JP6347536B2 - Sound synthesis method and sound synthesizer - Google Patents

Description

  The present invention relates to a sound synthesis method and a sound synthesis apparatus.

  The sound synthesizing method and the sound synthesizing apparatus of the present invention extend the voice analysis / synthesis method used as a substrate technology when converting the tone color of a voice signal. In many cases, a speech analysis / synthesis method changes in three times: a fundamental frequency that is information related to voice pitch, a spectrum envelope that represents information of the vocal tract, and voiced / unvoiced determination information that represents whether voiced or voiced. Information is analyzed and extracted from the speech waveform, and a synthesized sound is generated from the information.

  The spectrum envelope is obtained by removing detailed fluctuation information that changes in the frequency direction related to the voice pitch from the amplitude spectrum obtained in the analysis of the voice signal. The spectral envelope is deeply related to the tone color information. That is, in speech, the spectrum envelope is closely related to vocal tract information, and includes a lot of speaker information indicating who is speaking and phonological information indicating what is being spoken.

  Since the speech analysis and synthesis method separates and expresses vocal tract information from a speech signal, it is easy to control the vocal tract. However, the speech analysis and synthesis method hardly includes information on the phase spectrum that is paired with the amplitude spectrum in the information included in the speech waveform. The phase spectrum contains a lot of noise information in vocal cord vibration. Therefore, the speech analysis and synthesis method generates a synthesized sound by replacing the phase spectrum with degenerate information called voiced / unvoiced determination information. In this case, if vocal cord vibration like an announcer is regular voice, there is little problem with such degenerate information, but vocal cord vibration such as a husky voice or rattle voice inputs special voice. When re-synthesized, the synthesized sound deteriorates.

  Furthermore, in recent years, in a singing voice synthesizing system, there is an increasing expectation for realizing a system that can reproduce a shout method, a scream method, and the like in which a singer actively controls vibration of a vocal cord. These voices are also very special in vocal cord vibration, and it is difficult to reproduce with high quality by the existing voice analysis and synthesis method. There are cases where the vocal waveform itself is recorded in large quantities and the method of using it is used, but only the recorded one can be reproduced, so it is impossible to reproduce the vocal cord vibration desired by the user Have difficulty.

  Therefore, Non-Patent Document 1 discloses a conventional speech analysis and synthesis method that can re-synthesize speech having strong noisy vocal cord vibrations with high quality or enhance noise. This speech analysis and synthesis method can enhance the noise characteristics of vocal cord vibrations by extracting a group delay spectrum having information equivalent to the phase spectrum from the speech signal and amplifying the value.

  Patent Document 1 discloses a conventional speech synthesis method. This speech synthesis method changes the information of the phase spectrum.

JP-A-10-97287

Hideki Sakano, Rikukinbayashi, Satoshi Nakamura, Kiyohiro Shikano, Hideki Kawahara, "Voice quality control method using phase control by time domain smoothing group delay", IEICE Transactions, D-II, Vol. J83-D-II, no. 11, pp. 2276-2282, November 2000

  However, the speech analysis and synthesis method of Non-Patent Document 1 requires a pre-processing called pitch marking in order to stably extract the group delay spectrum. Since pitch marking is an automatic process with many errors, if many errors occur, high-quality speech synthesis cannot be performed without manual correction. Further, the speech synthesis method of Patent Document 1 does not use the phase spectrum information of the input speech, but simply changes the phase spectrum information.

  The present invention has been made in view of the above-described conventional situation, and it is an object to be solved to provide a sound synthesis method and a sound synthesis apparatus that can easily generate high-quality synthesized sound.

The sound synthesis method of the present invention is a sound synthesis method for generating a synthesized sound,
Predetermined spectral kurtosis that band division or based on the correspondence between the value and the variation amount of the group delay spectrum of the spectral flatness, to correspond to any value of kurtosis or spectral flatness of spectrum the band division A first step of generating a variation amount of the group delay spectrum,
A second step of generating a group delay spectrum using the variation amount of the group delay spectrum generated in the first step;
A third step of calculating an integral or cumulative sum of the group delay spectrum generated in the second step and converting it into a phase spectrum;
A complex spectrum is obtained by combining an arbitrary amplitude spectrum or an arbitrary spectral envelope and the phase spectrum converted in the third step, and the complex spectrum is subjected to inverse Fourier transform to obtain a one-pitch waveform which is a signal for one period. A fourth step of generating
A fifth step of generating a synthesized sound by superimposing and adding the 1 pitch waveform generated in the fourth step ;
With
The second to fifth steps are repeated to generate a synthesized sound having a frame length for synthesis.

The sound synthesizer of the present invention is a sound synthesizer that generates a synthesized sound,
A storage unit storing a correspondence relationship between a kurtosis or spectrum flatness value of a predetermined band-divided spectrum and a variation amount of the group delay spectrum;
Based on the correspondence relationship between the kurtosis or spectrum flatness value of the band-divided spectrum stored in the storage unit and the variation amount of the group delay spectrum, the kurtosis or spectrum flatness of the band-divided spectrum is calculated. A group delay variation generating unit for generating a group delay spectrum variation corresponding to an arbitrary value;
A group delay generation unit that generates a group delay spectrum using the variation amount of the group delay spectrum generated by the group delay variation generation unit;
A phase generation unit that calculates an integral or cumulative sum of the group delay spectrum generated by the group delay generation unit and converts it into a phase spectrum; and
A complex spectrum is obtained by combining an arbitrary amplitude spectrum or an arbitrary spectral envelope and the phase spectrum converted by the phase generation unit, and the complex spectrum is subjected to inverse Fourier transform to obtain a one-pitch waveform which is a signal for one period. A one-pitch waveform generating unit for generating
A superposition addition unit that superimposes and adds the one pitch waveform generated by the one pitch waveform generation unit to generate a synthesized sound ;
It is characterized by having.

This sound synthesizing method and sound synthesizing apparatus generates a synthesized sound by using the kurtosis or spectrum flatness of a spectrum obtained by dividing a band corresponding to the variation amount of the group delay spectrum. Since the kurtosis or spectral flatness value of the band-divided spectrum has a corresponding relationship with the fluctuation amount of the group delay spectrum, it is determined in advance and corresponds to any value of the kurtosis or spectral flatness of the band-divided spectrum Thus, the fluctuation amount of the group delay spectrum can be generated. This does not faithfully reproduce the group delay spectrum but reproduces the degree of fluctuation of the group delay spectrum in each band, thereby enabling generation of a synthesized sound having noise characteristics.

  Therefore, the sound synthesis method and sound synthesis apparatus of the present invention can easily generate high-quality synthesized sound.

1 is a block diagram illustrating a sound synthesizer according to a first embodiment. 3 is a flowchart illustrating a sound synthesis method according to the first embodiment. It is a graph which shows the value of kurtosis. It is a graph which shows the variation | change_quantity of a group delay spectrum. It is a graph which shows an index-group delay variation amount correspondence. It is a graph which shows a group delay spectrum. It is a graph which shows a phase spectrum. It is a graph which shows a spectrum envelope. It is a graph which shows 1 pitch waveform. It is a graph which shows a synthetic sound. It is a block diagram which shows the sound synthesizer of Example 2. 6 is a flowchart illustrating a sound synthesis method according to the second embodiment.

  A preferred embodiment of the present invention will be described.

In the sound synthesis method of the present invention, the arbitrary value of the kurtosis or spectrum flatness of the band-divided spectrum can be extracted from the analysis signal having the frame length of the time length set for the input sound signal. In this case, it is easier to analyze the input sound signal to extract the kurtosis or spectrum flatness of the band-divided spectrum that can be extracted more easily than extracting the group delay spectrum from the input sound signal. It can be carried out.

  In the sound synthesis method of the present invention, the arbitrary amplitude spectrum or the arbitrary spectrum envelope can be extracted from the analysis signal having the frame length of the time length set for the input sound signal. In this case, a synthesized sound closer to the input sound signal can be generated by using the amplitude spectrum or spectrum envelope extracted from the analysis signal in obtaining the complex spectrum in combination with the phase spectrum.

  In the sound synthesis method of the present invention, the group delay spectrum generated in the second step can be generated by multiplying a variation amount of the group delay spectrum by a predetermined coefficient. In this case, the noise characteristic of the synthesized sound can be amplified or attenuated by a predetermined coefficient that is multiplied by the fluctuation amount of the group delay spectrum.

  In the sound synthesis method of the present invention, the group delay spectrum generated in the second step may be generated by multiplying the variation amount of the group delay spectrum by a random number. In this case, the noise characteristic of the synthesized sound can be amplified or attenuated by a random number multiplied by the variation amount of the group delay spectrum, and the noise feeling of the synthesized sound can be generated better.

In the sound synthesizer of the present invention, an analysis signal extraction unit that extracts an analysis signal for each frame length of a time length set for an input sound signal, and the analysis signal extraction unit that extracts the analysis signal An index extraction unit that extracts a value of spectrum kurtosis or spectrum flatness obtained by band-dividing from the analysis signal. In this case, the kurtosis or spectral flatness value of the spectrum obtained by dividing the band extracted from the analysis signal can be extracted more easily than the group delay spectrum extracted from the analysis signal. Can be easily analyzed.

  In the sound synthesizer of the present invention, an analysis signal extraction unit that extracts an analysis signal for each frame length of a time length set for an input sound signal, and the analysis signal extraction unit that extracts the analysis signal And a spectrum extraction unit that extracts the amplitude spectrum or the spectrum envelope from the analysis signal. In this case, the analysis signal is extracted from the sound signal input by the analysis signal extraction unit, and the amplitude spectrum or spectrum envelope is extracted from the analysis signal by the spectrum extraction unit. As a result, it is possible to use an amplitude spectrum or spectrum envelope extracted from a signal for analysis when a complex spectrum is obtained in combination with a phase spectrum, and a synthesized sound closer to the input sound signal can be generated.

  Next, Embodiments 1 and 2 embodying the sound synthesis method and sound synthesis apparatus of the present invention will be described with reference to the drawings.

<Example 1>
As shown in FIG. 1, the sound synthesizer according to the first embodiment includes an analysis unit 10 and a synthesis unit 20. The analysis unit 10 includes an analysis signal extraction unit 11, a spectrum extraction unit 12, a fundamental frequency extraction unit 13, and an index extraction unit 14. The synthesizing unit 20 includes a group delay variation generating unit 21, a group delay generating unit 22, a phase generating unit 23, a 1 pitch waveform generating unit 24, a superposition adding unit 25, and a storage unit 26. In the sound synthesizing method using this sound synthesizer, the sound signal input to the sound synthesizer is analyzed by the analyzer 10, and the synthesized sound is generated by the synthesizer 20 based on information obtained by the analysis.

  As shown in FIG. 2, in the sound synthesis method using this sound synthesizer, first, the analysis signal extraction unit 11 sets the time length set from the analysis start point for the sound signal input to the sound synthesizer. The signal for analysis of the frame length is extracted (step S1). If necessary, the extracted analysis signal is multiplied by an analysis window. In the following, it is assumed that the frame number in this frame is m.

Next, the index extraction unit 14 extracts the kurtosis value K _m (f) of the band-divided spectrum, which is an index representing periodicity, from the analysis signal extracted by the analysis signal extraction unit 11 (step S2). ). As shown in FIG. 3, the kurtosis value K _m (f) of the band-divided spectrum depends on the frequency f and corresponds to the fluctuation amount W _m (f) of the group delay spectrum. Since the kurtosis value K _m (f) of the spectrum obtained by the band division can be easily and stably extracted from the sound signal, the input sound signal can be easily analyzed. Further, by using the kurtosis value K _m (f) of the spectrum obtained by band division that changes according to the frequency, a high-quality synthesized sound can be generated.

Next, the group delay variation generating unit 21 executes a first step of generating the group delay spectrum variation Wm (f) (step S3). The variation amount Wm (f) of the group delay spectrum generated in the first step is shown in FIG. The variation amount Wm (f) of the group delay spectrum is generated based on “index-group delay variation correspondence information” (see FIG. 5) accumulated in the storage unit 26. The correspondence relationship between the kurtosis value Km (f) of the band-divided spectrum and the group delay spectrum fluctuation amount Wm (f) is experimentally determined in advance and stored in the storage unit 26 as “index-group delay fluctuation amount correspondence”. Information "is accumulated. The function Ψ representing the correspondence between the index kurtosis value Km (f) of the band-divided spectrum and the group delay spectrum variation Wm (f) is used as the index, and the kurtosis value Km ( If it depends on f), it can be expressed as Wm (f) = Ψ (f, Km (f)).

Here, first, a kurtosis value Km (f) of a spectrum obtained by band-dividing from a signal created by artificially giving a variation amount Wm (f) of the group delay spectrum is observed, and these relationships are based on a sigmoid function. It approximated by Formula 1.

Here, b, c, and d are determined so that the kurtosis value Km (f) of the spectrum obtained by dividing the band observed from the experimental data and the group delay spectrum variation Wm (f) are best correlated. Constant. Also, a (f) is a function determined so that the kurtosis value Km (f) of the band-divided spectrum and the variation amount Wm (f) of the group delay spectrum are best correlated, for example, a sigmoid function Equation 2 based on can be used.

Here, similarly to b, c, and d, p and q have the best kurtosis value Km (f) of the band-divided spectrum observed from the actual data and the variation Wm (f) of the group delay spectrum. It is a constant determined to be associated. Then, Ψ (f, W) is obtained by using Ψ−1 (f, W) and obtaining an inverse function with respect to W. Expression 1 and Expression 2 become Expression 3. Note that a (f) may be a constant independent of f as necessary.

Next, the group delay generation unit 22 uses the group delay spectrum variation W _m (f) generated in the first step to use the group delay spectrum D _n (f ) Is generated (step S4). FIG. 6 shows the group delay spectrum D _n (f) generated in the second step. This group delay spectrum D _n (f) only needs to have a fluctuation amount depending on W _m (f), and a random number generator is used to calculate a random number N _n (f) having an average value of 0 and a variance of 1. It is generated and multiplied by the variation amount W _m (f) of the group delay spectrum. When amplifying or attenuating the noise characteristics of the synthesized sound, this weight may be multiplied by a non-zero coefficient α. In this case, the generated group delay spectrum D _n (f) is expressed as D _n (f) = αW _m (f) N _n (f). In this way, the noise of the synthesized sound can be generated satisfactorily by multiplying by the random number N _n (f).

Next, the phase generation unit 23 executes a third step of integrating the group delay spectrum D _n (f) generated in the second step and converting it to the phase spectrum θ _n (f) (step S5). FIG. 7 shows the phase spectrum θ _n (f) converted in the third step. Further, this conversion is expressed by Equation 4. The phase spectrum θ _n (f) may be subjected to other modifications such as a modification for reproducing a positional shift according to the value of the fundamental frequency.

Next, the 1-pitch waveform generation unit 24 combines the spectrum envelope A _m (f) shown in FIG. 8 with the phase spectrum θ _n (f) generated in the third step to generate a complex spectrum Y _n (f). Ask. The spectrum envelope A _m (f) is extracted from the analysis signal by the spectrum extraction unit 12 of the analysis unit 10 (step S6-1). For this reason, a synthesized sound closer to the input sound signal can be generated. The complex spectrum Y _n (f) is expressed in Equation 5. The fourth step of performing inverse Fourier transform on the obtained complex spectrum Y _n (f) to generate a signal (one pitch waveform) y _n (t) for one period shown in FIG. 9 is executed (step S6).

  Next, the 1-pitch waveform generated in the fourth step is superimposed and added using the fundamental frequency value (step S7-1) extracted from the analysis signal in the fundamental frequency extraction unit 13 of the analysis unit 10 to generate a synthesized sound. A fifth step is executed (step S7). The addition start position is updated based on the basic period. The second to fifth steps are repeated until the analysis frame needs to be updated.

When the analysis frame needs to be updated (step S8), the analysis start point is updated, the analysis signal extraction unit 11 extracts the next analysis signal (step S1), and the above-described processes are executed. Assuming that the addition start position in the synthesis frame n is t _n , the synthesized sound s _m (t) after the process from the second process to the fifth process is converted to the synthesized sound s _m-1 (t) before the repeated process. And is expressed as Equation 6. Here, n _m represents the first frame number of synthetic frames in the analysis frame m. N _m represents the number of synthesis repetitions in the analysis frame m.

The synthesized sound generated in this way is shown in FIG. This sound synthesizing method and sound synthesizing apparatus generates a synthesized sound by using the kurtosis of the spectrum obtained by band division as an index representing periodicity corresponding to the variation amount of the group delay spectrum. Since the kurtosis value Km (f) of the band-divided spectrum has a corresponding relationship with the fluctuation amount Wm (f) of the group delay spectrum, it is determined in advance and the kurtosis of the band-divided spectrum extracted from the analysis signal. The group delay spectrum variation Wm (f) can be generated in correspondence with the value Km (f). This does not faithfully reproduce the group delay spectrum but reproduces the degree of fluctuation of the group delay spectrum in each band, thereby enabling generation of a synthesized sound having noise characteristics.

  Therefore, the sound synthesizing method and sound synthesizing apparatus according to the first embodiment can easily generate a synthesized sound having noise characteristics.

<Example 2>
As shown in FIG. 11, the sound synthesizer of the second embodiment includes a linear prediction analysis unit 15 and a linear prediction residual extraction unit 16 in the analysis unit 110, and a residual drive synthesis unit 27 in the synthesis unit 120. This is different from the first embodiment. Other configurations are the same as those of the first embodiment, and the same configurations are denoted by the same reference numerals and detailed description thereof is omitted.

As shown in FIG. 12, a sound synthesis method using this sound synthesizer uses a linear prediction residual drive type analysis and synthesis method. That is, the 1-pitch waveform generation unit 24 generates the amplitude spectrum A _m (f) (step S6-2) of the linear prediction residual signal extracted by the linear prediction residual extraction unit 16 and the third step (step S5). A complex spectrum Y _n (f) is obtained by combining with the phase spectrum θ _n (f), and a fourth step of generating a signal (one pitch waveform) for one period by performing an inverse Fourier transform is executed (step) S6).

  Thereafter, the superimposing / adding unit 25 executes a fifth step of generating a synthesized sound (step S7) and uses it as a linear prediction residual signal to be given to the linear prediction residual drive type analysis / synthesis method. Then, the residual drive synthesis unit 27 uses the linear prediction coefficient (step S9-1) extracted by the linear prediction analysis unit 15 for each analysis frame, and drives with this linear prediction residual signal to generate a synthesized sound ( Step S9).

This sound synthesizing method and sound synthesizing apparatus also generates a synthesized sound by using the kurtosis of the spectrum obtained by band division as an index representing periodicity corresponding to the variation amount of the group delay spectrum. Since the kurtosis value Km (f) of the band-divided spectrum has a corresponding relationship with the fluctuation amount Wm (f) of the group delay spectrum, it is determined in advance and the kurtosis of the band-divided spectrum extracted from the analysis signal. The variation amount Wm (f) of the group delay spectrum can be generated in correspondence with the value Km (f). This does not faithfully reproduce the group delay spectrum but reproduces the degree of fluctuation of the group delay spectrum in each band, thereby enabling generation of a synthesized sound having noise characteristics.

  Therefore, the sound synthesizing method and sound synthesizing apparatus according to the second embodiment can also easily generate a synthesized sound having noise characteristics.

The present invention is not limited to the first and second embodiments described with reference to the above description and drawings. For example, the following embodiments are also included in the technical scope of the present invention.
(1) In the first and second embodiments, the sound signal input to the sound synthesizer having the analysis unit is analyzed, and the synthesized sound is generated by the synthesis unit based on the analyzed signals. Instead, the synthesized sound may be generated by the synthesis unit based on the accumulated signal.
(2) in Examples 1 and 2, but using spectral kurtosis that band division as an index representing the periodicity, using spectral flat Ne scan of spectrum band division instead of spectral kurtosis that band division May be.
(3) In Examples 1 and 2 have been extracted kurtosis from the analysis signal with respect to spectrum, and the amplitude spectrum after removal of the information of the spectral envelope, the kurtosis for the spectrum from the amplitude spectrum of the linear prediction residual signal It may be extracted.
( 4 ) In the first and second embodiments, the group delay generation unit multiplies the fluctuation amount of the group delay spectrum by the random number. However, the group delay spectrum generated appropriately instead of the random number is prepared as a group delay database. A similar process may be performed for this.
( 5 ) In the first and second embodiments, the phase generation unit integrates the group delay spectrum and converts it into the phase spectrum. However, the accumulated sum of the group delay spectrum may be calculated and converted into the phase spectrum.
( 6 ) In the first and second embodiments, the 1-pitch waveform generation unit obtains the complex spectrum by combining the spectrum envelope and the phase spectrum, but an amplitude spectrum may be used instead of the spectrum envelope.

  The present invention is a real-time voice quality conversion system capable of reproducing and coordinating huskyness, a singing voice synthesis system capable of reproducing and emphasizing the shout scream method, a text-to-speech synthesis system capable of controlling huskyness, and a timbre noise characteristic can be freely controlled. It can be used for music synthesizers.

S3 ... 1st process S4 ... 2nd process S5 ... 3rd process S6 ... 4th process S7 ... 5th process 11 ... Signal extraction part for analysis 12 ... Spectrum extraction part 14 ... Index extraction part 21 ... Group delay variation | change_quantity production | generation part DESCRIPTION OF SYMBOLS 22 ... Group delay production | generation part 23 ... Phase production | generation part 24 ... 1 pitch waveform production | generation part 25 ... Superimposition addition part 26 ... Memory | storage part

Claims (8)

A sound synthesis method for generating a synthesized sound,
Predetermined spectral kurtosis that band division or based on the correspondence between the value and the variation amount of the group delay spectrum of the spectral flatness, to correspond to any value of kurtosis or spectral flatness of spectrum the band division A first step of generating a variation amount of the group delay spectrum,
A second step of generating a group delay spectrum using the variation amount of the group delay spectrum generated in the first step;
A third step of calculating an integral or cumulative sum of the group delay spectrum generated in the second step and converting it into a phase spectrum;
A complex spectrum is obtained by combining an arbitrary amplitude spectrum or an arbitrary spectral envelope and the phase spectrum converted in the third step, and the complex spectrum is subjected to inverse Fourier transform to obtain a one-pitch waveform which is a signal for one period. A fourth step of generating
A fifth step of generating a synthesized sound by superimposing and adding the 1 pitch waveform generated in the fourth step ;
With
A sound synthesizing method, wherein the second to fifth steps are repeated to generate a synthesized sound having a synthesis frame length. The sound synthesis according to claim 1, wherein an arbitrary value of the kurtosis or the spectrum flatness of the spectrum obtained by the band division is extracted from an analysis signal having a frame length of a time length set for the input sound signal. Method. The sound synthesis method according to claim 1 or 2, wherein the arbitrary amplitude spectrum or the arbitrary spectral envelope is extracted from an analysis signal having a frame length of a time length set for an input sound signal . 4. The sound synthesis method according to claim 1, wherein the group delay spectrum generated in the second step is generated by multiplying a variation amount of the group delay spectrum by a predetermined coefficient . The sound synthesis method according to any one of claims 1 to 4, wherein the group delay spectrum generated in the second step is generated by multiplying a variation amount of the group delay spectrum by a random number . A sound synthesizer for generating a synthesized sound,
A storage unit storing a correspondence relationship between a kurtosis or spectrum flatness value of a predetermined band-divided spectrum and a variation amount of the group delay spectrum;
Based on the correspondence relationship between the kurtosis or spectrum flatness value of the band-divided spectrum stored in the storage unit and the variation amount of the group delay spectrum, the kurtosis or spectrum flatness of the band-divided spectrum is calculated. A group delay variation generating unit for generating a group delay spectrum variation corresponding to an arbitrary value;
A group delay generation unit that generates a group delay spectrum using the variation amount of the group delay spectrum generated by the group delay variation generation unit;
A phase generation unit that calculates an integral or cumulative sum of the group delay spectrum generated by the group delay generation unit and converts it into a phase spectrum; and
A complex spectrum is obtained by combining an arbitrary amplitude spectrum or an arbitrary spectral envelope and the phase spectrum converted by the phase generation unit, and the complex spectrum is subjected to inverse Fourier transform to obtain a one-pitch waveform which is a signal for one period. A one-pitch waveform generating unit for generating
A superposition addition unit that superimposes and adds the one pitch waveform generated by the one pitch waveform generation unit to generate a synthesized sound;
A sound synthesizer characterized by comprising: An analysis signal extraction unit that extracts an analysis signal for each frame length of a time length set for the input sound signal;
An index extraction unit for extracting a value of spectrum kurtosis or spectrum flatness obtained by band-dividing from the analysis signal extracted by the analysis signal extraction unit;
The sound synthesizer according to claim 6 . An analysis signal extraction unit that extracts an analysis signal for each frame length of a time length set for the input sound signal;
A spectrum extraction unit for extracting the amplitude spectrum or the spectrum envelope from the analysis signal extracted by the analysis signal extraction unit;
The sound synthesizer according to claim 6 or 7.