / LLH"
become that way. As a result, it can be seen that the rules can be used to convert the accents to those often found in Kansai dialect.
[0044]
In the present embodiment, by performing conversion according to the conversion rule, a dialect accent dictionary is not required, so that variations of the reading speech can be expanded at low cost.
[0045]
(Embodiment 2)
FIG. 4 is a diagram for explaining the configuration of the speech synthesizer according to the second embodiment. This speech synthesizer has a bypass switch 405 in the prosody conversion means 4 in addition to the configuration of FIG. When the text processing unit 2 instructs non-conversion, the text is output without performing accent conversion by the accent conversion unit 402 and phoneme sequence conversion by the phoneme sequence conversion unit 404. This function is effective when the language dictionary 1 has a dialect accent storage unit 103 as shown in FIG. 7 and stores correct accents only for some important words and frequent words. .
[0046]
In this configuration, it is desirable to input the same parameters to the text processing means 2 and the accent conversion unit 402 as dialect selection parameters to unify the accents.
[0047]
For example, when the dialect accent specified by the dialect selection parameter is present in the language dictionary 1 for the word to be output with the dialect accent in the text processing means 2, 0 is output as the prosodic conversion instruction code, When the designated dialect accent does not exist in the language dictionary 1, 1 is output as the prosodic conversion instruction code.
[0048]
The bypass switch 405 in the prosody conversion means 4 outputs information to the accent conversion unit 402 and the phoneme sequence conversion unit 404 when the prosody change instruction code is 1, and does not convert when the prosody conversion instruction code is 0. The information is output to the regular speech synthesis means 3 as it is.
[0049]
In the present embodiment, the output from the text processing means 2 for each phrase and the output from the prosody conversion means 4 for each phrase are the phoneme symbol string, accent type, phrase component strength, pause length, and voice volume. And accent type information. Further, the prosody conversion means 4 also outputs information on the prosody conversion instruction code. Note that the accent type is output from the text processing unit 2 and the prosody conversion unit 4, and when the conversion is not instructed by the prosody conversion instruction code, the accent type from the text processing unit has priority. When conversion is instructed by the prosodic conversion instruction code, the accent type from the text processing unit is ignored, and the accent type is determined by the prosody conversion unit.
[0050]
The operation of the bypass switch 405 may be performed by the prosody conversion unit 4 in accordance with the prosody conversion instruction code instructed from the text processing unit 2, or may be performed by the accent conversion unit 402 itself. Further, even when speech synthesis is performed using standard accents, accent conversion and phoneme string conversion are not performed using the bypass switch 405.
[0051]
(Embodiment 3)
FIG. 5 is a diagram showing a configuration example for realizing the speech synthesizer of the present invention using a computer. The input device 5 inputs text data to be synthesized by serial communication, network communication, or a keyboard.
[0052]
The storage medium 8 is a CDROM, floppy disk, flash memory, or the like in which a speech synthesis program comprising the functions of the language dictionary 1, the text processing unit 2, the regular speech synthesis unit 3 and the prosody conversion unit 4 shown in the first embodiment is recorded. It is a recording medium. The storage device 9 is a storage device such as a hard disk or a flash memory that has read the program of the storage medium 8. The RAM 10 is used for primary storage necessary for processing.
[0053]
The output device 7 includes a D / A converter, an amplifier, a speaker, and the like for outputting synthesized voice.
[0054]
The processing device 6 is a processing device such as a microcomputer or a DSP that performs speech synthesis processing in accordance with a speech synthesis program read into the storage medium 9 or the storage device 10.
[0055]
FIG. 6 shows a flowchart when the Kansai dialect accent conversion rule is realized by an algorithm. This algorithm itself performs the same processing operation as in the first embodiment. Steps 1101 to 1107 are processes for converting the position of the accent nucleus, and steps 1108 to 1114 are processes for selecting a high-rise type or a low-rise type. Steps 1101 to 1105 correspond to the conversion rule (1), and steps 1106 and 1107 correspond to the conversion rule (2). Steps 1108 and 1109 correspond to the conversion rule (3), and steps 1110 and 1111 correspond to the conversion rule (4). Steps 1112 to 1114 correspond to the conversion rules (5) and (6).
[0056]
The second embodiment can also be realized by the same configuration as that in FIG.
[0057]
【The invention's effect】
As described above in detail, according to the present invention, since the accent type analyzed and given based on the standard word accent is converted based on the conversion rule, it is not necessary to create a special accent dictionary. Speech synthesis with dialect accents that were difficult to implement due to excessive development costs is possible. Furthermore, since there is almost no increase in memory, a variety of speech synthesis such as a plurality of dialects and accents with personal characteristics can be performed with one apparatus.
[0058]
Also, Kansai dialect speech can be synthesized by using a conversion rule for converting from a standard word accent to a high-rise accent type or low-rise accent type characteristic of Kansai dialect.
[0059]
Furthermore, even when dialect accents are registered in the language dictionary in some words, it is possible to prevent the voice listener from being confused by mixing dialect accents and standard word accents.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating a configuration of a speech synthesizer according to a first embodiment.
FIG. 2 is a diagram illustrating an example of a phrase shape.
FIGS. 3A to 3C are diagrams illustrating an example of an accent type according to the first embodiment.
FIG. 4 is a diagram illustrating a configuration of a speech synthesizer according to a second embodiment.
FIG. 5 is a diagram illustrating a configuration for realizing the speech synthesizer according to the third embodiment by using a computer.
FIG. 6 is a flowchart for realizing an accent conversion rule by an algorithm.
FIG. 7 is a diagram illustrating a configuration of a conventional speech synthesizer.
[Explanation of symbols]
1 language dictionary
2 Text processing means
3. Regular speech synthesis means
4 Prosody conversion means
5 input devices
6 processing equipment
7 Output device
8 storage media
9 Storage device
10 RAM
101 Morphological information storage unit
102 Standard word accent storage
103 Dialect accent storage
301 Accent shape storage
302 Accent generator
303 Phrase shape storage unit
304 Phrase generator
305 Waveform Dictionary
306 Waveform generator
401 Accent conversion rule storage unit
402 Accent converter
403 phoneme string conversion rule storage unit
404 Phoneme sequence converter
405 Bypass switch

Claims (7)

A language dictionary storing morpheme information for each word and accent information of standard words for each word;
The input text data is analyzed by reference to the language dictionary, for each clause, and text processing means for outputting a prosodic information including accent type indicating the position of the accent nucleus clauses in the phoneme string information and Mandarin ,
Prosodic conversion means for converting prosodic information input from the text processing means into different prosodic information based on conversion rules;
Regular speech synthesis means for synthesizing a speech waveform from the phoneme sequence information and the converted prosodic information ,
The prosody conversion means includes an accent conversion rule storage unit and an accent conversion unit,
The accent conversion rule storage unit stores conversion rules for converting from the accent type in the standard language to the high-rise accent type or low-rise accent type characteristic of each dialect,
In the accent conversion unit, the conversion rule stored in the accent conversion rule storage unit is selected based on the input dialect selection parameter, and all the clauses in the input text data are selected based on the conversion rule. Then, the speech synthesizer performs conversion from the accent type in the standard language into a high-rise accent type or a low-rise accent type characteristic of the dialect corresponding to the dialect selection parameter . The prosody conversion means further includes a phoneme sequence conversion rule storage unit and a phoneme sequence conversion unit,
The phoneme string conversion unit selects a conversion rule stored in the phoneme string conversion rule storage unit based on the input rule selection parameter, and the phoneme string input from the text processing unit according to the selected conversion rule The speech synthesizer according to claim 1 , wherein the information is converted into a phoneme string different from the input phoneme string. The prosody conversion means includes a bypass switch that outputs the output of the text processing means by bypassing the accent conversion unit and the phoneme string conversion unit,
In the language dictionary , dialect accent types are stored in advance for some words,
When the dialect accent type is stored for the word included in the input text data, the text processing means selects and outputs the dialect accent type held in the language dictionary. The speech synthesizer according to claim 2 , wherein the prosody conversion unit causes the accent conversion unit and the phoneme sequence conversion unit to bypass the output of the text processing unit by the bypass switch . A method for performing speech synthesis using the speech synthesizer according to claim 1 ,
The accent conversion unit of the prosody conversion unit selects the conversion rule stored in the accent conversion rule storage unit based on the dialect selection parameter input, and the text data input based on the conversion rule A speech synthesis method comprising: converting all accents in a standard word from the accent type in a standard word into a high-rise accent type or a low-rise accent type characteristic of a dialect corresponding to the dialect selection parameter . The prosody conversion means includes a bypass switch that outputs the output of the text processing means by bypassing the accent conversion unit and the phoneme string conversion unit,
In the language dictionary , dialect accent types are stored in advance for some words,
When the dialect accent type is stored for the word included in the input text data, the text processing means selects and outputs the dialect accent type held in the language dictionary. The speech synthesis method according to claim 4 , wherein the prosody conversion means bypasses the accent conversion unit and the phoneme string conversion unit by using the bypass switch . A speech synthesis program for executing a method of performing speech synthesis using the speech synthesizer according to claim 1 ,
The accent conversion unit of the prosody conversion unit selects the conversion rule stored in the accent conversion rule storage unit based on the dialect selection parameter input, and the text data input based on the conversion rule A speech synthesis program for performing conversion from the accent type in a standard word to a high-rise accent type or a low-rise accent type characteristic of a dialect corresponding to the dialect selection parameter . The prosody conversion means includes a bypass switch that outputs the output of the text processing means by bypassing the accent conversion unit and the phoneme string conversion unit,
In the language dictionary , dialect accent types are stored in advance for some words,
When the dialect accent type is stored for the word included in the input text data, the text processing means selects and outputs the dialect accent type held in the language dictionary. The speech synthesis program according to claim 6 , wherein the prosody conversion unit causes the accent conversion unit and the phoneme sequence conversion unit to bypass the output of the text processing unit by the bypass switch .

Images