JP3077981B2 - Basic frequency pattern generator - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech synthesizer, and more particularly to a fundamental frequency pattern generator used in a speech rule synthesizer or a text synthesizer that generates a fundamental frequency pattern according to rules. is there.

[Conventional technology]

A conventional speech synthesizer that uses characters and symbols as input (for example, Proceedings of the 1986 IEICE General Conference)
FIG. 2 shows a processing block diagram of S26-5, March 1986). Table 1 shows an example of input to the speech synthesizer.

Inputs are accent symbols, phrase symbols, pause symbols, and syllable symbols.

The accent marks are A1 (0.40), A2 (0.26), and A0. A1 and A2 indicate the syllable boundary at the rising edge of the accent and the type of accent size, and A0 indicates the syllable boundary at the falling edge of the accent. The size of the accent is
There are two types indicated by the symbols A1 and A2. () Shows the size of the accent actually assigned.

Phrase symbols are P1 (0.43), P2 (0.26), P3 (0.1
2) and P0, where P1, P2, and P3 indicate the starting point and size of the phrase, and P0 is the preceding P1, P
2, indicates that the phrase component generated by P3 is reduced to zero. The size in parentheses indicates the size of the phrase to be assigned.

The pause symbol is “.” (0.7 seconds), “,” (0.3 seconds) “•”
(0.08 seconds), indicating that there is a pause at the syllable boundary, that is, there is a pause. The number in parentheses indicates the length of the pause.

The syllables are symbols represented by katakana such as "su", "zu", "me", "wa" ... and indicate the type of sound.

During the input, a symbol with a circle, for example, "shi" indicates a silenced "shi".

The voice to be synthesized is specified by the input, and the synthesis process is performed as follows.

(1) Generation of phonemic parameters A speech specified by a syllable symbol is selected from a storage pattern, and based on a time length specific to the syllable described in the storage pattern and a pause time specified by a pause symbol. To determine the time of the syllable.

Next, the phonetic parameters specific to the syllable described in the storage pattern, such as the time-varying pattern of the formant frequency and the bandwidth, are read out, and the phonemic parameters are expanded so that the time points of the respective syllables defined above are satisfied.・ Connect each other while compressing. For example, a syllable C ₁ V ₁
At time t = 0, the next syllable C ₂ V ₂ is t = 140 msec, the phonetic parameters of C ₁ V ₁ are described only up to t = 100 msec, and C ₂ V ₂ Phonemic parameter is t
= If we do not describe only the accumulation pattern of minute from 140Msec, between the t = 100 msec to t = 140Msec is compensated for by stretching a portion of the V _1.

Through the above processing, phonemic parameters of the sentence speech to be synthesized are obtained, sent to a speech synthesizer (for example, a formant synthesizer), and used for generating a speech signal.

(2) Generation of sound source intensity pattern The value of the sound source intensity must be determined for each type of syllable to be synthesized, and must be decreased or increased before and after pausing. The value unique to the syllable type is also described in the storage pattern, and by performing expansion and contraction processing similar to the phonemic parameters and joining them together, a basic sound source intensity pattern of the target sentence can be obtained. Further, before and after the pause, particularly before and after the pause of "." Which indicates a sentence-to-sentence separation, a certain amount is decreased or increased in accordance with the sound source intensity rule, thereby obtaining a final sound source intensity pattern. And is used for audio signal generation.

(3) fundamental frequency in (voice pitch, serial represented by F ₀₎ pattern generator input, the point of the phrase and accent, what is the syllable boundary is shown, moreover syllable as described above Since the time point is determined, the time point of the phrase and the accent can be determined based on the time point of the syllable. The actual value to be used is determined by the type of the phrase symbol and accent symbol being input (for example, A1
Is 0.40), so that the size and time of the phrase command and the accent command can be determined. The equation generation model of F ₀ pattern of time and size of a phrase and accent based, and generates the F ₀ pattern.

The F ₀ pattern generation model is shown in Figure 3. A time change pattern of the F ₀ expressed as F ₀ (t), performs calculation according to the following equation.

Where the impulse response and the step response of the critical damping quadratic linear system, respectively.

α and β are constants that determine the response speed, α = 3.0,
Use a value of about β = 20.0.

1 (1 + βt) exp (−βt) asymptotically approaches the target value 1.0 as t increases, but Ga (t) within a finite time
Is converged to the target value, the processing is performed with θ = 0.9. In the case of θ ≦ 1, the target value of Ga (t) is θ.

I indicates the number of phrases appearing in the sentence, Ap
i indicates the size of the i-th phrase command. For example, Ap _i = 0.43 where the phrase indicated by the symbol P1 (0.43) comes. To _i indicates the time of the phrase command.

J indicates the number of accents appearing in the text,
Aa _j indicates the size of the jth accent command. For example, if accent shown as the first accent symbol A1 (0.40) comes, the Aa ₁ = 0.40. T _1j and T _2j indicate the start point and end point of the j-th accent command.

1nFmin is a constant term and corresponds to the lowest vibrating frequency of the vocal cords. For example, when synthesizing male voice, Fmin ≒ 75Hz, and when synthesizing female voice, Fmin
≒ Set to about 115Hz.

When calculating F ₀ (t), the size of the phrase command determined by the above-described processing and the time Ap _i , To _i (1 ≦ i ≦
I), the size of the accent command and the time points Aa _j , T _1j , T
_The right side is calculated by applying _2j (1 ≦ j ≦ J) to the above-described equation, and the inverse function of the logarithm, that is, the exponential function, is used to calculate P ₀ (t).

F ₀ (t) obtained by the above processing, that is, the fundamental frequency pattern is sent to the speech synthesizer and used for generating a speech signal.

The hardware used for the above-described processing is such that a speech synthesizer (for example, a formant synthesizer) is realized by a signal processor, and the processing up to creating an input to the speech synthesizer from an input signal is performed by a microprocessor. It is processed. The storage pattern is stored in a ROM accessed by the microprocessor.

The calculation of the formula of F ₀ (t) is realized by a program of a microprocessor.

In the above description, the processing of synthesizing speech by inputting syllable symbols, pause symbols, accent symbols, and phrase symbols has been described. However, a speech synthesizer that inputs a sentence mixed with kanji kana is also known. In this case, it is necessary to convert the sentence mixed with the kanji kana into the above-mentioned phonetic symbols, pause symbols, accent symbols, and phrase symbols. In this process, the input sentence is divided into words, the reading is determined by searching the word dictionary, the syllable boundaries of the accents are determined by the accent type written in the word dictionary, and the accent size is determined. Is performed.

Next, a description will be given of how to give a pause symbol / phrase symbol to a sentence to be synthesized according to the prior art (see the above-mentioned paper).

This way of giving is based on the syntactic structure (syntax) of the sentence, and is as follows. There are various other modified and improved versions of the derivation rules for deriving pause symbols and phrase symbols from the syntactic structure.

(1) Put “.” And P0 at the period of the sentence and P1 at the beginning of the sentence.

(2) Put “,” and P0 and P1 at the reading point of the sentence.

(3) Put “•” and P2 on a word boundary that is syntactically larger than a reading point (or a punctuation mark if not) or a word boundary one step smaller.

(4) Put P3 at the word boundary two steps smaller than the reading point.

(5) However, at the boundary between words having a qualifying relationship,
Regardless of (4), do not leave pause / phrase symbols.

(6) When the pause / phrase interval set as described above is at least a certain distance apart (at a normal utterance speed of 13
P3 is added in the order of larger word boundaries.

(7) If it is necessary to specifically indicate a syntactic boundary, place P3 even on a small word boundary.

(8) Check the intervals between P1 and P2 immediately before for all P2s, and change them to P3 if the intervals are small (about 4 mora at normal utterance speed).

[Problems to be solved by the invention]

When calculating the phrase component by the formula, if the phrase command continues for a short time interval, the phrase component will be added by the next phrase command while the phrase component by the previous phrase command remains largely, and the entire phrase component will be added. As for the size becomes considerably large. However, in the utterance of natural speech, the pitch of the voice does not increase unnecessarily due to physiological constraints. For this reason, when the phrase command continues at short time intervals, it is necessary to reduce the size of the phrase command by the process (8).

In other words, it is considered that the phrase command itself is essentially determined by reflecting the syntactic structure of the sentence, but the size of the phrase command actually given needs to be corrected by the previous phrase command. For this reason, the correspondence between the syntactic structure of the sentence and the size of the phrase command was not clear, and the rules for giving the phrase command had to be poorly prospected.

An object of the present invention is to make the processing for correcting the size of a phrase command corresponding to the physiological constraint inherent in the phrase component generation processing, thereby enabling a prosody generation rule with good visibility.

[Means for solving the problem]

The present invention is characterized in that a target value of a phrase component is given as a phrase command, and the size of the phrase command is obtained from the size of the phrase component required to reach the target value.

As shown in FIG. 3, the configuration comprises a control unit 1, a phrase control mechanism 2, an accent control mechanism 3, and a glottal vibration mechanism 4.

The control unit 1 performs the following first processing and second processing. That is, the first process determines a phrase command and an accent command from input characters, symbols, and the like. The phrase command has a time point T0 _i and a size Ap _i , and the accent command has a time point T1 _i
-T2 _i and size Aa _i . The second process, as shown in FIG. 1, the maximum value of the phrase component caused by the i-th phrase command (time T0 _i) is determined as a target value Tp _i, to the target value Tp _i and i-1 th Is calculated in the target unreached portion calculation unit 1a. The size (Ap _i ) of the i-th phrase command is determined based on the unreached portion obtained by the calculation.

As shown in FIG. 1, the phrase control mechanism 2 determines the time point (T0 _i ) of the phrase command determined in the first process and the size (Ap _i ) of the phrase command determined in the second process.
The phrase component is obtained based on

The accent control mechanism 3 obtains an accent component from the accent command determined in the first processing.

The glottal vibration mechanism 4 determines and outputs a fundamental frequency pattern by combining the phrase component and the accent component.

As another example of the second processing, a first processing described later
As shown in the examples, i-th phrase the maximum value of the phrase component caused by the command is determined as the first target value Tpi, i-1 th phrase second target value the maximum value of the phrase component caused by a command Tpi _- determined as _1, first,
The magnitude of the i-th phrase command may be determined based on the second target values and their time intervals.

[First Embodiment] An embodiment in which there are two positive target values of a phrase component will be described.

Gp (t) in the equation is a maximum value Gp at t = 1 / α.
(1 / α) = α / e (e is the base of natural logarithm)
table). Therefore, the value of 0.43 assigned as the size of the phrase command for P1 in the prior art is multiplied by α / e,
The target value of the first phrase component is 0.43 × α / e. α =
Assuming that 3.0 and e = 2.71828, the target value of the phrase component is 0.
It becomes 47. The target value of the second phrase component is set to 80% of the first target value, that is, 0.47 × 0.8 = 0.376.

Next, an example of calculating the phrase component unattainment will be described with reference to FIG. 4 (b). When the i-th phrase command is generated at time t = 0 and the phrase component is added to the phrase component by the (i-1) -th phrase command, the time t _{max at} which the phrase component takes the maximum value is 1 / α. And different. However, considering that t _max cannot be obtained by a simple calculation, and that the magnitude of the phrase command does not necessarily have to be strictly controlled audibly, the unreached portion of the phrase component at t = 1 / α is considered. Is used for calculating the size of the i-th phrase command. That is, in FIG. 4 (b), Tp
_i- c is the phrase component unreached part.

Finally, an embodiment of calculating the size of the phrase command will be described. From FIG. 4 (a), the magnitude of the phrase command may be set to e / α for the phrase component unreached. Therefore, assuming that the phrase component has not been reached is Tp _i −c, the size of the phrase command to be given is (Tp _i −c) × e / α. However, when the size of the phrase command to be given becomes a negative value, the size of the phrase command is set to 0 (i-th phrase command is not generated).

Since the negative phrase symbol P0 is for realizing a decrease in the phrase component, -0.5 may be fixedly assigned in the same manner as in the related art. Alternatively, 0 is assigned as the target value of the phrase component, and a positive value is assigned. The size of the phrase command may be obtained in the same manner as the phrase command.

Second Embodiment Focusing on the fact that the size of a phrase command does not need to be strictly controlled in terms of hearing, an embodiment in which calculation is simpler than in the first embodiment will be described below.

To find the phrase components according to the (i-1) th phrase command, it is necessary to calculate Gp (t) = α ² texp (−αt), but the function takes a long time to calculate. Therefore, if this is described in a table and searched, the calculation time can be reduced. Even if the size of the phrase command is not strict, there is no problem in hearing. Therefore, the table can be described at relatively coarse time intervals (for example, 0.1 seconds) to reduce the storage capacity (second example). table).

Further, since the function of Gp (t) gradually approaches 0 with an increase in t, the phrase components according to the phrase commands up to the (i-1) th are mainly the phrase components according to the (i-1) th phrase command. The contribution of the phrase command up to the (i-2) th order is small (FIG. 5).

From the above point of view, the following describes an embodiment in which the size of the i-th phrase command is obtained using the size of the (i-1) -th phrase command.

First, as in the first embodiment, the magnitude of the phrase component 1 / α after the time of the i-th phrase command is obtained. The magnitude of the phrase command at this point is obtained by searching Table 2 with the elapsed time from the time of the (i-1) -th phrase command and obtaining Gp (t)
Is obtained by multiplying by the magnitude of the (i-1) -th phrase command. This value is hereinafter referred to as c.

Next, when the phrase component has not been reached, Tp _i −c is obtained. Finally, the magnitude of the i-th phrase command is (Tp _i -c) × e / α in the case of the first embodiment.

[Third Embodiment] In the second embodiment, the phrase component unreached portion is obtained based on the size of the immediately preceding phrase command. However, the phrase component unachieved portion is determined based on the target value of the immediately preceding phrase command. The arrival can be approximated.

Further, the time interval from the immediately preceding phrase command can be counted by the number of beats.

 An embodiment considering these points will be described below.

Assuming that the utterance speed is m beats / sec, the time interval after I beat is I / m. It is assumed that a phrase command having a magnitude of Tp _b × e / α has occurred with respect to the target value Tp _b of the immediately preceding phrase command. I beat is not reached fraction d with respect to the target value Tp _n of the next phrase command accustomed _{d = Tp n -Tp b × e} / α × Gp (I / m + 1 /
α). Therefore, the size of the command of the next phrase is Ap
_n = d × e / α.

Table 3 shows the values of Ap _n when there are two types of phrase commands, 0.47 and 0.376, for various I values. When actually giving a phrase command, Ap _n
If _n is negative, Ap _n = 0. That is, no phrase command occurs. Since the phrase component due to the immediately preceding phrase command is gradually attenuated, if I> 20, it may be treated the same as I = 20.

Fourth Embodiment In the third embodiment, the size of the phrase command is (Tp _i −
c) Although the value of xe / α is used as it is, it is also possible to quantize the size of the phrase command in several steps.
That is, as the size of the phrase command, 0.43, 0.26,
0.12 (P1, P2, P3, respectively)
It is also possible to prepare and use a value closest to Ap _n .

Table 4 rewrites Table 3 from this viewpoint. According to this table, the size of the phrase command can be determined very easily.

[Fifth Embodiment] In the first embodiment, 1 / from the time of the i-th phrase command.
After α hours, the size c of the phrase component according to the (i-1) th phrase command was required. In terms of simplicity in processing, it is more convenient to perform processing with a value known up to now than to use a predicted value at a future time point.
Therefore, an embodiment in which the size c 'of the phrase component obtained at the time of the i-th phrase command is used instead of c will be described below.

First, at the time point To _i of the i-th phrase command, the value c ′ of the phrase component is obtained. Since the processing for obtaining the fundamental frequency pattern is performed for generating the voice waveform (see FIG. 2), the value c ′ of the phrase component may be extracted at the time point To _i (FIG. 6).

Next, when the phrase component has not been reached, Tp _i −c is obtained. Finally i th the size of a phrase command and if the in the same manner _{(Tp i -c ') × e} / α of the first embodiment.

〔The invention's effect〕

According to the present invention, the phrase command size setting process for preventing the phrase component from becoming too large can be included in the phrase component generation process. As a result, the process of correcting the size of the phrase command, which was conventionally required, is not required, and the rule for generating the phrase command is simplified, and the visibility is improved.

[Brief description of the drawings]

FIG. 1 is a conceptual diagram of a process for calculating the size of a phrase command according to the present method. FIG. 2 is a processing block diagram of a speech synthesizer. FIG. 3 is a Fo pattern (fundamental frequency pattern) generation model. FIGS. 5A and 5B are diagrams for explaining additional phrase components. FIG. 5 is a diagram for explaining the calculation of the size of a phrase component. FIG. 1 Control unit 1a Target unreachable calculation unit 2 Phrase control mechanism 3 Accent control mechanism 4 Glottal vibration mechanism

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Tsune Kawai 5-50-8 Maeharahigashi, Funabashi City, Chiba Prefecture (56) References JP-A-2-48700 (JP, A) JP-A 64-35599 (JP) , A) JP-A-64-28695 (JP, A) JP-A-63-259692 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) G10L 13/08 G10L 13/04 G10L 13/06

Claims (7)

(57) [Claims] 1. A fundamental frequency pattern generation device comprising a control unit, a phrase control mechanism, an accent control mechanism, and a glottal vibration mechanism, wherein the control unit performs first and second processing, and the first processing includes inputting A phrase command and an accent command each having a time and a size are determined from the characters / symbols to be performed, and the second process is to determine a maximum value of a phrase component generated by the i-th phrase command as a target value Tpi. , Target value T
The difference between pi and the phrase component up to the (i-1) th phrase component is determined as the unarrived portion, and the size of the i-th phrase command is determined based on the unarrived portion. At the time of the phrase command determined in the above, a phrase component is obtained based on the size of the phrase command determined in the second process. The accent control mechanism obtains an accent component from the accent command determined in the first process. The fundamental frequency pattern generating apparatus, wherein the glottal vibration mechanism obtains and outputs a fundamental frequency pattern by combining the phrase component and the accent component. 2. A fundamental frequency pattern generator comprising a control unit, a phrase control mechanism, an accent control mechanism, and a glottal vibration mechanism, wherein the control unit performs first and second processing, and the first processing includes inputting A phrase command and an accent command each having a time and a size are determined from the characters / symbols to be executed, and the second processing is to use the maximum value of the phrase component generated by the i-th phrase command as a first target value Tpi. Decide, i
The maximum value of the phrase component generated by the -1st phrase command is determined as the second target value Tpi _- 1, and the magnitude of the i-th phrase command is determined based on the first and second target values and their time intervals. The phrase control mechanism obtains a phrase component based on the phrase command determined in the first processing and the size of the phrase command determined in the second processing. The accent control mechanism determines A fundamental frequency pattern generating device, wherein an accent component is obtained from the accent command determined in the first processing, and the glottal vibration mechanism obtains and outputs a fundamental frequency pattern by combining the phrase component and the accent component. 3. The fundamental frequency according to claim 1, wherein the phrase control mechanism obtains the phrase component by an impulse response of a critical braking quadratic linear system of the phrase command. Pattern generator. 4. The equation of the impulse response of the critical damping quadratic linear system is: The difference between the target value of the i-th phrase command and the phrase component from the time point of the i-th phrase command to the (i-1) -th time point 1 / α after the time point is determined as the unarrived part. Item (3): The fundamental frequency pattern generator. 5. The fundamental frequency pattern generation device according to claim 1, wherein the control unit determines the magnitude of the phrase command as 0 when the unreached portion is negative. . 6. The apparatus according to claim 1, wherein the control unit determines the unarrived part based on the size of the (i-1) th phrase command and the time interval between the i-th and (i-1) th phrase commands. Item (1): The fundamental frequency pattern generator. 7. The controller according to claim 1, wherein the difference between the target value of the i-th phrase command and the phrase component up to the (i-1) -th phrase component at the time of the i-th phrase command is determined as an unreached portion. The fundamental frequency pattern generation device according to claim (1).