JP3313310B2 - Speech synthesis apparatus and synthesis method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device and a method of voice compositing which allows a small capacity of memory for storing a data base by maintaining a high- quality sound of composited voices even when a small number of voice elements are stored in the data base. SOLUTION: Inputted characters are converted into phonetic notation (means 101). The rhythms of synthesized voice are calculated by the converted phonetic notation (means 102), and based on the converted phonetic notation, voice elements including at least one of two syllables in the heading of word which sharpen human hearing sense, the voice elements of high-pitch frequencies upto the accent kernel, the voice elements including the endings of words and other voice elements are selected (means 103) from a memory 104. The variations of the rhythms of these voice elements are executed (means 105) in the conditions that the fine structures of the rhythm at the time of recording voice elements are left with respect to the voice elements including at least one of the two heating syllables of words, high-pitch frequencies up to the accent kernel and endings of words and the variation of the rhythm of the other voice elements is executed by making the use of the fine structures of the rhythms at the time of recording or using calculated rhythms.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizing apparatus and a synthesizing method for converting input characters based on rules and synthesizing them so that they can be converted into speech.

[0002]

2. Description of the Related Art Conventionally, as this kind of speech synthesizer,
There is an apparatus disclosed in Japanese Patent Application Laid-Open No. 6-95592 "Speech synthesis apparatus". Hereinafter, a conventional speech synthesizer will be described with reference to FIG. FIG. 10 is a block diagram showing the configuration of the above-mentioned conventional speech synthesizer.

In FIG. 10, reference numeral 501 denotes an input unit for inputting a character string and its prosody information and converting it into a phonetic notation.
02 is a prosody parameter generation unit that generates prosody parameters for the phonetic notation from the input unit 501 based on the prosody rule dictionary 503, and 504 is a speech unit set selection unit 506 from a large-capacity speech unit database 505 containing many speech units.
A voice unit selection unit for selecting a voice unit from the selected voice unit set 507 using the prosodic parameters so that the spectrum change at the connection point of the voice unit is smooth, and a voice unit selection unit 508 connects the selected voice units to generate voice. An audio synthesizer 509 is an output unit that outputs synthesized audio.

Next, the operation of the above-mentioned conventional speech synthesizer will be described with reference to FIG. First, an input character string is converted into a phonetic notation in an input unit 501, and a prosody parameter generation unit 502 generates a prosody parameter based on a prosody rule dictionary 503. The speech unit selection unit uses a prosodic parameter from the speech unit set 507 selected by the speech unit set selection unit 506 from a large-capacity speech unit database 505 containing a large number of speech units (hereinafter referred to as speech units). A speech unit is selected such that the spectrum change at the connection point of the segments becomes smooth. The selected voice unit is connected by the voice synthesizer 508 and synthesized as continuous voice, and output from the output unit 509. In this way, a conventional voice synthesizer can provide a high-quality synthesized voice.

[0005]

However, in the above-mentioned conventional speech synthesizer, in order to secure a speech unit in which a spectrum change at a connection point of the speech unit becomes smooth, various speech spectrums are required. There is a problem that a large number of speech units must be prepared, and a large-capacity memory must be prepared to store such a large number of speech units.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned conventional problems, and is intended to maintain a high quality of synthesized speech even if the number of speech units stored in the database is small, so that the speech unit is improved. An object of the present invention is to make a memory for storing a database small.

[0007]

A speech synthesizing apparatus and a synthesizing method according to the present invention convert an input character string and prosody information into a phonetic notation, calculate a prosody of the synthesized speech based on the converted phonetic notation, and convert the speech. A speech unit containing at least one of the first two syllables, a speech unit having a high pitch frequency up to the accent nucleus, and a speech unit containing the ending, or a speech unit containing at least one of the two syllables at which the human auditory sense is sharpened from the memory based on the phonetic transcription Of the speech unit, and the prosodic transformation of these speech units is such that a speech unit including at least one of the first two syllables, which is a part where human hearing is sharp, and a high pitch frequency speech up to the accent nucleus. For the speech unit including the unit and the ending, it is performed while keeping the fine structure of the prosody at the time of recording the speech unit,
The modification of the prosody of the speech unit for other uses the fine structure of the prosody at the time of recording or uses the calculated prosody.

According to the present invention, a large amount of storage capacity is required for two syllables at the beginning, particularly those having a high pitch frequency up to the accent nucleus, or deformation of the prosody of the speech unit at the end of the word, which are particularly important for maintaining the sound quality of the synthesized speech. Use the fine structure of the prosody at the time of voice unit recording, and use the fine structure of the prosody at the time of the voice unit recording to transform the prosody of the other voice unit that does not significantly affect the sound quality of the synthesized speech. Or, by performing calculations, speech synthesis can reduce the number of speech units stored in the database and reduce the capacity of the memory for storing the speech unit database, while maintaining high quality of synthesized speech. An apparatus and a synthesis method are obtained.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The invention according to claim 1 of the present invention comprises input character string conversion means for converting an input character string mixed with kanji kana or a reading kana and prosodic information into phonetic notation, and Prosody calculation means for calculating the prosody of the synthesized speech using the phonetic notation, a speech unit containing at least one of the first two syllables, a speech unit having a high pitch frequency up to the accent nucleus, and a speech unit containing the ending, and these speeches Other than a fragment
A memory that stores at least the speech unit, the selection while remaining the speech unit selection means for selecting a speech unit corresponding from the memory, the prosody of the microstructure during the speech unit recorded in accordance with the phonetic Speech unit transformation means for transforming the prosody of the transformed speech unit, speech unit connection means for connecting the transformed speech unit, and synthesized speech output means for outputting the connected synthesized speech. The first two syllables, which are particularly important for maintaining the sound quality of synthesized speech,
Deformation of the prosody of a pitch unit with a high pitch frequency up to the accent nucleus, each speech unit at the end, or a speech unit for other uses, changes the fine structure of the prosody when recording speech units that does not require a large amount of storage capacity. By using this method, it is possible to reduce the number of speech units stored in the database and to reduce the capacity of the memory for storing the speech units while maintaining high sound quality of the synthesized speech.

According to a second aspect of the present invention, there is provided an input character string converting means for converting an input kanji kana mixed character string or a reading kana and prosody information into a phonetic notation, and using the converted phonetic notation. A prosody calculating means for calculating the prosody of the synthesized speech, a speech unit including at least one of the first two syllables, a speech unit having a high pitch frequency up to the accent nucleus, a speech unit including the ending and a speech unit other than these speech units A memory for storing at least a speech unit, a speech unit selection means for selecting a corresponding speech unit from the memory according to the phonetic notation, and the selected beginning 2 while retaining the fine structure of the prosody at the time of recording. syllables perform prosody modification of speech units including speech segment and ending high pitch frequencies up speech unit and accent core containing at least one, these sounds
Deformation of a prosody of a voice unit other than a voice unit is performed using the calculated prosody without using the fine structure of the prosody at the time of the voice unit recording, and Speech unit connection means to be connected, and synthesized speech output means for outputting the connected synthesized speech,
The first two syllables that are particularly important for maintaining the sound quality of synthesized speech, those with a high pitch frequency up to the accent nucleus, or the deformation of the prosody of the speech unit at the end of the word, can be used when recording speech units that do not require a large amount of storage capacity. By using the fine structure of the prosody, the prosody deformation of the other speech unit that does not significantly affect the sound quality of the synthesized speech is calculated without using the fine structure of the prosody at the time of recording In addition, while maintaining high quality of synthesized speech, the number of speech units stored in the database can be reduced, and the capacity of the memory for storing the speech segments can be reduced.

According to a third aspect of the present invention, there is provided an input character string converting means for converting an input character string mixed with kanji kana or a reading kana and prosodic information into a phonetic notation, and a converted phonetic notation. A prosody calculating means for calculating a prosody of the synthesized speech; a memory storing at least a speech unit including at least one of the first two syllables, a speech unit including the ending, and a speech unit other than these speech units ; Voice unit selection means for selecting a corresponding voice unit from the memory according to phonetic notation, and a voice unit for transforming the prosody of the selected voice unit while retaining the fine structure of the prosody at the time of voice unit recording One-side transforming means, a speech unit connecting means for connecting the transformed speech unit, and a synthesized speech output means for outputting the connected synthesized speech,
Modification of the prosody of two syllables at the beginning, each speech unit at the end, or a speech unit for other purposes, which is particularly important for maintaining the sound quality of synthesized speech, is a prosody at the time of speech unit recording that does not require a large amount of storage capacity. By using the microstructure of the above, it is possible to reduce the number of speech units stored in the database and the memory capacity for storing it while maintaining high sound quality of synthesized speech. Has an action.

According to a fourth aspect of the present invention, there is provided an input character string converting means for converting an input character string mixed with kanji kana or a reading kana and prosodic information into a phonetic notation, and using the converted phonetic notation. A prosody calculating means for calculating a prosody of the synthesized speech; a memory storing at least a speech unit including at least one of the first two syllables, a speech unit including the ending, and a speech unit other than these speech units ; A speech unit selecting means for selecting a corresponding speech unit from the memory according to the phonetic notation, and a speech unit and ending which include at least one of the selected two-syllables at the beginning while leaving the fine structure of the prosody at the time of recording. It performs a prosody of the deformation of the speech segment, including, this
The speech unit deformation means for performing the transformation of the prosody of the speech unit other than these speech units using the calculated prosody without using the fine structure of the prosody at the time of the speech unit recording; and A speech unit connecting means for connecting the speech units; and a synthesized speech output means for outputting the connected synthesized speech, wherein two initial syllables or endings particularly important for maintaining the sound quality of the synthesized speech. The prosody transformation of the speech unit uses the fine structure of the prosody at the time of speech unit recording that does not require a large amount of storage capacity, and the prosody of the other speech unit that does not significantly affect the sound quality of the synthesized speech. By using the one calculated without using the fine structure of the prosody at the time of recording,
While maintaining high sound quality of the synthesized speech, there is an effect that the number of speech units stored in the database can be reduced and the capacity of a memory for storing the speech segments can be reduced.

[0013] According to a fifth aspect of the present invention, a kanji kana-mixed character string or a reading kana and prosodic information are converted into phonetic notations, and based on the converted phonetic notations, the first two words are read from a memory. the phonetic from the speech unit and the sound <br/> Koemotohen other than those speech segments including speech unit and endings of high pitch frequencies up speech unit and accent core containing at least one syllable Select the corresponding speech unit, perform the prosody modification of the selected speech unit while leaving the fine structure of the prosody at the time of speech unit recording, connect the transformed speech unit as synthetic speech The first two syllables that are particularly important for maintaining the sound quality of synthesized speech, those with a high pitch frequency up to the accent nucleus,
By modifying the prosody of each speech unit at the end, or the speech unit for the rest, by using the fine structure of the prosody at the time of speech unit recording that does not require a large amount of storage capacity, While maintaining high sound quality of the synthesized speech, there is an effect that the number of speech units stored in the database can be reduced and the capacity of a memory for storing the speech segments can be reduced.

According to a sixth aspect of the present invention, a kanji kana mixed character string or a reading kana and prosody information are converted into phonetic notations, and the converted phonetic notation is used to convert the prosody of the synthesized speech. Calculated, based on the converted phonetic notation, a speech unit containing at least one of the first two syllables from the memory, a speech unit having a high pitch frequency up to the accent nucleus, a speech unit containing the ending, and other than these speech units A speech unit corresponding to the phonetic notation is selected from the speech units, and a speech including at least one of the selected two-syllables at the beginning while leaving the fine structure of the prosody at the time of recording the speech unit. performs a prosody of the deformation of the speech segment, including the speech segment and endings of high pitch frequency of up to fragment and the accent nucleus, these sounds
Modification of the prosody of the speech unit other than the voice unit is performed using the calculated prosody without using the fine structure of the prosody at the time of recording, and the transformed speech unit is connected and output as a synthesized speech. The first two syllables that are particularly important for maintaining the sound quality of the synthesized speech, those with a high pitch frequency up to the accent nucleus, or the prosodic transformation of the speech unit at the end do not require a large amount of storage capacity. Utilizes the fine structure of the prosody at the time of speech unit recording, and uses the one calculated without using the fine structure of the prosody at the time of recording to modify the prosody of the other speech unit that does not significantly affect the sound quality of the synthesized speech By doing so, it is possible to reduce the number of speech units stored in the database and to reduce the capacity of the memory for storing the speech units while maintaining high sound quality of the synthesized speech.

According to a seventh aspect of the present invention, a kanji kana mixed character string or a reading kana and prosody information are converted into phonetic notations, and based on the converted phonetic notations, the first two words are read from a memory. Speech units containing at least one of syllables, speech units containing endings, and other than these speech units
A speech unit corresponding to the phonetic transcription is selected from the speech units, and the prosody transformation of the selected speech unit is performed while leaving the fine structure of the prosody at the time of recording the speech unit. The modified speech units are connected to be output as a synthesized speech, and are particularly important for maintaining the sound quality of the synthesized speech. The prosodic transformation of speech units is performed using the fine structure of the prosody at the time of speech unit recording, which does not require a large amount of storage capacity. This has the effect that the number of speech units to be stored can be reduced and the capacity of the memory for storing the same can be reduced.

The invention according to claim 8 of the present invention converts the input kanji kana mixed character string or the reading kana and the prosody information into phonetic notation, and converts the prosody of the synthesized speech by the converted phonetic notation. Based on the converted phonetic notation, a speech unit including at least one of the first two syllables, a speech unit including the ending, and a speech unit including these words are calculated based on the converted phonetic notation.
A speech unit corresponding to the phonetic notation is selected from outside speech units and includes at least one of the selected two-syllables at the beginning while keeping the fine structure of the prosody at the time of recording the speech unit. It performs a prosody of the deformation of the speech segment, including the speech segment and endings, this
The transformation of the prosody of the speech unit other than these speech units is performed using the calculated prosody without using the fine structure of the prosody at the time of recording, and the transformed speech unit is connected as a synthesized speech. The prosodic transformation of the first two syllables or the ending speech unit, which is particularly important for maintaining the sound quality of the synthesized speech, is performed when the speech unit recording that does not require a large amount of storage capacity. By using the fine structure of, the deformation of the prosody of the other speech unit that does not significantly affect the sound quality of the synthesized speech uses the one calculated without using the fine structure of the prosody at the time of recording, While maintaining high sound quality of the synthesized speech, there is an effect that the number of speech units stored in the database can be reduced and the capacity of a memory for storing the speech segments can be reduced.

Hereinafter, based on the attached drawings, FIGS. 1 to 9,
An embodiment of the present invention will be described in detail. FIG. 1 is a block diagram illustrating a configuration of a speech synthesis device according to a first embodiment of the present invention, FIG. 2 is a block diagram illustrating a configuration of a speech synthesis device according to a second embodiment of the present invention, and FIG. FIG. 4 is a block diagram illustrating a configuration of a speech synthesizer according to a third embodiment of the present invention. FIG. 4 is a block diagram illustrating a configuration of a speech synthesizer according to a fourth embodiment of the present invention. FIG. 6 is a view for explaining how the synthesizer converts kanji character strings into phonetic notations when the input is performed. FIG. 6 is a flowchart showing a speech synthesis method according to a fifth embodiment of the present invention.
FIG. 7 is a flowchart showing a speech synthesis method according to the sixth embodiment of the present invention, FIG. 8 is a flowchart showing a speech synthesis method according to the seventh embodiment of the present invention, and FIG. 9 is an eighth embodiment of the present invention. 9 is a flowchart showing a speech synthesis method according to the embodiment.

First, with reference to FIG. 1, the configuration of a speech synthesizer according to a first embodiment of the present invention will be described. FIG.
101, input character string conversion means for inputting a character string mixed with kanji or kana or reading kana and prosodic information and converting them into phonetic transcription, and 102, a prosody calculation for calculating the prosody (pitch frequency) of the whole synthesized speech based on the phonetic transcription Means 104, a memory storing a database including a speech unit including at least one of the first two syllables, a speech unit up to an accent nucleus, a speech unit including a ending, and a speech unit for other use;
103 converts the phonetic notation output from the input character string converting means 101 into a speech unit notation in CV or VCV units,
This is a speech unit selection unit that searches the memory 104 and selects a corresponding speech unit.

Reference numeral 105 designates a speech unit including at least one of the selected two initial syllables, a speech unit having a high pitch frequency up to the accent nucleus, a speech unit including the ending, and a speech unit for the other. While retaining the fine structure of the prosody at the time of recording, the speech unit transforming means 106 for transforming the prosody (pitch frequency) according to the formula (1-1) or (1-2) (described later) is transformed. Speech unit connection means 107 for connecting speech units by averaging, and 107 a synthesized speech output unit for outputting a synthesized speech to which speech units are connected.

Next, the operation of the voice synthesizing apparatus according to the first embodiment of the present invention will be described with reference to FIG. The input character string conversion means 101 converts the input character string mixed with kanji kana or the reading kana and prosodic information into phonetic notation. An example of this conversion will be described later with reference to FIG. Next, the prosody (pitch frequency) of the entire synthesized speech is calculated by the prosody calculation means 102 based on the converted phonetic notation. The phonetic notation output from the input character string conversion means 101 is converted into a speech unit notation in units of CV or VCV by the speech unit selection means 103, and a speech unit including at least one of the two syllables at the beginning and an accent. A database in the memory 104 storing the speech unit up to the nucleus, the speech unit including the ending, and the speech unit for the other is searched, and the corresponding speech unit is selected.

Next, these speech segments are sent to speech segment transformation means 105, and a speech segment containing at least one of the two syllables at the beginning, a speech segment having a high pitch frequency up to the accent nucleus, and an ending are added. The speech unit to be included and the speech unit for the other are modeled for the synthesized speech in accordance with the formula (1-1) or (1-2) (described later) while keeping the fine structure of the prosody at the time of recording. The prosody (pitch frequency) is modified so that it matches the vowel start part and the vowel stationary part pitch frequency of the pitch frequency. These deformed speech units are connected by averaging in a range of about one pitch in the speech unit connection means 106, and the connected synthesized speech is output by the synthesized speech output means 1.
07. Also, the stored speech units have two syllables at the beginning, those with a higher pitch frequency up to the accent nucleus, and those with the endings are closer to the speech unit waveform at the time of recording, and the total speech unit amount is smaller. Nevertheless, it has the feature that the synthesized speech has high sound quality.

Next, the configuration of a speech synthesizer according to a second embodiment of the present invention will be described with reference to FIG. FIG.
, 201 is an input character string conversion means for inputting a character string mixed with kanji kana or reading kana and prosody information and converting it into phonetic notation, and 202 calculates a prosody (pitch frequency) of the entire synthesized speech based on the converted phonetic notation. A memory storing a database including a speech unit including at least one of the first two syllables, a speech unit up to the accent nucleus, a speech unit including the ending, and a speech unit for the other, Reference numeral 203 denotes a speech unit selection unit that converts the phonetic notation output from the input character string conversion unit 201 into a speech unit notation in CV or VCV units, searches the memory 204, and selects a corresponding speech unit. is there.

Reference numeral 205 denotes the fine structure of the prosody at the time of recording the selected speech unit including at least one of the first two syllables, the speech unit having a high pitch frequency up to the accent nucleus, and the speech unit including the ending. As it is, the prosody (pitch frequency) is transformed according to the formula (2-1) or (2-2) (described later), and the deformation of the speech unit for the other is performed by refining the prosody at the time of recording the speech unit. Speech unit transformation means for transforming using the prosody calculated by the prosody calculation means 202 without using the structure, speech unit connection means 206 for connecting the transformed speech units by averaging, This is a synthesized voice output unit that outputs a synthesized voice to which the segments are connected.

Next, the operation of the speech synthesizer according to the second embodiment of the present invention will be described with reference to FIG. The input character string conversion means 201 converts the input character string mixed with kanji kana or the reading kana and prosodic information into phonetic notation. An example of this conversion will be described later with reference to FIG. Next, the prosody (pitch frequency) of the entire synthesized speech is calculated by the prosody calculation means 202 based on the converted phonetic notation. The phonetic notation output from the input character string converting means 201 is converted into a phonemic unit notation in CV or VCV units by the voice unit selecting means 203, and a phoneme unit including at least one of the first two syllables and an accent The database in the memory 204 storing the speech unit up to the nucleus, the speech unit including the ending, and the speech unit for the others is searched, and the corresponding speech unit is selected.

Next, these speech segments are sent to speech segment transformation means 205, and a speech segment including at least one of the two syllables at the beginning, a speech segment having a high pitch frequency up to the accent nucleus, and an ending are added. The speech unit to be included is the expression (2-1) or (2-2) while keeping the fine structure of the prosody at the time of recording.
In accordance with (described later), the prosody (pitch frequency) is transformed so as to match the vowel start part and the vowel stationary part pitch frequency of the model pitch frequency of the synthesized speech, and the speech unit for the other is transformed by the speech unit recording. Instead of using the fine structure of the prosody at the time, the prosody is transformed using the prosody calculated by the prosody calculation means 202 first. These deformed speech units are connected by averaging in a range of about one pitch in the speech unit connection unit 206, and the connected synthesized speech is output to the synthesized speech output unit 207.
Output from

Next, with reference to FIG. 3, the configuration of a speech synthesizing apparatus according to a third embodiment of the present invention will be described. FIG.
, 301 is an input character string conversion means for inputting a character string mixed with kanji or kana or reading kana and prosody information and converting it into phonetic notation, and 302 is a prosody calculation for calculating the prosody (pitch frequency) of the whole synthesized speech based on the phonetic notation. Means 304, a memory storing a database including a speech unit including at least one of the first two syllables, a speech unit including the ending, and a speech unit for other words; 303, an input character string conversion unit 3;
This is a speech unit selection unit that converts the phonetic notation output from No. 01 into a speech unit notation in CV or VCV units, searches the memory 304 and selects a corresponding speech unit.

Reference numeral 305 denotes a speech unit including at least one of the selected two initial syllables and a speech unit including the ending and a speech unit for the other words, while retaining the fine structure of the prosody at the time of recording. The speech unit transforming means 306 transforms the prosody (pitch frequency) according to the formula (3-1) or (3-2) (to be described later). A speech unit 306 connects the transformed speech units by averaging. One-side connecting means 307 is a synthesized voice output means for outputting a synthesized voice to which a voice unit is connected.

Next, the operation of the speech synthesizer according to the third embodiment of the present invention will be described with reference to FIG. The input character string conversion means 301 converts the input character string mixed with kanji kana or the reading kana and prosodic information into phonetic notation. An example of this conversion will be described later with reference to FIG. Next, the prosody (pitch frequency) of the entire synthesized speech is calculated by the prosody calculation means 302 based on the converted phonetic notation. The phonetic notation output from the input character string conversion means 301 is converted into a speech unit notation in units of CV or VCV by the speech unit selection means 303, and a speech unit including at least one of the first two syllables and a suffix are added. Is searched in the database of the memory 304 in which the speech unit including the above and the other speech units are stored, and the corresponding speech unit is selected.

Next, these speech units are sent to speech unit transformation means 305, and a speech unit including at least one of the first two syllables and a speech unit including the ending and a speech unit for the other are used. Means the vowel start part and the vowel stationary part pitch frequency of the model pitch frequency of the synthesized voice according to the equation (3-1) or (3-2) (described later) while keeping the fine structure of the prosody at the time of recording. The prosody (pitch frequency) is modified to match. These deformed speech units are connected by averaging in a range of about one pitch in the speech unit connection unit 306, and the connected synthesized speech is output by the synthesized speech output unit 3.
07.

Next, with reference to FIG. 4, a configuration of a speech synthesizing apparatus according to a fourth embodiment of the present invention will be described. FIG.
, 401 is an input character string conversion means for inputting a character string mixed with kanji kana or reading kana and prosody information and converting it into a phonetic transcription, and 402 calculates a prosody (pitch frequency) of the whole synthesized speech based on the converted phonetic transcription. 404 is a memory storing a database including a speech unit including at least one of the first two syllables, a speech unit including the ending, and a speech unit for other words, and 403 is an input character string conversion unit 401. Is a speech unit selection unit that converts the phonetic notation output from the unit into a speech unit notation in CV or VCV units, searches the memory 404 and selects a corresponding speech unit.

Further, reference numeral 405 denotes an expression (4-1) in which the speech unit including at least one of the selected two syllables at the beginning and the speech unit including the ending are retained with the fine structure of the prosody at the time of recording.
Alternatively, the prosody (pitch frequency) is transformed in accordance with (4-2) (to be described later), and the prosody of the speech unit for the other is transformed without using the fine structure of the prosody at the time of recording the speech unit. Speech unit transformation means for transforming using the prosody calculated by the prosody calculation means 402, speech unit connection means 406 for connecting the transformed speech units by averaging, and 407 a synthesized speech to which speech units are connected. Is output as synthesized voice output means.

Next, the operation of the speech synthesizer according to the fourth embodiment of the present invention will be described with reference to FIG. The input character string conversion means 401 converts the input character string mixed with kanji kana or the reading kana and prosodic information into phonetic notation. An example of this conversion will be described later with reference to FIG. Next, the prosody calculation unit 402 calculates the prosody (pitch frequency) of the entire synthesized speech based on the converted phonetic notation. The phonetic notation output from the input character string conversion means 401 is converted into a speech unit notation in CV or VCV units by the speech unit selection means 403, and the phoneme unit including at least one of the first two syllables and the ending are added. The database in the memory 404 storing the speech units including the speech units and the speech units for other speech units is searched, and the corresponding speech units are selected.

Next, these speech units are sent to the speech unit transformation unit 405, and the speech unit including at least one of the first two syllables and the speech unit including the ending are combined with the fine structure of the prosody at the time of recording. (4-1) or (4-2)
In accordance with (described later), the prosody (pitch frequency) is transformed so as to match the vowel start part and the vowel stationary part pitch frequency of the model pitch frequency of the synthesized speech, and the speech unit for the other is transformed by the speech unit recording. Instead of using the fine structure of the prosody at the time, the prosody is transformed using the prosody calculated by the prosody calculation means 402 first. These transformed speech units are connected by averaging in a range of about one pitch in the speech unit connection unit 406, and the connected synthesized speech is output by the synthesized speech output unit 40.
7 is output.

Next, referring to the flowchart of FIG.
The flow of the speech synthesis method according to the fifth embodiment of the present invention will be described. In step 110, a phonetic notation is determined from a character string such as the input kanji kana spelling notation or its reading kana, and the phonetic notation is further converted to CV or VCV.
It is decomposed into units (C means consonant, V means vowel).
Next, proceeding to step 111, based on the Fujisaki model or the like, the prosody (particularly pitch frequency) of the word to be synthesized is calculated from information such as the kanji kana spelling of the word to be synthesized or its reading kana and accent information. (Hereinafter referred to as model pitch frequency).

Next, in step 112, a speech unit counter is initialized, and in step 113, for each speech unit, one of the first two syllables of the synthesized speech, or the pitch frequency of the pitch pitch up to the accent nucleus is high. Are selected in order from the beginning of the word, including the end of the word, or those other than these, and the speech unit is one of the first two syllables of the synthesized speech, the one with a high pitch frequency to the accent nucleus, or the end of the word. If they contain CV or VC with the same phonetic notation from the speech units recorded for them
The V speech unit is selected (step 114), and the speech unit used for the other parts is a flat speech unit, and a CV or VCV speech unit having the same phonetic notation is also selected. (Step 115).

FIG. 5 illustrates a case where "Yokosuka City" is synthesized using the character string "Yokosuka City" as an example. In FIG. 5, the character string notation is “Yokosuka City”, which is converted into a phonetic notation “yo ko sxu ka1 sh”.
i ”, and the corresponding synthesized speech unit or synthesized unit is“ yo (S) ko (S) osxu xuka
(A) ashi i (E) ".

Any one of the two initial syllables selected in step 114 and step 115, the speech unit having a high pitch frequency up to the accent nucleus, the speech unit including the ending, or the other speech unit is , Step 11
In step 6, it is deformed according to the prosody (pitch frequency) of the synthesized speech. That is, the pitch frequency of the speech unit is modified by the following equation [1-1] or [1-2] so as to match the vowel start part and the vowel stationary part pitch frequency of the model pitch frequency of the synthesized speech. The deformation of the pitch frequency is realized by superimposition of a pitch waveform.

In the case of a CV speech unit: p (t) = p0 (t) {p1 / p01 + (P2 / p02-p1 / p01) t / t1} (1-1) where p (t): Speech unit pitch frequency after deformation p0 (t): Speech unit pitch frequency before deformation p01: Vowel start pitch frequency of speech unit before deformation p02: Vowel steady part pitch frequency of speech unit before deformation p1 : Model pitch frequency at the position corresponding to the vowel start part of the unit p2: Model pitch frequency at the position corresponding to the vowel stationary part of the unit t: Time from the start of the vowel as the origin t1: Time of the vowel stationary part (the origin is Vowel start time)

In the case of a VCV speech unit: p (t) = p0 (t) {p1 / p01 + (P2 / p02-p1 / p01) t / t1} (1-2) where p (t): Speech unit pitch frequency after deformation p0 (t): Speech unit pitch frequency before deformation p01: First vowel stationary part pitch frequency of speech unit before deformation p02: Second vowel stationary state of speech unit before deformation Part pitch frequency p1: Model pitch frequency at the position corresponding to the first vowel stationary part of the unit p2: Model pitch frequency at the position corresponding to the second vowel stationary part of the unit t: The origin is the time of the first vowel stationary part Time t1: the second vowel steady part time (the origin is the time of the vowel start part)

After the above speech unit deformation, the pitch waveforms of the second steady part of the second vowel of the preceding speech unit and the steady part of the first vowel of the first speech unit are averaged and connected. (Step 117). After the connection is completed, the presence or absence of the next speech unit is determined (step 118), and if so, the next speech unit is selected (step 113). In this way,
A synthesized speech is created by deforming and connecting the speech units.

It goes without saying that such a speech synthesizing method is similarly applied not only to the creation of words but also to the production of synthesized speech such as phrases, phrases or sentences. Also, it goes without saying that the same processing can be performed by using a speech unit other than the CV or VCV format, for example, a speech unit of another format or type such as VCV or VV.

Next, referring to the flowchart of FIG.
The flow of the speech synthesis method according to the sixth embodiment of the present invention will be described. In step 120, a phonetic notation is determined from a character string such as the input kanji kana spelling or the reading kana, and the phonetic notation is converted to CV or VCV.
It is decomposed into units (C means consonant, V means vowel).
Next, proceeding to step 121, based on the Fujisaki model or the like, the prosody (particularly pitch frequency) of the word to be synthesized is calculated from information such as the kanji kana spelling of the word to be synthesized or its reading kana and accent information. (Hereinafter referred to as model pitch frequency).

Next, in step 122, a speech unit counter is initialized, and in step 123, for each speech unit, one of the two syllables at the beginning of the synthesized speech, or the pitch frequency up to the accent nucleus is high. Are selected in order from the beginning of the word, including the end of the word, or those other than these, and the speech unit is one of the first two syllables of the synthesized speech, the one with a high pitch frequency to the accent nucleus, or the end of the word. If they contain CV or VC with the same phonetic notation from the speech units recorded for them
A V-speech unit is selected (step 124), and speech units used for other parts are flat-speech-based speech units, and similarly, a CV or VCV speech unit having the same phonetic notation is selected. (Step 126).

FIG. 5 illustrates a case where "Yokosuka City" is synthesized using the character string "Yokosuka City" as an example. In FIG. 5, the character string notation is “Yokosuka City”, which is converted into a phonetic notation “yo ko sxu ka1 sh”.
i ”, and the corresponding synthesized speech unit or synthesized unit is“ yo (S) ko (S) osxu xuka
(A) ashi i (E) ".

Initial 2 selected in step 124
At step 125, any one of the syllables or a speech unit having a high pitch frequency up to the accent nucleus or a speech unit including the ending is deformed according to the prosody (pitch frequency) of the synthesized speech. That is, the following formula [2-
According to [1] or [2-2], the pitch frequency of the speech unit is changed so as to match the vowel start portion and the vowel stationary portion pitch frequency of the model pitch frequency of the synthesized speech. The deformation of the pitch frequency is realized by superimposition of a pitch waveform.

In the case of a CV speech unit: p (t) = p0 (t) {p1 / p01 + (P2 / p02-p1 / p01) t / t1} (2-1) where p (t): Speech unit pitch frequency after deformation p0 (t): Speech unit pitch frequency before deformation p01: Vowel start pitch frequency of speech unit before deformation p02: Vowel steady part pitch frequency of speech unit before deformation p1 : Model pitch frequency at the position corresponding to the vowel start part of the unit p2: Model pitch frequency at the position corresponding to the vowel stationary part of the unit t: Time from the start of the vowel as the origin t1: Time of the vowel stationary part (the origin is Vowel start time)

In the case of the VCV speech unit: p (t) = p0 (t) {p1 / p01 + (P2 / p02-p1 / p01) t / t1} (2-2) where p (t): Speech unit pitch frequency after deformation p0 (t): Speech unit pitch frequency before deformation p01: First vowel stationary part pitch frequency of speech unit before deformation p02: Second vowel stationary state of speech unit before deformation Part pitch frequency p1: Model pitch frequency at the position corresponding to the first vowel stationary part of the unit p2: Model pitch frequency at the position corresponding to the second vowel stationary part of the unit t: The origin is the time of the first vowel stationary part Time t1: the second vowel steady part time (the origin is the time of the vowel start part)

On the other hand, any other speech unit other than the speech unit including any one of the first two syllables or the speech ending at the end of the word is determined at step 127 at each time determined by the model pitch frequency of the synthesized speech. It is deformed by superimposing in units of pitch waveforms in accordance with the pitch cycle.

After the above speech unit deformation, the pitch waveforms of the second stationary part of the preceding vowel and the stationary part of the first vowel of the preceding speech unit are averaged and connected. (Step 128). After the connection is completed, the presence or absence of the next speech unit is determined (step 129), and if there is, the next speech unit is selected (step 123). In this way,
A synthesized speech is created by deforming and connecting the speech units.

It is needless to say that such a speech synthesizing method is similarly applied not only to the creation of words but also to the production of synthesized speech such as phrases, phrases or sentences. Also, it goes without saying that the same processing can be performed by using a speech unit other than the CV or VCV format, for example, a speech unit of another format or type such as VCV or VV.

Next, referring to the flowchart of FIG.
The flow of the speech synthesis method according to the seventh embodiment of the present invention will be described. In step 130, the phonetic notation is determined from the character string such as the input kanji kana spelling notation or its reading kana, and the phonetic notation is further converted to CV or VCV.
It is decomposed into units (C means consonant, V means vowel).
Next, proceeding to step 131, based on the Fujisaki model or the like, the prosody (particularly pitch frequency) of the word to be synthesized is calculated from the information such as the kanji kana spelling of the word to be synthesized or its reading kana and accent information. (Hereinafter referred to as model pitch frequency).

Next, in step 132, a speech unit counter is initialized, and in step 133, for each speech unit, whether they include any one of the first two syllables of the synthesized speech, the ending of the syllable, or the like. If the speech unit includes any one of the first two syllables or the end of the synthesized speech, the phonetic notation is the same from the speech unit recorded for them.
Alternatively, a VCV speech unit is selected (step 134),
Speech units used for other parts are plate-speech-based speech units, and also have the same phonetic notation of CV or VC.
A V speech unit is selected (step 135).

FIG. 5 illustrates a case where the character string "Yokosuka City" is used as an example and "Yokosuka City" is synthesized. In FIG. 5, the character string notation is “Yokosuka City”, which is converted into a phonetic notation “yo ko sxu ka1 sh”.
i ”, and the corresponding synthesized speech unit or synthesized unit is“ yo (S) ko (S) osxu xuka
(A) ashi i (E) ".

In step 136, any one of the first two syllables selected in steps 134 and 135, the speech unit including the ending, or the other speech unit is processed in step 136 by the prosody (pitch frequency) of the synthesized speech. ). That is, the pitch frequency of the speech unit is modified by the following equation [3-1] or [3-2] so as to match the vowel start portion and the vowel stationary portion pitch frequency of the model pitch frequency of the synthesized speech. The deformation of the pitch frequency is realized by superimposition of a pitch waveform.

In the case of a CV speech unit: p (t) = p0 (t) {p1 / p01 + (P2 / p02-p1 / p01) t / t1} (3-1) where p (t): Speech unit pitch frequency after deformation p0 (t): Speech unit pitch frequency before deformation p01: Vowel start pitch frequency of speech unit before deformation p02: Vowel steady part pitch frequency of speech unit before deformation p1 : Model pitch frequency at the position corresponding to the vowel start part of the unit p2: Model pitch frequency at the position corresponding to the vowel stationary part of the unit t: Time from the start of the vowel as the origin t1: Time of the vowel stationary part (the origin is Vowel start time)

In the case of a VCV speech unit: p (t) = p0 (t) {p1 / p01 + (P2 / p02-p1 / p01) t / t1} (3-2) where p (t): Speech unit pitch frequency after deformation p0 (t): Speech unit pitch frequency before deformation p01: First vowel stationary part pitch frequency of speech unit before deformation p02: Second vowel stationary state of speech unit before deformation Part pitch frequency p1: Model pitch frequency at the position corresponding to the first vowel stationary part of the unit p2: Model pitch frequency at the position corresponding to the second vowel stationary part of the unit t: The origin is the time of the first vowel stationary part Time t1: the second vowel steady part time (the origin is the time of the vowel start part)

After the above speech unit deformation, the pitch waveforms of the second steady part of the second vowel of the preceding speech unit and the steady part of the first vowel of the first speech unit are averaged and connected. (Step 137). After the connection is completed, the presence or absence of the next speech unit is determined (step 138), and if there is, the next speech unit is selected (step 133). In this way,
A synthesized speech is created by deforming and connecting the speech units.

It is needless to say that such a speech synthesizing method is similarly applied not only to the production of words but also to the production of synthesized speech such as phrases, phrases or sentences. Also, it goes without saying that the same processing can be performed by using a speech unit other than the CV or VCV format, for example, a speech unit of another format or type such as VCV or VV.

Next, referring to the flowchart of FIG.
The flow of the speech synthesis method according to the eighth embodiment of the present invention will be described. In step 140, a phonetic notation is determined from a character string such as the input kanji kana spelling notation or its reading kana, and the phonetic notation is further converted to CV or VCV.
It is decomposed into units (C means consonant, V means vowel).
Next, proceeding to step 141, based on the Fujisaki model or the like, the prosody (particularly pitch frequency) of the word to be synthesized is calculated from information such as the kanji kana spelling of the word to be synthesized or its reading kana and accent information. (Hereinafter referred to as model pitch frequency).

Next, in step 142, a speech unit counter is initialized, and in step 143, for each speech unit, whether they include any one of the first two syllables of the synthesized speech, the end of the syllable, or the like. If the speech unit includes any one of the first two syllables or the end of the synthesized speech, the phonetic notation is the same from the speech unit recorded for them.
Alternatively, a VCV speech unit is selected (step 144),
Speech units used for other parts are plate-speech-based speech units, and also have the same phonetic notation of CV or VC.
A V speech unit is selected (step 146).

FIG. 5 illustrates a case where "Yokosuka City" is synthesized using the character string "Yokosuka City" as an example. In FIG. 5, the character string notation is “Yokosuka City”, which is converted into a phonetic notation “yo ko sxu ka1 sh”.
i ”, and the corresponding synthesized speech unit or synthesized unit is“ yo (S) ko (S) osxu xuka
(A) ashi i (E) ".

The head 2 selected in step 144
In step 145, any one of the syllable speech units or the speech unit including the ending is deformed according to the prosody (pitch frequency) of the synthesized speech. That is, the pitch frequency of the speech unit is modified by the following equation [4-1] or [4-2] so as to match the vowel start part and the vowel stationary part pitch frequency of the model pitch frequency of the synthesized speech. The deformation of the pitch frequency is realized by superimposition of a pitch waveform.

In the case of the CV speech unit: p (t) = p0 (t) {p1 / p01 + (P2 / p02-p1 / p01) t / t1} (4-1) where p (t): Speech unit pitch frequency after deformation p0 (t): Speech unit pitch frequency before deformation p01: Vowel start pitch frequency of speech unit before deformation p02: Vowel steady part pitch frequency of speech unit before deformation p1 : Model pitch frequency at the position corresponding to the vowel start part of the unit p2: Model pitch frequency at the position corresponding to the vowel stationary part of the unit t: Time from the start of the vowel as the origin t1: Time of the vowel stationary part (the origin is Vowel start time)

In the case of a VCV speech unit: p (t) = p0 (t) {p1 / p01 + (P2 / p02-p1 / p01) t / t1} (4-2) where p (t): Speech unit pitch frequency after deformation p0 (t): Speech unit pitch frequency before deformation p01: First vowel stationary part pitch frequency of speech unit before deformation p02: Second vowel stationary state of speech unit before deformation Part pitch frequency p1: Model pitch frequency at the position corresponding to the first vowel stationary part of the unit p2: Model pitch frequency at the position corresponding to the second vowel stationary part of the unit t: The origin is the time of the first vowel stationary part Time t1: the second vowel steady part time (the origin is the time of the vowel start part)

On the other hand, in step 147, a speech unit other than any one speech unit of the first two syllables or a speech unit including the ending is added at step 147 at each time determined by the model pitch frequency of the synthesized speech. It is deformed by superimposing in units of pitch waveforms in accordance with the pitch cycle.

After the above speech unit deformation, the pitch waveforms of the second steady part of the second vowel of the preceding speech unit and the steady part of the first vowel of the first speech unit are averaged and connected. (Step 148). After the connection is completed, the presence or absence of the next speech unit is determined (step 149). If there is, the next speech unit is selected (step 143). In this way,
A synthesized speech is created by deforming and connecting the speech units.

It is needless to say that such a speech synthesizing method is similarly applied not only to the production of words but also to the production of synthesized speech such as phrases, phrases or sentences. Also, it goes without saying that the same processing can be performed by using a speech unit other than the CV or VCV format, for example, a speech unit of another format or type such as VCV or VV.

[0068]

The present invention is configured as described above, and in particular,
Instead of using spectral transformations that require large storage capacity as in the prior art, two initial syllables, one with a high pitch frequency up to the accent nucleus, or a speech unit at the end, which are particularly important for maintaining the sound quality of the synthesized speech Uses the fine structure of prosody at the time of recording speech units that do not require as much storage capacity as the case of using spectrum conversion to transform (prosody), and uses other speech units that do not significantly affect the sound quality of synthesized speech The (prosody) transformation is performed by calculation without using the fine structure of the prosody at the time of speech unit recording, so that the speech units stored in the database can be reduced while maintaining the high quality of synthesized speech. Thus, the capacity of the memory for storing the speech unit database can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech synthesizer according to a first embodiment of the present invention.

FIG. 2 is a block diagram illustrating a configuration of a speech synthesizer according to a second embodiment of the present invention;

FIG. 3 is a block diagram showing a configuration of a speech synthesizer according to a third embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a speech synthesizer according to a fourth embodiment of the present invention.

FIG. 5 is a view for explaining a state of conversion into phonetic notation when an input is a kanji character string in the speech synthesizer shown in FIG. 1;

FIG. 6 is a flowchart illustrating a speech synthesis method according to a fifth embodiment of the present invention.

FIG. 7 is a flowchart illustrating a speech synthesis method according to a sixth embodiment of the present invention.

FIG. 8 is a flowchart showing a speech synthesis method according to a seventh embodiment of the present invention.

FIG. 9 is a flowchart illustrating a speech synthesis method according to an eighth embodiment of the present invention.

FIG. 10 is a block diagram showing the configuration of a conventional speech synthesizer.

[Explanation of symbols]

101, 201 input character string conversion means 102, 202 prosody calculation means 103, 203 speech unit selection means 301, 401 input character string conversion means 302, 402 prosody calculation means 303, 403 speech unit selection means 104, 204 105, 205 Speech unit transformation means 106, 206 Speech unit connection means 107, 207 Synthetic speech output means 305, 405 Speech unit database for speech units with high pitch frequencies up to accent nuclei, endings, etc. Unit transforming means 306, 406 Speech unit connecting means 307, 407 Synthetic speech output means 304, 404 Memory (stores speech prefix, ending, and other speech unit databases for other uses) 501 Input unit 502 Prosodic parameter generation unit 503 Prosodic rule dictionary 504 Audio unit selection unit 505 Large capacity audio Unit database (DB) 506 Speech unit set selection unit 507 Speech unit set 508 Speech synthesizer 509 Output unit

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-6-95692 (JP, A) JP-A-4-281495 (JP, A) JP-A 1-284898 (JP, A) JP-A-2- 238497 (JP, A) JP-A-7-181995 (JP, A) JP-A-7-56597 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 13/06 G10L 13 / 08

Claims (8)

(57) [Claims] An input character string converting means for converting an input kanji kana mixed character string or a reading kana and prosody information into a phonetic notation, and a prosody calculating means for calculating a prosody of a synthesized speech based on the converted phonetic notation. If, prefix 2 syllables of speech units and these audio including speech unit and endings of high pitch frequencies up speech unit and accent core containing at least one
A memory storing at least speech units other than the speech unit, a speech unit selection unit for selecting a corresponding speech unit from the memory according to the phonetic notation, and a fine structure of the prosody at the time of the speech unit recording. Speech unit transforming means for transforming the prosody of the selected speech unit, speech unit connecting means for connecting the transformed speech unit, and synthesized speech output means for outputting the connected synthesized speech. A speech synthesizer comprising: 2. An input character string converting means for converting an input character string mixed with kanji kana or a reading kana and prosodic information into a phonetic notation, and a prosody calculating means for calculating a prosody of a synthesized speech based on the converted phonetic notation. If, prefix 2 syllables of speech units and these audio including speech unit and endings of high pitch frequencies up speech unit and accent core containing at least one
A memory storing at least a speech unit other than a unit, a speech unit selection unit for selecting a corresponding speech unit from the memory according to the phonetic notation, and the selection while retaining the fine structure of the prosody at the time of recording. been prefixes perform prosody modification of speech units including speech segment and ending high pitch frequencies up speech unit and accent nucleus containing 2 syllables least one of, speech units other than those of speech units Transformation of the prosody is performed using the calculated prosody without using the fine structure of the prosody at the time of the speech unit recording, a speech unit transformation unit, and a speech unit connection unit that connects the transformed speech unit, A speech synthesizer comprising a connected synthesized speech output unit for outputting the synthesized speech. 3. An input character string converting means for converting an input character string mixed with kanji kana or a reading kana and prosodic information into a phonetic notation, and a prosody calculating means for calculating a prosody of a synthesized speech based on the converted phonetic notation. When, the prefix 2 syllables speech unit and of including speech unit and endings including at least one
A memory storing at least a speech unit other than the speech unit, a speech unit selection means for selecting a corresponding speech unit from the memory according to the phonetic notation, and a fine structure of a prosody at the time of speech unit recording. Speech unit transformation means for transforming the prosody of the selected speech unit as it is, speech unit connection means for connecting the transformed speech unit, and synthesized speech output for outputting the connected synthesized speech And a voice synthesizing device. 4. An input character string converting means for converting an input character string mixed with kanji kana or a reading kana and prosodic information into a phonetic notation, and a prosody calculating means for calculating a prosody of a synthesized speech based on the converted phonetic notation. When, the prefix 2 syllables speech unit and of including speech unit and endings including at least one
A memory storing at least a speech unit other than the speech unit, a speech unit selection means for selecting a corresponding speech unit from the memory according to the phonetic notation, and the fine structure of the prosody at the time of recording is retained. The prosody of the speech unit including at least one of the selected two initial syllables and the speech unit including the ending is modified, and the prosody of the speech unit other than these speech units is modified according to the prosody at the time of recording the speech unit. Speech unit deformation means for performing using the calculated prosody without using the fine structure of, speech unit connection means for connecting the deformed speech unit,
A speech synthesizer comprising a connected synthesized speech output unit for outputting the synthesized speech. 5. A speech unit including at least one of the first two syllables from a memory based on the input kanji kana mixed character string or reading kana and prosodic information and converted into phonetic notation based on the converted phonetic notation. Speech units with high pitch frequency up to the accent nucleus and speech units including endings and these
Be applicable from the speech segment of the non-voice segment in the phonetic transcription
The speech unit is selected, the prosody of the selected speech unit is transformed while the fine structure of the prosody at the time of recording the speech unit is left, and the transformed speech unit is connected and output as a synthesized speech. A speech synthesis method comprising the steps of: 6. Converting the input kanji kana mixed character string or reading kana and prosodic information into a phonetic notation, calculating the prosody of the synthesized speech based on the converted phonetic notation, and converting the phonetic notation into the converted phonetic notation. This and speech units from based memory containing speech unit and endings of high pitch frequencies up speech unit and accent core containing at least one of the prefix 2 syllables
The above phonetic notation from speech units other than these speech units
And a speech unit including at least one of the selected two syllables and a speech unit having a high pitch frequency up to the accent nucleus while retaining the fine structure of the prosody at the time of recording the speech unit. Transform the prosody with the speech unit including the ending, and perform the transformation of the prosody of the speech unit other than these speech units using the calculated prosody without using the fine structure of the prosody at the time of recording, A speech synthesis method comprising the steps of connecting the transformed speech units and outputting the speech as synthesized speech. 7. A speech unit including at least one of the first two syllables from a memory based on the input kanji kana mixed character string or reading kana and prosodic information and converted into phonetic notation based on the converted phonetic notation. this and the speech segment, including the endings
The phonetic unit other than these phonemic units is used as the phonetic notation.
Select the speech unit equivalent to perform the deformation of prosody of the selected speech unit while leaving the prosody of the microstructure during the speech unit recording, as synthesized speech by connecting the modified speech unit A speech synthesis method comprising the steps of outputting. 8. Converting the input kanji kana mixed character string or reading kana and prosodic information into a phonetic notation, calculating the prosody of the synthesized speech based on the converted phonetic notation, and converting the phonetic notation into the converted phonetic notation. A speech unit containing at least one of the first two syllables from the memory and a speech unit containing the ending
From these speech units other than these speech units ,
Selecting a corresponding speech unit and transforming the prosody between the speech unit including at least one of the selected two initial syllables and the speech unit including the ending while retaining the fine structure of the prosody at the time of recording the speech unit. The prosody modification of the speech units other than these speech units is performed using the calculated prosody without using the fine structure of the prosody at the time of recording, and the transformed speech units are connected and synthesized. A speech synthesis method comprising the steps of outputting as speech.