JP2002358090A - Speech synthesis method, speech synthesis device, and recording medium - Google Patents

Abstract

(57) [Summary] An object of the present invention is to provide a speech synthesis method, a speech synthesis device, and a recording medium that can change voice quality and the like flexibly while having good sound quality. A voice synthesis method, a voice synthesis device, and a recording medium for generating a voice signal by superimposing a pitch waveform in accordance with pitch period information, wherein the pitch waveform includes:
The approximate formant waveform is generated by summing a plurality of formant waveforms, and the approximate formant waveform is generated by applying a window function to a sine wave of the formant frequency.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to text-to-speech synthesis, and more particularly to speech synthesis for generating a speech signal from information such as a phoneme symbol string, pitch, and phoneme duration.

[0002]

2. Description of the Related Art Creating a speech signal artificially from an arbitrary sentence is called text-to-speech synthesis. Usually, this text-to-speech synthesis system includes three stages: a language processing unit, a phoneme processing unit, and a speech signal generation unit.

The input text is first subjected to morphological analysis and syntactic analysis in a language processing unit, and then to accent and intonation processing in a phonological processing unit to obtain a phonological symbol string and a pitch pattern (voice pitch). Information such as a change pattern and a phoneme duration time is output. Finally, the voice signal generation unit, that is, the voice synthesizer, synthesizes a voice signal from information such as a phoneme symbol string, a pitch pattern, and a phoneme duration.

[0004] The principle of a synthesizer capable of synthesizing an arbitrary phoneme symbol string is that if a vowel is represented by V and a consonant is represented by C, the characteristic parameters of small basic units such as CV, CVC, VCV, etc. A speech unit is stored, and the speech is synthesized by controlling the pitch and the duration of the connection.

[0005] In such a speech synthesizer, a PSOLA (Pitch-Synchronous) method is used as a method for generating a speech signal having a desired pitch pattern or duration from speech unit information.
ousOverlap-add) is well known. For example, the pitch cycle of a speech waveform stored as a speech unit is represented by PSO
Japanese Patent Laid-Open No. 8-202395, "Pitch conversion method and device"
Is disclosed.

FIG. 18 shows the principle of generating an output audio signal 104 by changing the pitch period of an input audio signal 101 using the PSOLA method. First, the input audio signal 101
The pitch period is obtained by performing pitch analysis. Then, a pitch waveform 103 is generated by applying a window function having a window length of about twice the pitch period to the input audio signal 101 at a position synchronized with the pitch. Next, an output audio signal 104 having a changed pitch cycle is generated by superposing the pitch waveforms 103 at a desired pitch cycle interval.

When the PSOLA method is applied to a speech synthesizer, an input speech signal 101 corresponds to a speech unit stored in advance, and an output speech signal 104 corresponds to a synthesized speech signal. It is known that when the degree of change in the pitch cycle of a synthesized voice by the PSOLA method is small, sound quality deterioration due to the change in the pitch cycle is small and the sound quality is good.

Another form of the speech synthesizer is a formant synthesis method. The formant synthesis method is a model that simulates a human vocal mechanism, and generates a voice signal by driving a filter that models vocal tract characteristics with a sound source signal that models a signal generated from a vocal cord. As an example, Japanese Patent Application Laid-Open No. 7-152396 "Speech synthesizer" discloses a speech synthesizer using a formant synthesis method.

FIG. 19 shows the principle of generating an audio signal by the formant synthesis method. Resonator 21, 2
A synthesized voice 208 is generated by driving a vocal tract filter composed of 2, 23 cascade connections with a pulse train 207 arranged at a desired pitch period interval. The frequency characteristic 204 of the resonator 21 is determined by the formant frequency Fl and the formant bandwidth Bl. Similarly, the frequency characteristic 205 of the resonator 22 is determined by the formant frequency F2 and the formant bandwidth B2, and the frequency characteristic 206 of the resonator 23 is determined by the formant frequency F3 and the formant bandwidth B3.

As described above, in the formant synthesis method, the phoneme (/ a /, / i /, / u /, etc.) and voice quality (male,
Female voice). Therefore, the speech unit information is not a waveform but a combination of the formant frequency and the bandwidth value. Since the formant synthesis method can control parameters directly related to phonemes and voice qualities, there is an advantage that flexible control such as changing voice qualities is possible.

[0011]

As described above, the PS
The OLA method has a problem that the sound quality is relatively good in a range where the change amount of the pitch period is small, but the sound quality is deteriorated when the change range is large.

[0012] In a voice uttered by a human, the spectral envelope of the same phoneme changes when the pitch period changes, whereas the PSOLA method cannot model this change, which causes deterioration. In addition, when the discontinuity of the spectrum occurs at the connection part of the speech unit, there is a problem that the distortion is generated in the spectrum by performing the smoothing processing, and the sound quality is deteriorated. Furthermore, since the waveform itself is used as a speech unit, it is difficult to change the voice quality, and lacks flexibility.

On the other hand, although the formant synthesis method is flexible, there is a problem that the accuracy of the model is poor. In other words, only the formant frequency and the bandwidth cannot express the fine structure of the spectrum of the actual audio signal, and the sound quality is poor and lacks real voice feeling (humanity).

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a speech synthesizer which has good sound quality and can flexibly change voice quality and the like.

[0015]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, a voice synthesizing method according to the present invention includes a voice synthesizing method for generating a voice signal by superimposing a pitch waveform according to pitch period information. A plurality of formant waveforms are generated by applying a window function to a sine wave, and the pitch waveform is generated by the sum of the plurality of formant waveforms.

Further, in the speech synthesizer of the present invention, a pitch pattern, a phoneme duration and a phoneme symbol string are inputted,
In a voice synthesizing apparatus for generating a voice signal by superimposing a pitch waveform formed by a pitch waveform generating section on a pitch mark generated in accordance with information on a pitch period, the pitch waveform generating section is provided for each speech unit. A storage unit in which formant parameters are stored, and by referring to the pitch pattern, the phoneme duration and the phoneme symbol string, the formant parameters for one frame corresponding to the pitch mark are selected from the storage unit. A parameter selecting unit for reading out the data, a sine wave generating unit for generating a sine wave having the read formant frequency, and a multiplier for generating a formant waveform by multiplying the generated sine wave by the selected window function. And an adder for adding each of these formants. It is intended.

According to the recording medium of the present invention, a program for realizing an audio synthesizing method for generating an audio signal by superimposing a pitch waveform according to pitch period information is recorded. Generating a plurality of formant waveforms by applying a function, and recording a program for realizing a voice synthesis method for generating the pitch waveform by summing up the plurality of formant waveforms. It is a feature.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a configuration of a speech synthesis apparatus for realizing a speech synthesis method according to an embodiment of the present invention. The voice synthesizer has a pitch pattern 30
6, a phoneme duration 307, a phoneme symbol string 308 is input, and a synthesized speech signal 305 is output. The voice synthesizing apparatus according to the present embodiment includes an unvoiced voice synthesis unit 32 and a voiced voice synthesis unit 31, and generates a synthesized voice signal 305 by adding the unvoiced voice signal 304 and the voiced voice signal 303 output from each of them.

The unvoiced sound synthesizer 32 has a phoneme duration 30
7 and the phoneme symbol string 308, an unvoiced speech signal 304 is generated mainly when the phoneme is an unvoiced consonant or voiced fricative. The unvoiced sound synthesis unit 32 can be realized by a known technique such as a method of driving an LPC synthesis filter with white noise.

The voiced sound synthesizer 31 comprises a pitch mark generator 33, a pitch waveform generator 34, and a waveform superimposing unit 35. The pitch mark generation unit 33 generates a pitch mark 302 as shown in FIG. 2 with reference to the pitch pattern 306 and the phoneme duration 307. The pitch mark 302 indicates the position where the pitch waveform 301 is superimposed, and the interval between the pitch marks corresponds to the pitch cycle.
The pitch waveform generation unit generates a pitch waveform 301 corresponding to each of the pitch marks 302 as shown in FIG. 2 with reference to the pitch pattern 306, the phoneme duration 307, and the phoneme symbol string 308. The waveform superimposing unit 35 places the corresponding pitch waveform 30 at the position indicated by the pitch mark 302.
A voiced speech signal 303 is generated by superimposing 1.

Next, the configuration of the pitch waveform generator of FIG. 1 will be described in detail.

FIG. 3 is a block diagram showing the configuration of one embodiment of the pitch waveform generator 34. Pitch waveform generator 34
Comprises a formant parameter storage unit 41, a parameter selection unit 42, and a sine wave generation unit (43, 44, 45). The formant parameter storage unit 41 stores formant parameters for each speech unit.

FIG. 4 shows an example of a formant parameter of a phoneme / a / fragment. In this example, the fragment of / a / is composed of three frames, and each frame is composed of three formants. A formant frequency, a formant phase, and a window function are stored as parameters representing the characteristics of each formant.

The formant parameter selection unit 42 refers to the pitch pattern 306, the phoneme duration 307, and the phoneme symbol string 308 input to the pitch waveform generation unit 34, and forms one formant parameter 401 corresponding to the pitch mark 302. Is selected from the formant parameter storage unit 41 and read out.

The formant parameter 401 is a parameter corresponding to the formant number 1 and the formant frequency 40
2, the formant phase 403 and the window function 411 are output. Similarly, the parameters corresponding to the formant number 2 are the formant frequency 404, the formant phase 405, and the window function 412, and the parameters corresponding to the formant number 3 are the formant frequency 406. Formant phase 4
07, output as a window function 413.

The sine wave generator 43 has a formant frequency 40
2 and outputs a sine wave 408 according to the formant phase 403. The sine wave 408 is windowed by a window function 411 to generate a formant waveform 414. When the formant frequency 402 is represented by ω, the formant phase 403 is represented by φ, and the window function 411 is represented by w (t), the formant waveform y (t)
Is represented by the following equation.

Y (t): ΔV (t) · sin (ωt + φ) Similarly, the sine wave generator 44 outputs a sine wave 409 according to the formant frequency 404 and the formant phase 405,
Formant waveform 4 after windowing by window function 412
15 is generated. The sine wave generator 45 generates a sine wave 410 according to the formant frequency 406 and the formant phase 407.
Is output, and a formant waveform 416 is generated through windowing processing by a window function 413.

Further, the pitch waveform 301 is generated by adding the formant waveforms (414, 415, 416).

FIG. 6 shows an example of a sine wave, a window function, a formant waveform, and a pitch waveform. FIG. 7 shows the power spectra of these waveforms. 6, the horizontal axis represents time, and the vertical axis represents amplitude. In FIG. 7, the horizontal axis represents frequency, and the vertical axis represents amplitude.

The sine wave is a line spectrum having a sharp peak, and the window function is a spectrum concentrated in a low band. Since windowing (multiplication) in the time domain corresponds to convolution in the frequency domain, the spectrum of the formant waveform has a shape obtained by translating the spectrum of the window function to the position of the frequency of the sine wave. Therefore, the center frequency and phase of the pitch waveform formant can be changed by controlling the frequency and phase of the sine wave, and the form shape of the pitch waveform can be changed by controlling the shape of the window function. it can.

As described above, since the center frequency, phase, and spectrum shape of each formant can be controlled independently, it can be said that this is a highly flexible model. At the same time, since the fine structure of the spectrum can be expressed by the shape of the window function, the spectrum structure of the real voice can be approximated with high accuracy, and a voice with a real voice feeling can be synthesized.

Next, a second embodiment of the pitch waveform generator 34 of the present invention will be described with reference to FIG. The parts corresponding to those in FIG. 3 are denoted by the same reference numerals and mainly the differences will be described. In the present embodiment, the window function is expanded to a basis function, and instead of storing the window function as a formant parameter, a weighting coefficient is used. Are stored. Then, the newly added window function generator 56 generates a window function from the set of weighting coefficients.

FIG. 5 shows an example of the formant parameters stored in the formant parameter storage unit 51. In this example, a window function is developed into a weighted sum of three basis functions, and a set of three coefficients is stored as a set of window function weighting coefficients. The parameter selection unit 42 selects the formant frequency (4 () 2, 404, 406) and the formant phase (40) among the selected formant parameters (formant frequency, formant phase, window function weighting coefficient) 501.
3, 405, 407) are sine wave generators (43, 44, 4).
5), a window function weighting coefficient set (517, 518, 51)
9) is output to the window function generator 56.

The window function generating section 56 has a weighting coefficient set (5
17, 518, 519), the window function (51
1, 512, 513) respectively. The weight coefficient sets are a1, a2, and a3, respectively, and the basis function is b1
(T), b2 (t), b3 (t), the window function W
(T) is represented by the following equation.

W (t) = a1 · b1 (t) tens a2 · b2
(T) 10 a3 · b3 (t) The basis function is DC
A T-basis or the like may be used, or a basis function generated by performing KL expansion of a window function may be used. In the present embodiment, the order of the basis is set to 3, but the order may be any number. By expanding the window function into the basis function, there is an advantage that the storage capacity of the formant parameter storage unit is reduced.

Next, a third embodiment of the pitch waveform generator 34 of the present invention will be described with reference to FIG. 3 will be described with the same reference numerals given to the parts corresponding to FIG. 3. In this embodiment, a parameter deforming unit 67 is newly added, and the formant parameters change according to the pitch pattern 306. The points are different.

The parameter deforming section 67 includes a formant frequency 402, a formant phase 403, a window function 411, a formant frequency 404, a formant phase 405, and a window function 4
12, formant frequency 406, formant phase 40
7. The window function 413 is changed in accordance with the pitch pattern 306 so that the formant frequency 720 and the formant phase 7
21, window function 717, formant frequency 722, formant phase 723, window function 718, formant frequency 72
4. Output formant phase 725 and window function 719. All parameters may be changed, or only some parameters may be changed.

FIG. 10 shows an example of a control function for controlling the formant frequency according to the pitch period. Such a control function may be set for each phoneme, or may be set and used for each frame or formant number.

Further, instead of the formant frequency itself, a control function for controlling a difference value or a ratio value between the input formant frequency and the output formant frequency may be used.

FIG. 11 shows a control function for controlling the power of the formant by multiplying the window function by a gain corresponding to the pitch period. Thus, by changing the parameters according to the pitch cycle, it is possible to model the change in the spectrum of the voice due to the change in the pitch cycle, and to generate a synthesized voice with high sound quality regardless of the pitch of the voice. Can be.

Alternatively, the phoneme symbol string 308 may be input to the parameter transformation section 67 so that the formant parameters are changed according to the type of the preceding or succeeding phoneme. Accordingly, it is possible to model a change in the spectrum of the voice due to the phoneme environment, and it is possible to improve the sound quality.

Further, the parameters may be changed in accordance with the voice quality information 309 input from the outside to the parameter deforming section 67. As a result, it is possible to generate synthesized voices of various voice qualities.

FIG. 12 shows an example of a control function when the thickness of the voice is changed by changing the formant frequency. If all the formant frequencies are converted using the control function of (a), a thin voice is generated by shifting the formant to a high frequency band, and in the case of (b), the voice becomes slightly thin. Conversely, when the control function of (d) is used, a thick voice is generated by shifting the formant frequency to a low band, and in the case of (c), the voice becomes slightly thick.

Next, a fourth embodiment of the pitch waveform generator 34 of the present invention will be described with reference to FIG. The parts corresponding to those in FIG. 3 are denoted by the same reference numerals and mainly the differences will be described. In the present embodiment, a parameter smoothing unit 77 is newly added, and the temporal change of each formant parameter is changed. The difference is that the parameters are smoothed so as to be smooth.

The parameter smoothing section 77 includes a formant frequency 402, a formant phase 403, a window function 411, a formant frequency 404, a formant phase 405, a window function 412, a formant frequency 406, and a formant phase 40.
7. The window function 413 is smoothed to formant frequency 820, formant phase 821, window function 817,
Formant frequency 822, formant phase 823, window function 818, formant frequency 824, formant phase 8
25, and outputs a window function 819, respectively. All parameters may be smoothed, or only some parameters may be smoothed.

FIG. 14 shows an example of formant smoothing. × represents the formant frequency 4 before smoothing
02, 404, and 406. The smoothed formant frequencies 820, 822, and 824 represented by ○ are smoothed so that the change from the corresponding formant frequency of the preceding or subsequent frame is smooth. Generated.

In the case where the formant cannot be handled at the connection part of the speech unit, the X in FIG.
, The formant corresponding to the formant frequency 404 may disappear. in this case,
Since a large discontinuity occurs in the spectrum and the sound quality is degraded, a formant is added to generate a formant frequency 822 as indicated by a circle. At this time, the window function 8 corresponding to the formant frequency 822 as shown in FIG.
The power of 18 is attenuated so that the discontinuity of the formant power does not occur.

FIG. 16 shows an example of smoothing the window function position. A window function 817 is generated by smoothing the window function position so that the peak position of the window function 411 changes smoothly between frames. In addition, the shape of the window function and the power of the window function may be smoothed.

In the above-described embodiment of the present invention, the case where the number of formants is 3 has been described. However, the number of formants may be any number, and the number of formants may change for each frame.

Although the sine wave generator according to the embodiment of the present invention outputs a sine wave, the waveform may not be a perfect sine wave as long as the waveform has a power spectrum close to a line spectrum. For example, when the calculation accuracy is lowered for the purpose of reducing the calculation amount, or when a table is formed, a perfect sine wave may not be obtained due to an error.

The spectrum of the formant waveform does not always represent the peak of the spectrum of the audio signal, and the spectrum of the pitch waveform, which is the sum of a plurality of formant waveforms, represents the spectrum of the audio.

Although the synthesizer for text-to-speech synthesis has been described as an embodiment of the present invention, there is a decoder for speech coding as another embodiment of the present invention. That is, the encoder determines formant parameters such as formant frequency, formant phase, and window function and the pitch period from the audio signal by analysis, encodes them and transmits or stores them, and the decoder encodes the formant parameters and the pitch period. Can be decoded to reproduce an audio signal in the same manner as the synthesizer described above.

The above-described speech synthesis can be performed by program-controlling a computer according to a program stored in a recording medium. This program control will be described with reference to FIG.

FIG. 17A shows a flowchart of the voice synthesis process, and FIG. 17B shows a flowchart of the voiced voice generation process in the voice synthesis process.
FIG. 7C shows a flowchart of the pitch waveform generation processing of the voiced voice generation processing of FIG. 17B.

In the speech synthesis processing in FIG. 17A, a pitch pattern 306, a phoneme duration 307, and a phoneme symbol string 308 are input (S11). A voiced speech signal 303 is generated based on the pitch pattern 306, phoneme duration 307, and phoneme symbol string 308 (S1).
2). An unvoiced speech signal 304 is generated with reference to the phoneme duration 307 and phoneme symbol string 308 (S13). A synthesized voice signal 3 is obtained by adding the voiced voice signal and the unvoiced voice signal.
05 is generated (S14).

In the voiced voice generation processing in FIG. 17B, the pitch mark 302 is generated with reference to the pitch pattern 306 and the phoneme duration 307 (S21). A pitch waveform 301 corresponding to the pitch mark 302 is generated with reference to the pitch pattern 306, the phoneme duration 307, and the phoneme symbol string 308 (S22). Pitch waveform 3 corresponding to the position indicated by pitch mark 302
01 is superimposed to generate voiced speech (S23).

In the pitch waveform generation processing in FIG. 17C, the pitch pattern 306 and the phoneme duration 3
07 and the phonological symbol string 308 with reference to the pitch mark 3
Formant parameter 4 for one frame corresponding to 02
01 is selected from the formant parameter storage unit 41 (S31). A plurality of sine waves are generated according to the formant frequency and formant phase corresponding to the formant number of the selected formant parameter 401 (S32). A plurality of sine waves are windowed by a window function to generate formant waveforms 414, 415, 416 (S33). A pitch waveform is generated by adding these formant waveforms (S
34).

[0058]

As described above, according to the present invention, since the formant frequency and the formant shape are controlled independently for each formant, it is possible to express a change in the spectrum of a voice due to a difference in pitch period or voice quality. High flexibility can be realized. Alternatively, since the fine structure of the formant spectrum is expressed by the shape of the window function, a high-quality synthesized sound having a real voice feeling can be generated.

[Brief description of the drawings]

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

FIG. 2 is a schematic diagram showing generation of voiced speech by superposition of a pitch waveform.

FIG. 3 is a block diagram of a pitch waveform generator according to one embodiment of the present invention.

FIG. 4 is a schematic diagram showing an example of a formant parameter.

FIG. 5 is a schematic diagram showing an example of a formant parameter.

FIG. 6 is a schematic diagram showing an example of a sine wave, a window function, a formant waveform, and a pitch waveform.

FIG. 7 is a schematic diagram showing an example of a power spectrum of a sine wave, a window function, a formant waveform, and a pitch waveform.

FIG. 8 is a block diagram of a pitch waveform generator according to an embodiment of the present invention.

FIG. 9 is a block diagram of a pitch waveform generator according to an embodiment of the present invention.

FIG. 10 is a schematic diagram illustrating an example of a control function of a formant frequency.

FIG. 11 is a schematic diagram showing an example of a control function of a formant gain.

FIG. 12 is a schematic diagram showing an example of a formant frequency mapping function for voice quality conversion;

FIG. 13 is a block diagram of a pitch waveform generator according to one embodiment of the present invention.

FIG. 14 is a schematic diagram showing an example of smoothing a formant frequency.

FIG. 15 is a schematic diagram showing an example of smoothing a formant frequency.

FIG. 16 is a schematic diagram showing an example of smoothing a window function position.

FIG. 17 is a flowchart showing processing of the speech synthesizer of the present invention.

FIG. 18 is a schematic diagram showing speech synthesis by a conventional PSOLA method.

FIG. 19 is a block diagram of a conventional formant synthesizer.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 31 ... Voiced sound synthesis part 32 ... Unvoiced sound synthesis part 33 ... Pitch mark generation part 34 ... Pitch waveform generation part 35 ... Waveform superposition part 41, 51 ... Formant parameter storage part 42 ... Parameter selection part 43, 44, 45 ... Sine wave generation Unit 56: window function generating unit 67: parameter deforming unit 77: parameter smoothing unit

Claims (17)

[Claims] In a voice synthesizing method for generating a voice signal by superimposing a pitch waveform according to pitch period information, a plurality of formant waveforms are generated by applying a window function to a sine wave of a formant frequency. Wherein the pitch waveform is generated by the sum of the formant waveforms. 2. The speech synthesis method according to claim 1, wherein said window function is generated by weighted addition of a plurality of basis functions. 3. The voice according to claim 1, wherein at least one of the power of the formant waveform, the shape of the window function, the position of the window function, and the formant frequency changes in accordance with the pitch period. Synthesis method. 4. The apparatus according to claim 1, wherein at least one of the power of the formant waveform, the shape of the window function, the position of the window function, and the formant frequency changes in accordance with at least a type of a preceding or succeeding phoneme. Item 1. The speech synthesis method according to Item 1. 5. The apparatus according to claim 1, wherein at least one of the power of the formant waveform, the shape of the window function, the position of the window function, and the formant frequency changes in accordance with given voice quality information. Described speech synthesis method. 6. The method according to claim 6, wherein at least one of the formant frequency, the power of the formant waveform, the shape of the window function, the phase of the sine wave, and the position of the window function is at least one of a corresponding formant of a preceding or succeeding pitch waveform. 2. The speech synthesis method according to claim 1, wherein the voice synthesis method changes in accordance with at least one of a formant frequency, a power of a formant waveform, a shape of a window function, a phase of a sine wave, and a position of the window function. 7. At least one of the formant frequency, the power of the formant waveform, the shape of the window function, the phase of the sine wave, and the position of the window function is the presence or absence of a corresponding formant of at least a preceding or succeeding pitch waveform. 2. The speech synthesis method according to claim 1, wherein the speech synthesis method changes according to the following. 8. A voice signal is inputted by inputting a pitch pattern, a phoneme duration and a phoneme symbol string, and superimposing a pitch waveform formed by a pitch waveform generator on a pitch mark generated according to pitch period information. In the speech synthesizer that generates, the pitch waveform generating unit, a storage unit in which formant parameters are stored for each unit of the speech unit, and the pitch pattern, the phoneme duration and the phoneme symbol string as a reference, A parameter selection unit that selects and reads the formant parameters for one frame corresponding to the pitch mark from the storage unit, a sine wave generation unit that generates a sine wave of the read formant frequency, A multiplier for generating a formant waveform by multiplying the sine wave by the selected window function; Speech synthesis apparatus characterized by comprising an adder for adding the mantle, respectively. 9. A speech synthesizer according to claim 8, wherein said window function is stored in said storage unit. 10. A window function generating unit for storing a weighting factor of a window function in the storage unit, and generating the window function by introducing the weighting factor and adding weights of basis functions. The speech synthesizer according to claim 8, wherein 11. A speech synthesizer according to claim 8, further comprising a parameter deforming section for changing said selected formant parameter in accordance with said pitch period. 12. A speech synthesizer according to claim 8, further comprising a parameter transforming unit for changing said selected formant parameter in accordance with preceding or succeeding phoneme information. 13. A speech synthesizer according to claim 8, further comprising a parameter deforming unit for changing said selected formant parameter according to a given voice quality. 14. The speech synthesizer according to claim 8, further comprising a parameter smoothing unit for smoothing a temporal change of the selected formant parameter. 15. A recording medium on which a program for realizing a voice synthesizing method for generating a voice signal by superimposing a pitch waveform in accordance with pitch period information is recorded, by applying a window function to a sine wave of formant frequency. A recording medium recording a voice synthesis method, wherein a program for generating a formant waveform and a voice synthesis method for generating the pitch waveform by a sum of the plurality of formant waveforms is recorded. 16. A predetermined formant parameter is selected from said formant parameters according to a command for storing a plurality of formant parameters representing a formant frequency, a formant phase and a window function in a storage device, and a pitch pattern, a phoneme duration and a phoneme symbol string. Instructions to generate a plurality of sine waves based on the formant frequency and formant phase corresponding to the selected formant parameter, and a window function corresponding to the selected formant parameter to generate a plurality of formant waveforms. Speech synthesis including a command to multiply by the sine wave, a command to add the formant waveform to generate a plurality of pitch waveforms, and a command to superimpose the pitch waveform according to a pitch period to generate a voice signal program. 17. The speech synthesis program according to claim 16, further comprising an instruction to add a prescribed function weighted by a weight coefficient to generate the window function.