JP3468337B2 - Interpolated tone synthesis method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention synthesizes two kinds of original tones from all sounds including voice and music, and uses the two original tones as both ends to synthesize interpolated tones at an arbitrary ratio. The present invention relates to an interpolation tone color synthesis method.

[0002]

2. Description of the Related Art In a conventional electronic musical instrument, a user adds various tone colors (1) preset by various methods such as PCM sound source and FM sound source, and (2) addition by user newly recording sound ( Sampling), (3) re-synthesis in which a parameter expression such as an FM sound source is modified and edited, (4) various filter processes, and processing such as distortion addition has been performed.

Further, continuous processing of timbre and subtle processing could be realized to some extent by processing such as filter processing and distortion addition. However, these are to destroy and transform a certain timbre, and it was not possible to perform more advanced timbre control by giving a target specific timbre and interpolating toward that timbre.

[0004]

SUMMARY OF THE INVENTION The object of the present invention is to synthesize a tone color which is an arbitrary perceptual internal division point including both ends of two given tone colors, and one of the two tone colors given. It is an object of the present invention to provide an interpolated timbre synthesizing method capable of realizing timbre synthesizing that continuously changes from to the other end.

[0005]

According to the present invention, for two original sounds (those having different tone colors), parameters for expressing the original sound as a model are estimated at regular intervals, and these estimated two original sounds are estimated. The corresponding time points between the parameters of the two are extracted, the corresponding ones between the estimated parameters of the two original sounds at these extracted time points are found, and the corresponding parameters found are
Interpolation is performed according to the desired degree of proximity of the synthesized tone color to the two original tones.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 FIG. 1 shows the functional configuration of an interpolated tone color synthesis apparatus to which the method of the present invention is applied. With respect to the two original sound waveforms x and y, a parameter used for expressing the original sound by the signal model, for example, a partial sound is estimated by the partial sound decomposition unit 11. In this example, the original sounds x and y are waveforms separated for each frame of, for example, about 5 msec, and these waveforms are expressed as follows.

X = {x _n | n = 0,1, ..., N _x −1} (1) y = {y _n | n = 0,1, ..., N _y −1} (2) where N _x and N _y are the total number of frames of the original sound waveforms x and y, respectively. In this example, the partial sound decomposing unit 11 performs spectrum analysis and rearrangement necessary for synthesis by the sine wave superimposed model representation on the original sound waveform, and decomposes into partial sounds (sine waves). The sinusoidal superposition model is expressed as follows.

X (m) = ΣA _p (m) cos (θ _p (m)) (3) Σ is from p = 0 to P−1, x (m) is the signal of the m-th sample in the frame, p is the partial number, P is the number of partials, θ _p (m) is the instantaneous phase of partial p of the m-th sample, and A _p (m) is the instantaneous amplitude of partial p.
In equation (3), frame number n is omitted. In addition to A and θ, P takes different values depending on the analysis frame.

The information necessary for the expression (3) is the spectral information of the partial sound at the boundary of each frame, and the instantaneous (for each sample: subscript m) information can be calculated by interpolation during synthesis. It is useless. From this, the following is output for the spectrum parameter of the original sound waveform x. X = {X _n | n = 0,1, ..., N _x −1} (4) X _n = {(f _xp ⁿ , A _xp ⁿ , θ _xp ⁿ ) | p = 0, 1, ..., P _x (n) −1} (5) f is an instantaneous frequency, which is a time derivative of θ, where f, A, and θ are boundary values, respectively, and f _xp ⁿ = f _xp ⁿ (m = 0) , A _xp ⁿ = A _xp ⁿ (m =
0), θ _xp ⁿ = θ _xp ⁿ (m = 0). Similarly, for the spectral parameter of the original sound waveform y, Y represented below is calculated.

Y = {Y _n | n = 0,1, ..., N _y −1} (6) Y _n = {(f _yp ⁿ , A _yp ⁿ , θ _yp ⁿ ) | p = 0, 1, ... , P _y (n) −1} (7) Several parameter estimations of the equations (4) to (7) have been proposed since Moorer's algorithm (1977).
As a typical example, McAulay and Qu for obtaining X and Y by detecting the frequency, amplitude and phase of a local peak (each peak other than noise component) obtained from FFT.
algorithm by atieri (MQ algorithm,
1986).

Then, the corresponding time points between the spectral parameters X and Y of the original sound waveform thus obtained are extracted by the time point correspondence extraction unit 12. Here, the original sound waveforms x and y are vocal sounds of the same phoneme (sentence), the same melody sound, the same rhythm sound, the musical sounds by the same playing method, etc., but those having different tone colors are targeted. , Y, the correspondence at the time when the same phenomenon occurs perceptually.

FIG. 2 shows the result obtained after the correspondence between the time points of the parameters X and Y is taken. The horizontal axis represents time by frame number. In this example, N _x = 9 and N _y = 11, and X ₂ and Y _{1 are set} as the second (k = 2) corresponding data.
And X ₃ and Y ₁ correspond as the third (k = 3) corresponding data. The correspondence between the parameters X and Y is matching by nonlinear expansion / contraction of time, and can be realized by using a conventionally known DTW (Dynamic Time Worping) method. As a result, information C (k) representing the correspondence at the time point is obtained.

C (k) = (C _x (k), C _y (k)), k = 0, 1, ..., K−1 (8) C _x (k) ε {i | i = 0,1 , ..., N _x −1} (9) C _y (k) ε {j | j = 0,1, ..., N _y −1} (10) In the example of FIG. 2, (C ₁ (0), C ₀ (0)), (C ₂ (1), C
₁ (1)), (C ₃ (3), C ₁ (3)), (C ₄ (4), C ₂ (4)),
(C ₄ (5), C ₃ (5)), (C ₅ (6), C ₄ (6)) ...

Next, the corresponding time point C extracted in this way
The interpolation tone color synthesis section 13 synthesizes the interpolation tone color waveform Z using (k), the estimated parameters X and Y, and the degree of the synthesized tone color, that is, the ratio α of the original sound waveforms x and y in the synthesized tone. As shown in FIG. 3, the interpolated timbre synthesizer 13 is roughly divided into an initial value calculator 15 and a single-frame interpolated timbre recurrence calculator 16. The initial value calculation unit 15 is operated only when the count value of the counter storage unit 19 that counts the frame n is n = 0. The spectrum correspondence search unit 17 searches for a corresponding one of the parameters X ₀ and Y _{0 in} the 0th frame (C _x (0), C _y (0)).

Next, the spectrum interpolator 18 interpolates the corresponding spectra O _x and O _{y according} to the interpolation information, that is, the synthesis ratio α to obtain a new spectrum (parameter) Z ₀ .
A specific method for searching and interpolating the corresponding spectrum will be described later. This new spectrum Z ₀ is supplied as Z _{n to} the single-frame interpolation timbre synthesis unit 21 in the single-frame interpolation timbre recurrence calculation unit 16. On the other hand, in the reference time point calculation unit 22 in the single-frame interpolation timbre recurrence calculation unit 16, n + 1
, That is, the frame number k of Z _{n + 1} which is associated with the frame numbers of the spectra X and Y of the original sounds x and y used in the calculation of the interpolated spectrum Z _{n + 1 when} n = 1 at the beginning.
The mi is calculated using C (k) of the equation (8) previously obtained and the interpolation information α _n .

This state will be described with reference to FIG. In a typical frame, the spectral parameters at time n are known in the analysis of the previous frame and calculate time n + 1. Now, n + 1 = 3. When the interpolation information α _n changes with time as shown by the curve 23 in FIG.
In the case of n = 3, the spectrum Z ₃ ′ of the synthesized tone color is
It is on a line 24 connecting C _x (4) and C _y (4), this line 24 is the k = fourth value, and the corresponding time (frame) C _x (4),
Since each spectrum of C _y (4) is X ₄ and Y ₂ , these X ₄ and Y ₂ are interpolated by α to obtain a new spectrum Z ₃ ′.

On the other hand, for example, with α = 0.3 and α fixed, n
+ 1 = 3 synthesized tone color spectrum Z₃Is the time point extraction unit
Correspondence information (C
_x(k), C _yNot on the line connecting (k)). So interpolation (composite)
Tone) Spectrum Z₃ Of the original sound needed for the calculation of
C to find the tor X, Y_x(k), C_ystraight tie (k)
Find the number k = kmi close to the line. In the example of FIG. 4, α = 0.
3 and n + 1 = 3 interpolation spectrum Z₃ Close to C_x(k), C
_yThe line connecting (k) is a straight line 24 with k = 4, so
Frame C at both ends of straight line 24_x(4), C_yIn (4)
X and Y spectra X_Four And Y₂ Using Z₃ Seeking
It This synthetic tone Z_{n + 1} Passing through C_x(k), C_ytie (k)
Z if there is no line, that is, k_{n + 1} Closest to k = kmi
Is calculated by the following formula.

Kmi = arg min | (1-α) · C _x (k) + α · C _y (k) −n | (11) kmiε (k | k = 0, 1, ..., K−1) ( 12) arg min | A (k) | is to find k having the smallest value of A (k) among given k. C in this example
_{x (kmi = 3) = 4} , C y (kmi = 3) = 2 and composed. It can be said that the expressions (11) and (12) mean the following. That is, the corresponding frames (C _x (k), C _y (k)) of the two original sounds are extracted for each frame, and the corresponding frames extracted by the degree α of the desired proximity to the two original sounds of the synthesized tone are interpolated. (This is not limited to linear interpolation) is to select the extracted corresponding frame so that the time point when it is performed is closest to the time point represented by the frame to be newly calculated for the synthesized timbre.

In this way, the interpolation information α_nIn
Response time (frame) C_x(k), C_yspectrum of (k)
X_i, Y_j(I = C_x(k), j = C_y(k)) is obtained,
Both spectra X_i, Y_jΑ_nInterpolated according to
New spectrum Z_{n + 1} But the interpolation is as follows
To do. First, the spectrum correspondence search unit 26
X _i, Y_jAsk for the corresponding. X here_i, Y
_jThe number of elements in each
Although the elements are different in time, these two different dimension vectors
Tor, define the distance between its elements, optimal as a whole
Determine who will match. In this case, the other person sees
If the cost of failure is defined in the same dimension as distance,
With DP (Dynamic Programming) using the idea of total cost minimum
You can make a decision. As this distance measure, the instantaneous frequency,
Weighted Euclidean, with levels as elements of the vector
Distance can be considered. This is the moment when considering the case of the same pitch
Normalized the instantaneous frequency with the fundamental frequency rather than the time frequency itself
And it is more natural to use a numerical value (real number) corresponding to the harmonic
It will be a correspondence. That is, in the latter, each of the same harmonics
They will correspond to each other.

Given the spectrum X _i of frame _i and the spectrum Y _{j of} frame j corresponding to km _i ,
The dimension P _x (n), P _y (n) of each spectrum has a variable length,
It depends on the frame number n and the original sound waveforms x and y. This X
Find the best partner for _i , Y _{j by the} above algorithm. The corresponding one found in this way is recorded by the following formula. In other words, the corresponding one found is temporarily stored.

O _x (p _x ), p _x = 0,1, ..., P _x (i) O _y (p _y ), p _y = 0,1, ..., P _y (j) If it holds, it can be expressed by the following equation. O _x (p _x ) ε {p _y | p _y = 0,1, ..., P _y (j) −
1} O _y (p _y ) ε {p _x | p _x = 0,1, ..., P _x (i) −
1} Those that do not correspond are distinguished by defining a value other than the value used in the case of correspondence, such as setting O _x (p _x ) = − 1. If there is no other party, O _y (p _y ) does not have to be recorded because of redundancy.

For example, X_iThe spectrum of is shown in Figure 5A.
As the basic frequency is 100 Hz,
Each element O of wave number 0, 1, 2, 3, 4_x(0), O_x(1)
 , O _x(2), O_x(3), O_xFor (4), Y_jSpa
As shown in Fig. 5B, Khutor has a fundamental frequency of 200 Hz.
And each element O of harmonic numbers 0, 1, 2
_y(0), O_y(1), O_y(2). O at this time
_x(0) is O_yCorresponds to (0), O_x(0) = 0, O
_xO is because there is no correspondence in (1)_x(1) = -1 and O_x(2)
Is O_yCorresponds to (1), O_x(2) = 1 and O_x(3) is O
_yCorresponding to (2), O_x(3) = 2 and O_x(4) is compatible
O because there is no_x(4) = -1. Similarly, O_y(0) is O
_xCorresponds to (0) O_y(0) = 0, O_y(1) = 2, O_y(2)
= 3, O_y(3) = -1.

When the corresponding X _i , Y _j is found and recorded as described above, the recorded X _i , Y _j and α _{n are} recorded.
Using the physical interpolation by the spectrum between the alpha _n marked with correspondence, spectrum corresponding in doubt performs interpolation to zero value in the spectrum interpolation unit 27, to create a new composite tone spectrum Z _{n + 1.} That is, for example, O _x () and O
_{When y} () corresponds, αO _x () + (1-α) O
_y () = O _z (), and if there is no corresponding value, then O
_z ().

As described above, by the management by the harmonic number,
You don't have to make noise. When the unvoiced part is included in the original sound, (1) In the unvoiced part and unvoiced part, for the sake of convenience, an arbitrary harmonic band width is commonly set for the two original sounds. (2) In the unvoiced part and unvoiced part May set the harmonic band of the unvoiced part to be the same as the harmonic band of the voiced part. The new tone color spectrum Z _{n + 1} thus synthesized is calculated by the single frame interpolated tone color synthesizer 21 from the tone color waveform Z _n calculated from the previous frame Z _n . As the method of synthesizing the waveform, for example, the following various known methods can be used.

1. A method of performing sine wave synthesis within a frame using the parameters obtained from analysis for each frame and then performing overlap add on adjacent frames. 2. MQ mentioned above
An algorithm that expresses the instantaneous phase function by a cubic expression. 3. As shown in FIG. 6, in the MQ algorithm, the spectrum correspondence search unit 26 used in FIG. 3 is used again in the connection of the spectrum local peak of the frame and the next frame.

Since the last method uses the same calculation unit, the device configuration becomes easy. Corresponding spectral O _zn in the spectrum corresponding search unit 26 and Z _n and Z _{n + 1} in FIG. 6, O _{zn + 1}
And the corresponding O _zn and O _{zn + 1} determine the spectral information that becomes the boundary in the equation (3). Therefore, the interpolation data, that is, the equation (3), corresponds to the sample point m between the two. The single-frame sound synthesizer 29 obtains the synthesized waveform.

In the description of FIG. 3, the spectrum interpolation unit 2
7. Synthetic timbre spectrum Z_{n + 1} Remember this
Z to part 28_nAs the frame of the storage unit 19
The count value n is incremented by 1 and the frame is calculated again by the reference time point calculation unit 22.
Α at m n + 1_nAnd kmi corresponding to is calculated, and k
Corresponding spectrum X of mi_i, Y_jThe corresponding spectrum of
And perform spectrum interpolation to Z_{n + 1} Ask for.Example 2 There are cases where the two original sounds at both ends have the same sentence but different voice qualities.
Or two original sounds with different pitches or powers
Two original sounds that are more complex than simple stationary sounds such as cases
And the pitch or volume is perceptually
It is an independent element, and when interpolating, these are interpolated individually
Must have been done. Also, the spect that controls phonology
The le envelope also preserves the phoneme and is a physical supplement that interpolates the timbre.
A time should be made. Therefore, in this second embodiment, FIG.
As shown in, first, in the initial stage of analysis, partial sound decomposition is performed.
In addition to the above, the pitch extraction unit 31 uses the original waveforms x and y.
Pitch P_x, P_yRespectively, power extraction
Power a by the unit 32_x, A_yRespectively, and further
Spectral envelope E by spectral envelope extraction unit 33 _x，
E_yRespectively.

Parameter interpolation, that is, spectrum interpolation is performed by the spectrum interpolating unit 27 in FIG. 3 using these parameters. The spectrum interpolation in this case is performed as shown in FIG. In other words, the spectrum interpolator 27 has a frame number n _x (= C) of the corresponding frame.
_x (k _mi )), n _y (= C _y (k _mi )), the numbers p _x and p _y (= O _x (p _x )) of corresponding partial tones of the spectrum of these frames, user , The physical interpolation constant α, the pitches P _x and P _y , the powers a _x and a _y , and the spectral envelopes E _x and E _y are input.

The spectral envelopes E _x and E _y are interpolated to give a desired spectral envelope to a desired synthesized tone color. That is, the corresponding frame numbers n _x and n _y , the spectrum envelopes E _x and E _y , and the interpolation information α _n are assigned to the peak-to-peak correspondence search unit 35.
By inputting the peak frequency of E _x and E _y and the distance using the level as a vector, the corresponding peak is calculated using the same method as the search method in the spectrum corresponding search unit 26, and the corresponding peak is calculated. To the spectrum envelope interpolation unit 36
Then, linear interpolation corresponding to α _n is performed to obtain the target spectral envelope E _z .

On the other hand, the amplitude A of the original sound waveform x of the frame n _x
xpx ^nx and the amplitude A _ypy ^ny of the original sound waveform y of the frame n _y are linearly interpolated by the amplitude interpolator 37 with the interpolation information α _n , and the spectrum envelope of the interpolated amplitude is calculated by the spectrum envelope calculator 38. The shift of the spectrum envelope from the previously obtained target spectrum envelope E _z is calculated by the spectrum envelope correction amount calculation unit 39, and this shift is added by the adder 41 to the interpolated instantaneous amplitude obtained by the amplitude interpolation unit 37. Thus, the spectral envelope of the interpolated instantaneous amplitude is set to be a target. In this way, the desired spectral envelope is given to the desired synthesized tone.

The following processing may be added to give a target spectral envelope to this desired synthesized tone color. That is, the power a _x (d) extracted in each of the frames n _x and n _y
B), a _y (dB), with the interpolation information α _n
2 linearly interpolates to obtain the target power value a _z , and the level adjuster 43 adjusts the level of the interpolated amplitude whose spectrum envelope has been corrected by the adder 41 to be a _{z. Get zp} ^{n + 1} .

When the pitch is interpolated, the correspondence with the perception is
For better results, take a logarithmic pitch and use frame n_x, N
_yExtraction pitch P at_x, P_yInterpolation information α_nThen,
Linear interpolation is performed by the pitch interpolation unit 45, and the interpolated pitch
Chi P_zIs used to compensate the interpolation constant when interpolating the instantaneous frequency.
Recalculation is performed by the inter-coefficient recalculation unit 46, and the calculation result α_n′
It is supplied to the frequency interpolation unit 47. That is, the interpolation of the instantaneous frequency
Α_n, The pitch of the interpolated composite waveform will shift a little.
That's the pitch P_zInterpolation information α according to_nIt is better to fix
Aurally good. In this way, the frequency interpolation unit 47
Hour frequency f_xpx ^nx, F_ypy ^nyΑ_nInterpolate with ’
Hour frequency f_zp ^{n + 1}To get

Further, in order to generate the tone color synthesized waveform z _{n + 1} , in addition to the instantaneous frequency f _zp ^{n + 1} and the instantaneous amplitude A _zp ^{n + 1} , the instantaneous phase θ _zp.
Requires ^{n + 1} . In ^order to obtain this instantaneous phase θ _zp ^{n + 1} , the phase calculator 48 calculates the instantaneous phases θ _xpx ^nx and θ _ypy of the original sound waveform.
^ny is input, and without interpolating with α _n , the partial sound is regarded as a chirp signal in the frame and is determined only by the instantaneous frequency, or is made random, or the phase difference between each adjacent frame of the two original sounds Is determined by, for example, interpolating.

The level adjustment by the level adjusting unit 43 may be omitted. Further, α _n may be used as the interpolation information in the frequency interpolation unit 47, and the pitch interpolation unit 45 and the interpolation coefficient recalculation unit 46 may be omitted. The spectrum interpolation unit 27 in the first embodiment (FIG. 3) is the spectrum interpolation as described above, and is performed by the amplitude interpolation unit 37, the frequency interpolation unit 47, and the phase calculation unit 48.

In the above description, the time point correspondence extraction unit 12 extracts the corresponding frames as the corresponding time points of the original sound waveforms x and y, but it is predicted that more accurate timbre synthesis will be obtained if the corresponding time points are sought. Therefore, the time corresponding to each sampling time of each original sound waveform may be obtained. In this case, partial sound decomposition may be performed for a certain section at each sample time point. Further, in the above description, the parameter for expressing the original sound as a model is not limited to the partial sound, and a spectrum, a waveform or the like may be used.

In FIG. 1, the partial sound decomposition unit 11, the time point correspondence extraction unit 12, and the interpolated timbre synthesis unit 13 are basically based on offline calculation rather than real time. By calculating the correspondence extraction unit 12 in advance and using an appropriately fast arithmetic unit,
It is also possible that the user specifies α in real time to cause the interpolation tone color synthesis unit 13 to perform the calculation in real time, that is, the interpolation tone color synthesis sound waveform is generated in real time. Depending on the purpose of use, an interpolated synthesized tone color parameter Z _{n + 1} may be obtained and waveforms may be synthesized at another place. That is, according to the present invention, the interpolated synthesized tone color parameter Z _{n + 1} is obtained. There are features.

[0037]

As described above, according to the present invention, the interpolated timbres of the two tones, for example, the adult male voice and the adult female voice, are human voices, that is, voices, if the two tones have a common attribute. The interpolated timbre can also keep it. This is different from tone color control in the direction where there is no target, such as simply adding distortion or adding noise, and creates a synthesized tone color between one original sound and the other original sound to an arbitrary degree close to that original sound. Of course, one of the two original sounds can also be output, and the degree of freedom in controlling the timbre becomes greater.
Further, it is possible to perform tone color morphing in which the tone color is continuously changed from one original tone to the other original tone.

In particular, according to the second embodiment, it is possible to secure even an interpolated tone color in which linguistic features such as phonemes and prosody common to two original sounds are synthesized, and perceptually independent elements such as pitch and power are also included in the two original sounds. When they are different, they are also interpolated. As a result, the musical instrument can have a function as an electronic musical instrument that has never existed before. Further, in computer graphics, perceptual continuous deformation is called morphing, which has already been put to practical use and widely used, but sound morphing can be performed in combination with morphing of computer graphics.

[Brief description of drawings]

FIG. 1 is a block diagram showing a functional configuration of an interpolation tone color synthesis apparatus to which the method of the present invention is applied.

FIG. 2 is a diagram showing an example of a correspondence relationship between two original sounds in a time point correspondence extraction unit 12 in FIG.

FIG. 3 is a block diagram showing a functional configuration of an interpolated tone color synthesis unit 13 in FIG.

FIG. 4 is an explanatory diagram for obtaining each time point of two original sounds used for interpolation from a time point to be synthesized and interpolation information α in a reference time point calculation unit 22 in FIG.

5 is a diagram showing a spectrum example of two original sounds for explaining a corresponding spectrum obtained by a spectrum correspondence search unit 26 in FIG. 3;

6 is a block diagram showing an example of the functional configuration of a single-frame interpolation timbre synthesizer 21 in FIG.

FIG. 7 is a block diagram showing a functional configuration of an initial process in the second embodiment.

FIG. 8 is a block diagram showing functional information of a spectrum interpolation unit in FIG. 3 in the second embodiment.

Claims (7)

(57) [Claims] 1. A method for generating a synthesized timbre of two original sounds with a desired closeness to the original sounds, wherein a parameter for modeling the original sounds of the two original sounds is represented by a sine wave superposed model expression at regular intervals. Based on
Hazuki comprising the steps of: estimating each of these estimated two and the process of extracting the corresponding time points of the parameter of the original sound, different the two original at the time corresponding to the extracted
A dimension vector defines the distance between the elements,
The cost of failing is defined in the same dimension as the distance,
By using the idea of minimum total cost,
The process of finding the corresponding one between the estimated parameters and the process of interpolating the found corresponding parameter according to the degree of the desired proximity of the synthesized tone to the two tones to create a new parameter. And an interpolated tone color synthesizing method. 2. A method for generating a synthesized timbre from two original sounds with a desired closeness to the original sounds, a process of estimating parameters for modeling the original sounds of the two original sounds at regular time intervals. , The process of extracting the corresponding time points between the parameters of these two estimated original sounds, the process of finding the corresponding ones between the estimated parameters of the two original sounds at the extracted corresponding time points, and the correspondence found above the parameters, and a process of creating a new parameter by interpolation according to the desired degree of proximity to the two-original sound of the synthesized tone, a new interpolated depending on the degree of the desired closeness Parame
The process of creating the
And the extracted spectral envelopes.
Interpolate the target spectral envelope to the desired synthesized tone
The process of applying and interpolating the instantaneous frequencies of the above two original sounds
And a step of determining the instantaneous phase of the synthesized timbre . 3. The step of extracting the corresponding time points includes the steps of extracting the corresponding frames for each fixed time (frame) and numbering them in order of time, and the desired proximity of the synthesized tone color to the two original sounds. The extracted corresponding frame is selected so that the time when the above-mentioned extracted corresponding frame is interpolated is closest to the time represented by the frame to be newly calculated in the synthesized tone color, and this is extracted as described above. 3. The interpolated tone color synthesizing method according to claim 2 , further comprising the step of setting corresponding time points, wherein each parameter in both the selected corresponding frames is set as the found corresponding parameter. 4. The process of applying the target spectral envelope to a desired synthesized tone color is a process of interpolating the amplitudes of the two original sounds, a process of obtaining a spectral envelope of the interpolated amplitudes, and a process of obtaining the spectral envelope of the interpolated amplitudes. 4. The interpolated tone color synthesis according to claim 2 , further comprising a step of extracting a difference between the spectrum envelope and the interpolated spectrum envelope, and a step of adding the difference of the spectrum envelope to the interpolated amplitude. Method. 5. The process of extracting the powers of the two original sounds and the process of interpolating the extracted powers to determine the target power in the process of imparting the target spectral envelope to a desired synthesized tone color. process and interpolation tone synthesis method according to any one of claims 2 to 4 powers of the desired composite tone comprising the step of adjusting so that the power of the target. 6. The process of interpolating the instantaneous frequency includes a process of extracting the pitches of the two original sounds, and a process of interpolating the extracted pitches at corresponding time points to determine a target pitch. From the target pitch, the process of recalculating a constant representing the desired closeness to the two original sounds, and the instantaneous frequency of the two original sounds is interpolated using the recalculated constant representing the closeness. 3. The process according to claim 2
6. The interpolated tone color synthesizing method according to any one of items 1 to 5. 7. The interpolated tone color synthesizing method according to claim 2 , further comprising a step of synthesizing a waveform using the new parameter.