JPH0518117B2 - - Google Patents

Description

[Detailed description of the invention]

INDUSTRIAL APPLICATION FIELD The present invention relates to a waveform generation method that can be used in speech synthesis devices, electronic musical instruments, and the like. Configuration of conventional example and its problems In recent years, as a method of speech synthesis, a segment editing type speech synthesis device has been proposed that edits and outputs speech segments cut out and stored in pitch intervals. In this type of speech synthesizer, a speech output consisting of many pitch sections must necessarily be obtained using fewer speech segments. However, in this case, it is known that if the same elemental piece is repeated multiple times and connected to the following elemental pieces, discontinuity occurs in the connected part, resulting in unnaturalness in the output speech. Therefore, conventional waveform generation methods have been proposed to reduce the discontinuity of the connecting portion. In this conventional example, instead of repeating a selected speech segment, a waveform obtained by weighted averaging of this speech segment and a speech segment that is subsequently selected to be output is
Discontinuity is reduced by inserting one or more waves between these two selected speech segments.
For example, if a selected speech segment is W ₁ and a speech segment selected to be output subsequently is W ₂ ,
When inserting three speech segments between W ₁ and W ₂ , for example, W ₁ is output as is as the first wave, 3/4W ₁ + 1/4W ₂ is output as the second wave, and 1/2W ₁ +1/2W ₂ is output as the third wave, 1/4W ₁ +3/4W ₂ is output as the fourth wave, and W ₂ is output as is as the fifth wave. However, for example, 1/4 W ₁ +3/4W ₂ represents a waveform obtained by adding weights of 1/4 and 3/4 to W ₁ and W ₂ , respectively. With this configuration, compared to the case where W ₁ is repeated 4 times and then suddenly W ₂ is output, the waveform approaches from W ₁ to W ₂ in 4 steps, reducing waveform discontinuity. be done. However, for example, as shown in Figure 1, W ₁ and W ₂
Considering the case where both are cosine waves or sine waves, if both are cosine waves, as is clear from Figure 1a,
If the amplitudes of W ₁ and W ₂ are different, discontinuity will always occur at the connection portion of each waveform. If both are sine waves, each waveform is a perfect sine wave, although there is no discontinuity in the waveform itself. For example, the adjacent waveforms in Figure 1b can be expressed as a ₁ sinωt and a ₂ sinωt. Then, these 2
The differential coefficients at the connection of two sinusoids are each a ₁ ω
and a ₂ ω, which do not match. However, a ₁ and a ₂ are the amplitudes of these two sine waves, and ω is the angular frequency. Therefore, if the two waveforms a and b in FIG. 1 are acoustically output, noise due to waveform discontinuity and differential coefficient discontinuity will be sensed. In general, a speech segment Wn is expressed as Wn= _I 〓 ⁱ⁼⁰ [a _i cos ωin+b _i sin ωin] by Fourier representation, and the above-mentioned weighted average operation results in each corresponding harmonic Therefore _, as long as this conventional waveform generation method is used, the noise described in the example of _FIG . Even if they are, they will always exist unless they are exactly the same, and this will give the generated voice an unnatural feel. OBJECT OF THE INVENTION The object of the present invention is to provide a waveform generation method that can be used in a speech synthesizer or an electronic musical instrument, etc.
When expanding and outputting limited waveform information using interpolation processing, there is no discontinuity in the output waveform, and
It is an object of the present invention to provide a waveform generation method capable of generating a noise-free and natural acoustic signal by preventing discontinuity from occurring even in the differential coefficient of an output waveform. Another object of the present invention is to provide a waveform generation method suitable for an electronic musical instrument that generates a pseudo-synthesized sound in which the amplitude envelope of each harmonic of an instrument sound is approximated by a polygonal line, for example. Still another object of the present invention is to provide a waveform generation method that can produce effects such as tremolo, vibrato, and non-harmonics of musical instrument sounds. Structure of the Invention The waveform generation method of the present invention cuts out a plurality of waveforms each having a length of 1 pitch from different points in the musical sound waveform.
The phase of the harmonics included in each cut out waveform is processed so that the phase difference between harmonics of the same order in temporally adjacent waveforms takes a desired value. leading the way 2
The two waveforms are sequentially defined as a first waveform and a second waveform, and a difference waveform is obtained by subtracting the first waveform from the second waveform, and the value of this difference waveform gradually increases from approximately zero to approximately 1. Multiply the weight and apply the multiplication result to the first
The processed waveform is added to the waveform of Among the waveforms that follow, the waveform that temporally follows the second waveform is set as a new second waveform. With this configuration, the temporal movement of the harmonic spectrum of a musical tone can be accurately approximated, and it is very easy to do so. It is possible to synthesize generated sounds similar to the original musical tones. DESCRIPTION OF EMBODIMENTS An embodiment of the present invention will be described below with reference to FIG. 2 and the following drawings. In the following description, a waveform that is divided by a length of approximately one pitch period from a waveform having approximately periodicity will be referred to as a waveform of one pitch section. Also, 1
A waveform in which M waves of pitch interval waveforms are connected will be referred to as an M pitch interval waveform, and the m-th waveform will be referred to as an m-th pitch interval waveform. Further, the number of samples included in the waveform of each pitch section is represented by N.
Therefore, the nth sample of the m-th pitch section is the (m-1)N+nth sample from the beginning among the samples included in the waveform of the M-pitch section. FIG. 2 is an example of a waveform generation device embodying the waveform generation method of the present invention. In FIG. 2, reference numeral 1 denotes a timing pulse generator (hereinafter abbreviated as TPG) which defines the operation timing of this device and generates address signals to be applied to several memories to be described later. TPG1 uses CLK as a clock input.
Consists of a binary counter with LSB T0
~ Outputs 10 MSB T9 signals. or,
This TPG1 is the INIT output when waveform generation starts.
Initialized by a signal. The waveforms from T0 to T9 are shown in FIG. 2 and 3 in FIG. 2 are waveform memories that store discrete waveforms cut out from the acoustic signal in pitch intervals and discretized, and waveform designation signals WD1 and WD2.
are respectively the upper addresses, and the output T0~ of TPG1
The sample values of the discrete waveform specified by the address are each output with T5 as the lower address. 4
is a subtracter which subtracts the output of the waveform memory 2 from the output of the waveform memory 3 and outputs the result. 5 is TPG
This is a bit shifter that shifts the outputs T0 to T9 of 1 to higher r bits based on the repetition signal r.
The bit shifter 5 is, for example, as shown in FIG.
It can be configured with ROM (read-only memory). A multiplier memory 6 stores 1024 words of multiplier values, and outputs the contents stored at the address designated by the output of the bit shifter 5. The contents of the multiplier memory 6 are shown, for example, in Table 1.

【table】

[Table] Reference numeral 7 in FIG. 2 is a multiplier, which multiplies the output of the subtracter 4 and the output of the multiplier memory 6 and outputs the result. An adder 8 adds the output of the waveform memory 2 and the output of the multiplier 7 and outputs the result to a digital/analog converter (not shown). Next, the operation of the embodiment of the waveform generator configured as described above will be explained. In waveform generation, first, WD ₁ and WD ₂ signals specifying two waveforms to be used are normally supplied to waveform memory 2 and waveform memory 3 by a microcomputer. At the same time, an r signal specifying the wave number for each pitch section of the waveform generated from these two waveforms is supplied to the bit shifter 5, and TPG1 is initialized by the _INIT signal. Then TPG1 is CLK
Start the coefficients by clocking the signal. TPG
According to the count of 1, the waveform memories 2 and 3 each
Outputs one sample at a time from the first sample of the waveform specified by WD ₁ and WD _2. If WD ₁ and WD ₂ do not change even after outputting 64 samples, the waveform that has just been output is output. Output again from the first sample. The n-th sample value of the waveform output from waveform memory 2 and waveform memory 3 is W _1o ,
When W _2o is assumed, the subtracter 4 always outputs W _2o −W _1o . Next, the procedure for generating a multiplier will be described. Table 2 shows the correspondence between the r signal and the wave number for each pitch section of the generated waveform.

[Table] The operations of the bit shifter 5, multiplier memory 6, and multiplier 7 will be explained according to FIG. The outputs T0 to T9 of TPG1 are transferred to the upper level by bit shifter 5.
bit shifted. As an example, a case will be described in which the number of pitch sections of the generated waveform is four. In this case, 2 is specified as r, and the bit shifter 5 inputs T0 to T9 by 2.
Shift to higher bits. Therefore, the relationship between the outputs T0 to T9 of the TPG 1 and the outputs M0 to M9 of the multiplier memory 6 is as shown in Table 3.

[Table] Therefore, as shown in Figure 5a, TPG1 is 0~
While counting 255, T0 to T5 count 0 to 63 four times. The waveform memories 2 and 3 to which T0 to T5 are given as lower addresses each output the same waveform repeatedly four times. Further, as shown in FIG. 5b, T0 to T9 of TPG 1 counts 0 to 255, and while waveform memories 2 and 3 each repeat and output the same waveform four times, outputs M0 to M9 of multiplier memory 6
increases linearly by 4 from 0 to 1020. Therefore, M0 to M9 are expressed as follows. That is, M0 to M9 used for calculating the n-th sample of the m-th waveform of the output discrete waveform is [64×(m-1)+(n-1)]×4. However, the last multiplier 4 indicates that the outputs M0 to M9 of the multiplier memory 6 are incremented by 4 from 0. Generally, the wave number of the generated waveform is M,
If the number of samples in one wave is N, the multiplier can be expressed as [(n-1)N+(n-1)]×R (2) using R=1024/M・N (1) . The 10-bit multiplier M0 obtained in this way
M9 is multiplied by the 16-bit output signal of the subtracter 4 in the multiplier 7, and the upper 16 bits of the resulting 26-bit output signal are supplied to the adder 8.
The output to the adder 8 is a value equal to the 26-bit multiplication result shifted 10 bits lower, in other words, divided by 1024. Therefore, in reality, while TPG1 counts from 0 to 255, the output of the subtracter 4 is multiplied by a numerical value that increases linearly from 0 to 1020/1024≈0.996. Next, among the N samples (64 samples in this example) of the m-th pitch section of the M pitch section of the generated waveform,
The procedure for obtaining the n-th sample W^mn will be explained with reference to FIG. In this case, the waveform memories 2 and 3 are each outputting the n-th sample of the waveform being output. these
Assuming W _1o and W _2o , the subtracter 4 is
Outputs W _2o −W _1o . The output of the subtracter 4 is multiplied by the multiplier expressed by equation (2) by the multiplier 7, and the result is (W _2o −W _1o ) [(m-1)N+(n-1)]R
get. However, as can be seen from equation (1), 1024=
Since it is M・N・R, the upper 16 outputs of multiplier 7
The value represented by the bit is (W _2o −W _1o ) [(m-1)N+(n-1)]/M・N
(3) becomes. This value and the output W _1o of the waveform memory 2 are added in the adder 8, and the final result is W^mn=W _1o + (W _2o - W _1o ) [(m-1)N+(n-1)]/M・Obtain N (4). As described above, in the waveform generator of FIG. 1 applied to the present invention, when generating a waveform of M pitch intervals from two selected waveforms, it takes (4 ) expression is executed. Here, the analog waveforms corresponding to W _1o and W _2o are
If W ₁ (t) and W ₂ (t), then W ₁ (t)=∽ 〓 ⁱ⁼ ∽C _1i e ^j2 〓 ^fit (5) W ₂ (t)=∽ 〓 ⁱ⁼ ∽C _2i e ^j2 〓 It can be expressed as ^fit (6). However, C _2i and C _2i are the complex Fourier coefficients of the i-th harmonics of W _1o and W _2o , f is the fundamental frequency, and j is equal to √−1. Therefore, if the analog waveform corresponding to W^mn is W^(t), then W^ _(t) =∽ 〓 〓 ⁱ⁼ ∽[c _1i + (c _2i −c _1i ) (m-1) + (n -1)/M
・N〕×e ^j2 〓 ^fit (7a) =∽ 〓 ⁱ⁼ ∽C^mni e ^j2 〓 ^fit (7b) However, C^mni=C _1i + (C _2i −C _1i (m−1)N+(n− 1)/M
・N (7c). By the way, (m-1)N+(n-1)/M in formula (7c)
- The numerator (m-1)N+(n-1) in N increases by 1 from 0 to MN-1 from the first sample value _W11 until the last sample value _WMN is output. Therefore, equation (7c) shows that the instantaneous Fourier coefficient Cmni of Wmn gradually and continuously approaches C _2i from C _1i . FIG. 7a shows the complex Fourier coefficients expressed as vectors on the complex plane.
The tip of the vector Cmni moves continuously from P to Q on the straight line PQ while generating a waveform of MN samples. Therefore, as seen from equation (7b), W^
(t), the amplitude and phase of each harmonic thereof are completely continuous, resulting in a very smooth and natural sound signal as the output sound signal. For example, as mentioned above, when a waveform with a 4-pitch interval is generated from two cosine waves with different amplitudes using the waveform generator of this embodiment, the waveform shown in FIG. When a waveform with a 4-pitch interval is generated from two sine waves using the waveform generator of this embodiment, it becomes as shown in FIG. 8b, and in both cases, the discontinuity that existed in FIGS. does not exist. Furthermore, if the phases of the first discrete waveform and the second discrete waveform are adjusted in advance so that the phases of the corresponding harmonics are equal, then Equation (5) and
Equation (6) is
W ₁ (t)=∽ 〓 ⁱ⁼ ∽｜C _1i ｜e ^j 〓 ⁱ e ^j2 〓 ^fit (8) W ₂ (t)=∽ 〓 ⁱ⁼ ∽｜C _2i ｜e ^j 〓 ⁱ e ^j2 〓 ^fit ( 9), and formula (7a) is W^ _no =∽ 〓 〓 ⁱ⁼ ∽〔｜c _1i ｜+(｜c _2i ｜−｜c _1i ｜) (m−1)N+
(n-1)/M・N〕×e ^j 〓 ⁱ e ^j2 〓 ^fit (10a) =∽ 〓 ⁱ⁼ ∽C^mni e ^j2 〓 ^fit (10b) where c^ _noi = {｜c _1i ｜＋( ｜c _2i ｜−｜c _1i ｜) (m−1
)N+(n-1)/M・N}×e ^j 〓 ⁱ (10c). Equation (10) is the instantaneous Fourier coefficient Wmni of Wmn
This shows that the amplitude of -C _1i - approaches -C _2i - gradually and continuously. FIG. 7b illustrates this state, in which complex Fourier coefficients are expressed as vectors on a complex plane. As can be seen from this example, if the phases of the first and second discrete waveforms are adjusted in advance so that the phases of the corresponding harmonics are equal, the amplitude envelope of each harmonic can be approximated by a broken line. be able to. For example, Figure 9 shows the amplitude envelope of a certain musical instrument's sound up to the fifth harmonic. When approximating,
A waveform with an amplitude spectrum at point P is used as the first discrete waveform, and a waveform with an amplitude spectrum at point P is used as the second discrete waveform.
Using a waveform with an amplitude spectrum at points,
This is achieved by equalizing the phase of each corresponding harmonic of the first and second discrete waveforms. In addition, if you want to approximate the amplitude envelope of each harmonic between points P and Q of a musical tone with amplitude modulation or tremolo as shown in FIG. A waveform with an amplitude spectrum at point P is used as the second discrete waveform, a waveform with an amplitude spectrum at point Q is used as the second discrete waveform, and the phase difference between each corresponding harmonic of the first and second discrete waveforms is This can be achieved by making appropriate choices. As can be seen from Figure 7a, when there is a phase difference between the corresponding harmonics of the first discrete waveform and the second discrete waveform, |Cmni
This is because | once becomes smaller than |C _1i | and then approaches |C _2i |, and the shape of this curve can be selected by appropriately setting the phase difference. Furthermore, as can be seen from Figure 7a, if the phase of a corresponding harmonic of the second discrete waveform is ahead of the phase of a certain harmonic of the first discrete waveform, then the corresponding harmonic of the generated waveform The phase of the wave gradually advances, and the frequency of the harmonic becomes slightly higher. Conversely, if the phase of a corresponding harmonic of the second discrete waveform lags behind the phase of a certain harmonic of the first discrete waveform, the phase of the corresponding harmonic of the generated waveform will gradually lag. , the frequency of the harmonic becomes slightly lower. By utilizing this fact, it is possible to impart vibrato or non-harmonic properties to the generated sound. In addition, in the above embodiment, the contents of the multiplier memory are set to output the same value as the output of the bit shifter 5 which is the address input as shown in Table 1, so as shown in FIG. 5b, W _2o The weight given to −W _1o increases linearly, but the manner in which this weight increases can be set in any way by changing the contents of the multiplier memory. When approximating the amplitude envelope to the Q point, curve approximation can be realized instead of linear approximation. In this embodiment, as shown in FIG. 2, the first discrete waveform and the second discrete waveform are stored separately in the waveform memory 2 and the waveform memory 3. The memories 3 may be different memory areas of the same memory. In addition, in the description of this embodiment, a case has been described in which the output discrete waveform is generated from two waveforms, the first discrete waveform and the second discrete waveform. For example, the waveform at point P in FIG. After generating an output discrete waveform consisting of M/N samples using the waveform at point Q as the second discrete waveform, the waveform at point Q is the first discrete waveform, and the waveform at point R is the second discrete waveform. In the step of generating an output discrete waveform as a discrete waveform of
It is also possible to update the discrete waveform of . In this way, a musical tone can be obtained in which the harmonic time envelope is approximated by a polygonal line as shown in FIG. Therefore, the time envelope of the harmonics shown in Fig. 9 is
When obtained from the musical tones of a natural musical instrument, the generated tones are very similar to the original musical tones. In addition, although a method for generating an output waveform from two discrete waveforms has been disclosed here, the number of original waveforms is 2.
It is not necessary to limit the output waveform to three or more discrete waveforms, and the output waveform can be generated in a similar manner from three or more discrete waveforms. Effects of the Invention As is clear from the above description, the present invention extracts a plurality of waveforms each having a length of 1 pitch from different points in time of a musical sound waveform, and adjusts the phases of harmonics included in each of the extracted waveforms in a temporal manner. Processing is performed so that the phase difference between harmonics of the same order in adjacent waveforms takes a desired value, and among these processed waveforms, two temporally preceding ones are sequentially used as a first waveform and a second waveform. Then, subtract the first waveform from the second waveform to obtain a difference waveform, multiply this difference waveform by a weight whose value gradually increases from approximately zero to approximately 1, and use this multiplication result as the first waveform. Further, when the value of the weight becomes approximately 1, the second waveform is newly set as the first waveform, and the processed waveform is added to the output waveform instead of the second waveform. Since the waveform that temporally follows the second waveform is set as the new second waveform, the movement of the harmonic spectrum of the musical tone can be accurately approximated using linear approximation or curve approximation, which makes it easier to synthesize musical tones. You can easily obtain a synthesized sound that is very close to the original sound.

[Brief explanation of the drawing]

Fig. 1 is a waveform diagram for explaining the problems of the conventional waveform generation method, Fig. 2 is a block diagram showing an embodiment of the present invention, and Fig. 3 is the output of the timing pulse generator 1 in Fig. 2. A waveform diagram of the waveform, Figure 4 is a conceptual diagram used to explain the operation of Figure 2, and Figure 5 is a waveform diagram.
The figure shows a waveform diagram of the output signals of the waveform memory 2, waveform memory 3, and multiplier memory 6 in Figure 2, Figure 6 is a block diagram of the bit shifter 5, and Figure 7 shows the complex Fourier coefficient of a certain harmonic of the generated waveform. FIG. 8 is a waveform diagram for explaining the effects of the present invention, and FIGS. 9 and 10 are waveform diagrams of amplitude envelopes of harmonics of musical tones used to explain the application method of the present invention. It is. 1... Timing pulse generator, 2, 3... Waveform memory, 4... Subtractor, 5... Bit shifter,
6... Multiplier memory, 7... Multiplier, 8... Adder.

Claims (1)

[Claims] 1. A plurality of waveforms each having a length of 1 pitch are cut out from different points in time of a musical sound waveform, and the phase of harmonics included in each cut-out waveform is determined by comparing the phase of harmonics of the same order in temporally adjacent waveforms. Processing is performed so that the phase difference takes a desired value, and two of these processed waveforms that precede each other in time are sequentially designated as a first waveform and a second waveform, and the second waveform is then processed as the second waveform. 1 waveform is subtracted to obtain a difference waveform, this difference waveform is multiplied by a weight whose value gradually increases from approximately zero to approximately 1, and this multiplication result is added to the first waveform to obtain the output waveform. , Furthermore, when the value of the weight becomes approximately 1, the second waveform is newly set as the first waveform, and instead of the second waveform, the second waveform of the processed waveform is temporally A waveform generation method characterized in that a subsequent waveform is newly set as a second waveform. 2. The method according to claim 1, characterized in that the phase of harmonics included in the cut out waveform is processed so that the phase difference between harmonics of the same order in temporally adjacent waveforms becomes approximately zero. Waveform generation method.