JPS6367196B2 - - Google Patents

Description

[Detailed Description of the Invention] Industrial Application Field The present invention is a waveform generation method that freely creates temporal changes through interpolation calculations of a plurality of waveforms to create natural-sounding voices and musical sounds.
It is suitable for voice synthesizers and electronic musical instruments.

Conventional configuration and problems Conventionally, speech waves are synthesized by storing multiple waveforms, repeating one of the waveforms a predetermined number of times, and sequentially switching to the next new waveform. There is a way. In this method, since waveforms with completely different spectra are connected when switching waveforms, smooth changes in sound quality cannot be achieved and unnecessary noise is generated.

To prevent this, there is a method of interpolating multiple waveforms, as in Japanese Patent Application No. 55-155053 (Japanese Unexamined Patent Publication No. 57-78599), but in order to generate free waveforms, further ingenuity is required. is necessary.

OBJECTS OF THE INVENTION An object of the present invention is to provide a waveform generation method in which the transition from one waveform to another is smooth and can be freely performed without being restricted by the cycle of the waveform.

Structure of the Invention The present invention provides one or more waveform generators that sequentially generate a plurality of types of waveforms, a window function generator that generates a window function, and a multiplier. The product is multiplied by the function, and the output waveform is obtained by adding the product and any one of the original waveforms, and in the zero point interval of the window function,
The above waveforms are switched. By doing this, the spectrum at the waveform switching point changes smoothly, so no unnecessary noise is generated.

DESCRIPTION OF THE EMBODIMENTS FIG. 1 is a block diagram of one embodiment of the present invention.

1 and 2 are waveform generators that generate musical tone waveforms, and a memory 5 that stores multiple types of W ₁ to _{W 5}
Then, waveform samples are read out in a predetermined order from the memory 6 that stores W ₁ to _{W 5} . The read waveform samples are supplied to multipliers 7 and 8.
3 and 4 are window function generators that output window function waves to be described later. Multipliers 7 and 8 multiply the waveform by the window function, and adder 9 calculates the sum of the products. The envelope generator 10 and the multiplier 11 add an envelope change to the output waveform of the adder 9. The output of the multiplier 11 is digital-to-analog converted and output as a sound wave.

Next, the waveform and window function will be explained. Waveform W
It is assumed that ₁ to W ₅ and W ₁ to _{W 5} each consist of one period of natural speech or musical instrument sound. Then, as shown in FIG. 2a, W ₁ to _{W 5}
In each section, each of waveforms W ₁ to W ₅ is repeated. On the other hand, the window function is a triangular function such as F ₁ to F ₅ as shown in FIG. 2b. Waveforms W ₁ to _{W 5} and W ₁ to _{W 5} and window functions F ₁ to F ₅ and F ₁ to F ₅ are shown in Figure 2 a, c, and b.
As can be seen by comparing and d, the waveform switching time and the phase of the window function are different from each other.

The sample value of waveform W _i at time nT is W _i
(nT), the window function is F _i (nT), the sample value of the waveform W _i is W _j (nT), and the window function is F _j (nT), then the output waveform WO (nT) is given by the following equation.

WO (nT) = W _i (nT) × F _i (nT) + W _j (nT) × F _j (nT) ... (1) Here, j = i or i-1 In the W _i interval, the waveform W _i is read out repeatedly, generating R _i waves. The value of R _i depends on the time length of the window function F _i , and may be an integer or a decimal number. When the number is a decimal, the waveform W _i switches to the waveform W _i+1 from the middle of the waveform W i. In this case, the phases of the waveforms W _i and W _i+1 are not necessarily continuous, so there is not necessarily a smooth transition between the waveforms, but since the window function passes through a zero value when moving from the window function F _i to F _i+1 ,
The multiplication output W _i ×F _i changes smoothly. The same can be said about the multiplication output W _j ×F _j . That is, regardless of the phase of the waveforms W _i and W _j and the number of repetitions, unnecessary noise is not generated in the multiplication output because there is no discontinuity in the instantaneous value of the output waveform or discontinuity in the differential coefficient. This situation is shown in Fig. 2 e, f, and g. e is a waveform, f is a window function, and g is a product.

As for the waveform W _i , one cycle of the waveform is repeated in the interval W _i , but instead
All waveforms within the section may be stored in the memory 5 and read out. At this time, no discontinuity occurs at the contact points of the sections.

When transitioning from the waveform W _i to W _i+1 , the waveforms may have the same shape but differ in initial phase. In this case, it is sufficient to omit the waveform W _i+1 and jump the read address of the waveform W _i discontinuously, so that memory can be saved. Such control is performed by the waveform generators 1 and 2.
This is possible by manipulating the address code output by .

FIGS. 3a, b, c, and d are other examples of waveform sections and window functions. Window function F ₁ in interval W ₁
Since the waveform W 1 is set to a constant value, the waveform W ₁ appears as it is as the output of the multiplier 7. On the other hand, the window function F
Since ₁ is taken as zero, the waveform W ₁ is unnecessary. In switching from section W ₁ to W ₂ , since there is no zero between window functions F ₁ and F ₂ , waveforms W ₁ and W ₂ need to be continuous waveforms. That is, the sections W ₁ and W ₂ may be regarded as one section, and the window function may be regarded as a trapezoidal wave that is the sum of F ₁ and F ₂ .

In the examples shown in FIGS. 2 and 3, F _i +F _j =1 (2) (j=i or i-1). Therefore, the following equation may be used instead of equation (1).

WO(nT)=W _i (nT)+{W _j (nT) −W _i (nT)}F _j (nT) ...(3) (j=i or i-1) or WO(nT)={ W _i (nT) - W _j (nT)} F _i (nT) + W _j (nT) ...(4) (j=i or i+1) In other words, two sample values are added to one of the two waveforms. The operation is to add the difference value of the waveform sample and the product of the window function.

Next, using FIG. 2, waveforms W _i to W _i+1 ,
Alternatively, interpolation from waveform W _i to W _i will be explained. Focusing on the fall of F _i (time T ₀ to _{T 1} ) in FIG. 2b, the level of waveform W ₁ monotonically decreases. On the other hand, since F ₂ rises, the level of waveform W ₂ increases. W ₁
When the phases of each frequency component of waveform W 1 and W ₂ match, each frequency component changes linearly from the level of waveform W ₁ to the level of W ₂ . The phase is preserved as is. That is, each frequency spectrum level is linearly interpolated. When the phases of the frequency components of waveforms W ₁ and W ₂ do not match, it becomes necessary to consider not only level interpolation but also vector addition including phase angle. At this time, both the phase and amplitude change. An example of this situation is shown in FIG. A composite vector on a straight line from the vector of waveform W→ ₁ to the vector of waveform W→ ₂
WO → is moving.

FIG. 5 is another example of a window function. F _i and F
A finite zero interval is provided between _i+1 , and the waveform W
_i and W _i+1 switch within this zero interval.
Therefore, even if there is a discontinuity between the waveforms W _i and W _i+1 , no discontinuity occurs in the connection between W _i ×F _i and W _i+1 ×F _i+1 . Since there is a zero interval,
Although there is a slight deviation from linear interpolation between the waveforms W _i and W _i , there is no practical problem.

In FIG. 6, each of F _i and F _i is a trapezoid, and F _i +F _i =1...(5) or F _i +F _i+1 =1...(6). In this case, only one waveform is output at the top of the trapezoid. At the sloped part of the trapezoid, both waveforms are linearly interpolated.

FIG. 7 is another embodiment of the invention. 101
is a memory that stores the waveform of each section. 100
is a waveform generator that supplies an address to the memory 101, reads out a waveform sample at a predetermined address, and outputs it. The output of the waveform generator 100 is
It is supplied to a multiplier 102 and an adder 104. The output of multiplier 102 is provided to adder 104. The output of adder 104 becomes output waveform data. 10
3 is a window function generator which supplies window function data to be described later to the multiplier 103, and which also supplies the waveform generator 1 with
A waveform switching command is output to 00.

The memory 101 stores waveforms W ₁ to W ₆ , W ₁
~ _W6 are stored in order. FIG. 8 shows the procedure of arithmetic processing.

WO (nT) = W ₁ (nT) + {W ₂ (nT) - W ₁ (nT)
}F ₁ (F ₁ section) WO (nT) = W ₂ (nT) + {W ₂ (nT) - W ₂ (nT)
}F ₂ (F ₂ section) WO (nT) = W ₂ (nT) + {W ₃ (nT) − WI ₂ (nT)}F
₃ (F ₃ sections) 〓 WO (nT) = W _i (nT) + {W _i+1 (nT) − W _i (nT
)}F _2i-1 (F _2i-1 section) WO (nT) = W _i+1 (nT) + {W _i+1 (nT) − W _i+1
(nT)}F _2i (F _2i interval) ...(7) If the above calculation is performed for each sample value of the waveform, W _i and W _i+1 or W _i and W _i
It is possible to realize a smooth waveform transition between . The window functions F _2i and F _2i-1 are monotonically linear decreasing waveforms.

Instead of F _2i-1 and F _2i , _2i +F _2i = 1 ... (8) _2i-1 + F _2i-1 = 1 ... (9) is satisfied. Another arithmetic expression obtained by transforming the expression (7) using _2i-1 and _2i may be executed.

FIG. 9 is an example of another form of the window function _j . Here, a flat portion is provided at the vertex of a triangle and the starting point of the next triangle. The waveform is switched in this flat portion.

In the above description, examples have been given in which triangles, trapezoids, right triangles, etc. are used as window functions. These functions can easily be generated by digital circuits. For example, it can be generated by dividing the system clock pulse and counting it with a counter. The up-down counter creates a symmetrical triangular wave. A right triangle is created by an up counter or a down counter. By changing the clock frequency to the counter, the slope can be changed. When the counter output reaches zero, the waveform switching command output is supplied to the waveform generators 1, 2, and 100. If the clock is temporarily stopped when the counter output is zero or all "1", a zero section can be provided. Alternatively, a linear monotonically increasing function can be generated by accumulating a predetermined small value ΔF using an adder. If the cumulative value is reset to zero when it exceeds a predetermined value, the eighth
A function as shown in Figure c is obtained. If an adder/subtracter is used and addition is switched to subtraction, the functions shown in FIG. 2b and d can be obtained. In this method, the time length of each section can be set by changing ΔF.

We will explain the termination process when it is necessary to continue generating musical tones for a long time, such as when playing a musical instrument. If the memory 101 is very large, it will be possible to produce long sounds, but sooner or later the memory will be exhausted. When the end section of the memory is reached, the following processing is performed.

(1) Hold the final value of the window function and continue reading the waveform in that section repeatedly as before.

(2) When the final value of the window function is reached, return to the previous window function and read out the waveform in the section where it returned.

According to (1), the last waveform is repeated, so
There are no temporal changes. According to (2), since the waveform of a predetermined section is repeatedly output, the sound is accompanied by temporal fluctuations.

Furthermore, as a third method, the readout phase of the final waveform may be changed randomly. At this time, since phase modulation is added according to the window function, some time fluctuation occurs.

In the above explanation, the interpolation calculation between the waveforms of type B was explained. The following general formula may be implemented by further increasing the number of waveforms and window functions.

WO (nT) = 〓 ^N W _i (nT) FN _i (nT) ...(10) N = ,, i = section number In this case, it deviates from simple linear interpolation and can be considered as high-order interpolation. can.

Further, although the window function has been described using a triangle and a trapezoid, it is also possible to use a quadratic curve or a function having another shape. Generally, any waveform with a zero interval as shown in FIG. 10 can be used. Since the interpolation formula changes depending on the shape of the waveform, it will deviate from linear interpolation, but it is possible to create sounds with more natural temporal changes. If an appropriate fluctuation component is superimposed on the window function, amplitude fluctuations will occur between multiple waveforms, which can be used for amplitude modulation effects. This can be expressed by the following formula.

F^=F+AM...(11) F is the original window function, AM is the superimposed component, and F^ is the new window function. However, at the point of waveform switching, F^
Of course, AM must be determined so that is zero.

Instead of equation (11), an arbitrary function E may be used to generate a window function as shown in the following equation.

F^=E×F...(12) In Figure 1, using equation (12), the window function F^ _i and F^
If _i is created, E can be regarded as envelope data, so envelope control can be performed using a window function. It can also be used for amplitude modulation.

In the embodiments shown in FIGS. 1 and 7, the window function generators 3, 4, and 103 generate functions, but the window function waveforms may be stored in memory and read out.

Although not shown in FIGS. 1 and 7, since the time length of each window function corresponds to the section length of the waveform, the section length data of the waveform is stored in the memories 5, 6, and 101. It is preferable that the period length data be stored together with the waveform, and that the window function generator sequentially reads out the interval length data to generate a window function of a predetermined time length.

Although the waveform generators 1, 2, and 100 are configured to read out waveform data stored in the memory, the waveform generators may also process data or generate waveforms themselves.

The waveform is normally generated at a period corresponding to the number of samples of one period of the musical tone frequency or at an average period. On the other hand, window functions may be generated in response to real time. At this time, the respective time positions in real time are not necessarily completely synchronized. However, since changes in the window function are often slower than changes in the waveform, even if values before and after the original time position are used as sample values for the window function, there may not be a significant error. many. That is, the generation of the waveform and the generation of the window function do not need to be perfectly synchronized. In this case, even if a linear linear window function is used, some increase/decrease will occur on the straight line, but this is not a practical problem.

In the above explanation, we mainly focused on one wave of musical tones or audio, but we can also apply to waveforms consisting of multiple waves other than one wave, other arbitrary waveforms, and waveforms other than sound, such as video signals and various other signals. can also be applied.

Effects of the Invention The present invention generates a plurality of waveform samples and their difference samples, multiplies the difference samples by a window function having a zero value or substantially zero value, and adds the product to the waveform sample. to obtain an output waveform, and when the window function has a zero value or a substantially zero value, the types of the plurality of waveforms are switched.

Therefore, it is possible to synthesize constantly changing natural voices and the sounds of natural musical instruments without producing any unnecessary noise. Since there is no need to align the phases of multiple types of waveforms, not only amplitude transitions but also phase transitions can be reproduced in an interpolative manner. If the phase transition is continuous, the frequency changes, so it is also possible to create a sound that is deviated from the original waveform period and has no harmonic properties. Furthermore, vibrato, phase modulation effects, or delay time modulation can be created by shifting the phase relationships of multiple types of waveforms or by sequentially changing the readout address of one waveform for each section.

[Brief explanation of drawings]

Fig. 1 is a block diagram of an embodiment of the waveform generation method of the present invention, Figs. 2 and 3 are diagrams for explaining the waveform calculation method, and Fig. 4 also explains interpolation of phase and amplitude. Vector diagrams for Figures 5 and 6
The figure is a diagram for explaining a waveform calculation method when using another window function, FIG. 7 is a block diagram of another embodiment of the present invention, and FIG. 8 is a waveform calculation method of the embodiment of FIG. 7. 9 and 10 are diagrams of other window function waveforms. 1, 2, 100...Waveform generator, 3, 4, 100
... Window function generator, 7, 8, 102 ... Multiplier, 9,
104...Adder.

Claims (1)

[Claims] 1. Generating a plurality of waveform samples and difference samples thereof, multiplying the difference samples by a window function having a zero value or substantially zero value, and adding the product to the waveform samples. to obtain an output waveform, and the types of the plurality of waveforms are switched when the window function has a zero value or a substantially zero value. 2. The waveform generation method according to claim 1, wherein the window function is substantially triangular or trapezoidal. 3 A plurality of waveform generators that sequentially generate a plurality of types of waveforms, a plurality of window function generators that generate a window function, and a plurality of multipliers are provided, and an adder is further provided to perform the multiplication of the waveform and the window function. A method for generating a waveform, characterized in that the product is multiplied in the adder to obtain an output waveform, and the type of waveform in each waveform generator is switched in a substantially zero section of the window function. 4. Claim 3, characterized in that the plurality of waveform generators, window function generators, and multipliers are a single waveform generator, window function generator, and multiplier that are operated in a time-sharing manner. The waveform generation method described. 5. The waveform generation method according to claim 3, wherein the waveform generator generates the waveform by reading out the waveform from a memory in which the waveform is written. 6. The waveform generation method according to claim 3, wherein the sum of the window functions is substantially constant. 7. The waveform generation method according to claim 3, wherein the window function is substantially triangular or trapezoidal.