JPH02271398A - Noise sound generating device - Google Patents

Abstract

PURPOSE:To generate a synthetic sound which has a sharp peak in the center of a formant shape by adding a DC component to a noise waveform which has specific spectrum characteristics and multiplying the result by the period waveform of specific frequency. CONSTITUTION:A white noise from a white noise generator 1 becomes a white noise with a right-down waveform to which the DC component is added by an adder which performs offset processing and a digital filter 3 which performs spectrum processing for giving low-pass characteristics. A multiplier 5 multiplies this noise by the waveform such as a sine wave of the period waveform corresponding to the specified center frequency of the formant outputted by a period waveform generating circuit 4. Consequently, the noise having the sharp peak in the center of the formant shape is generated by frequency convolution and the center frequency of the formant shape can be varied optionally and easily.

Description

[Detailed Description of the Invention] Field of Application in the CA Industry] This invention is a method for electrically synthesizing noise sounds having a formant shape such as unvoiced sounds (consonants), whistling sounds, or attack parts of natural musical instruments. Regarding sound generators,
In particular, the present invention relates to a noise generation device capable of synthesizing noise having spectral characteristics with steep formant peaks.

[Prior Art] An unvoiced sound (consonant), a whistling sound, or an attack sound of a natural musical instrument has a formant shape in which the amplitude spectrum distribution has a peak at a specific frequency.

Conventionally, noise sounds having such a formant shape have been realized by using a bandpass filter or the like having a resonance point at a specific frequency.

By the way, the process of a voice transitioning from an unvoiced sound (noise sound) to a voiced sound (vowel or formant sound) (formant locus), the process of changing the pitch of a whistling sound, or the process of changing the pitch of a natural instrument from the attack part to the stationary or decay part (formant sound) When observing the process of transition to (part), the peak frequency of the formant shape (hereinafter referred to as formant center frequency) changes continuously. Especially, 1 from voiceless to voiced sounds! ! When the voice continues, the formant center frequency of the unvoiced sound changes continuously to the formant center frequency of the voiced sound.

Therefore, in order to form faithful synthetic sounds such as human voices, whistle sounds, and natural instrument sounds, it is necessary to continuously change the formant center frequency of unvoiced sounds.

However, it is difficult to continuously change the resonant frequency of a bandpass filter in real time, especially in a digital filter. That is, in the conventional method described above, it is difficult or impossible to finely and arbitrarily shift the formant center frequency.

Another method for synthesizing a noise sound having a formant shape is to multiply a noise waveform having a downward sloping spectrum characteristic as shown in FIG. 2(b) by a periodic waveform (for example, a sine wave) having a frequency fo. A method (hereinafter referred to as the preceding method) can be considered. In this case, a noise sound having a frequency fo of a periodic waveform as shown in FIG. 2(c) as a center frequency and having an amplitude spectrum of a formant shape in which the above-mentioned right-sloping shape appears symmetrically is formed. According to this prior method, the conventional method using a bandpass filter is 1/1
A filter with a 2nd order falling characteristic on one side (to the right) can synthesize noise sounds that fall to the left and right as in the conventional example,
Since the order of the filter is small, it is possible to perform high-speed calculations that can be considered to be substantially real-time.

The basic principle of this advanced method is the frequency convolution theorem that "convolution in the frequency domain can be transformed into a product in the time domain."

Figure 1 shows this frequency convolution theorem with a cosine waveform h (B
and a square waveform x (t).

In the figure, the hexagon of the 2m line indicates that the two above and below are Fourier transform pairs.

From Figure 1, by performing the multiplication h (t) *x (t) in the time domain, the cosine wave spectrum H (f) and the square wave spectrum X (f) are convolved in the frequency domain. It is understood that

By the way, the spectrum of the noise sound actually synthesized by this advance method has a formant center frequency f as shown in FIG. 2(f). It becomes what is missing. That is, with the previous method, it was not possible to obtain a noise sound that has a sharp peak in an arbitrary frequency band, such as a whistle.

[Problems to be Solved by the Invention] The present invention provides a noise sound generation device that can synthesize a noise sound having a sharp peak at the center of a formant shape and can arbitrarily and easily adjust the formant center frequency. The purpose is to

[Means for Solving the Problems] In order to achieve the above object, the noise sound generation device of the present invention includes a noise signal generation means for generating a noise signal waveform having a downward sloping spectrum characteristic, and a DC voltage applied to the noise signal waveform. The apparatus includes DC component adding means for adding a DC component, periodic function generating means for generating a periodic waveform of a predetermined frequency, and multiplication means for multiplying the noise signal to which the DC component is added and the periodic waveform.

[Operation] In the present invention, a noise waveform having a right-sloping characteristic is multiplied by the above-mentioned periodic waveform to synthesize a noise sound. In this case, as described above, the formant center frequency is determined by the periodic waveform, and can be arbitrarily and easily varied by varying the frequency and wave number of this periodic waveform.

As a conventional digital noise generator, for example, a white noise generator to which an M-sequence generation method and a multiplication congruence method are applied is well known.

The M-sequence generation method is composed of a 1-bit N-stage shift register including a feedback loop, and generates an M-sequence code with one period of 2N-1 bits.For example, the frequency of the shift register driving clock is set to 50kHz. Then, the repetition frequency of this M-sequence code is approximately 0.045Hz when N-20, and approximately 0.0OO023 when N=30.
Hz.

The multiplicative congruence method is Xy1+1 xax, (+od 2
This is a well-known random number generation method in a computer that generates noise waveform data by calculating ``).In the above equation, S is the calculation precision (number of bits), and ax,
(gaod 2") indicates the remainder when ax is divided by 2'. Also in this multiplication congruence method, the lowest frequency is determined by the calculation rate and the number of bits S, and is of approximately the same order as in the above M-sequence generation method.

Either method does not include a DC component with a frequency of 0, so the noise sound obtained by multiplying this noise waveform has f. A component of ±0, that is, a formant center frequency component is not included.

In this invention, since a DC component is added to the noise waveform having a downward-sloping spectrum before multiplication, this noise waveform has a steep spectrum distribution toward the DC side, as shown in FIG. 2(d). The formant shape in which this spectral distribution appears symmetrically around the formant center frequency fo also has a steep peak as shown in FIG. 2(e).

In one aspect of the invention, first, white noise is generated. Such white noise can be generated relatively easily by applying the above-mentioned M-sequence code generation method, multiplication congruence method, or the like. Further, in this embodiment, next, the spectral characteristics of the generated white noise are processed in a downward-sloping manner using a spectral processing means such as a low-pass filter. As such a low-pass filter, for example, a digital filter can be used, and by providing a small number of parameters, the slope of the downward-sloping characteristic, the cutoff frequency, etc. can be relatively freely and easily varied.

[Example] Hereinafter, this invention will be explained in detail based on examples.

FIG. 3 is a block diagram showing the configuration of a noise generating device according to an embodiment of the present invention.

The device shown in the figure is used, for example, as a sound source for an electronic musical instrument, and includes a white noise generation circuit 1, an adder 2, a digital filter 3, a periodic waveform generation circuit 4, multipliers 5, 7,
Envelope waveform generation circuit 6 and accumulator 8
etc.

The white noise generation circuit 1 generates white noise, that is, noise having a flat spectrum as shown in FIG. 2(a). This white noise generation circuit 1 may be of any type as long as it generates white noise.
For example, the M-sequence generator described above can be used.

The adder 2 adds an offset constant to the white noise data output from the white noise generation circuit 1, thereby adding a DC component to the white noise isometrically. This offset constant may be subtracted; in this case, a negative DC component will be added, but since the amplitude appears as an absolute value, there is no difference depending on the sign of the DC component.

The digital filter 3 includes adders 31, 32, 33, a multiplier 34, an inverter 35, and a delay circuit 36. This digital filter 3 is of a type called an IIR filter, and here forms a low-pass filter with a transmission frequency H(z) expressed as follows.

This IIR filter has only one parameter (ew
), the degree of spectral envelope can be controlled, and it can be attenuated fairly steeply using simple hardware, making it suitable for obtaining formant shapes similar to those of ordinary unvoiced sounds. FIG. 2(b) shows the amplitude spectrum of the noise signal outputted from the digital filter 3 and having a downward-sloping spectral characteristic. Note that by using a high-order FIR filter as the digital filter 3, it is also possible to more finely control the formant shape (sloping spectrum characteristic).

The periodic waveform generation circuit 4 includes a pitch data generator 41, a selector 42, a phase accumulator 43, a sine wave memory 44, and the like. The pitch data generator 41 generates pitch information values that change according to a function according to pitch data parameters set inside the device. Selector 42
is for selecting one of the internal pitch information value output from the pitch data generator 41 and the external pitch information value input from the external device. Accumulator 43
is the selector 4 in synchronization with a predetermined clock pulse φ.
The pitch information value F0 output from 2 is accumulated. Here, the pitch information value F0 is set to a value corresponding to the formant center frequency f0 of the noise sound to be generated and its change, and the accumulator 43 accumulates (aq
F.

(q = 1.2, 3...) as a sine wave 44
are sequentially output as read address signals. In the sine wave memory 44, sequential sample point amplitude values of one cycle of the sine wave are stored at each address, and are stored at the address specified by the address signal (accumulated value qFo) output from the accumulator 43. The sine wave amplitude values are read out sequentially. As a result, the sine wave memory 44 sequentially outputs sample point amplitude values sin 2πfot of a sine wave having a frequency f0 that transitions corresponding to the pitch information value F0 at each time point in accordance with the clock pulse φ.

As described above, the periodic waveform generating circuit 4 can use the external pitch information value, so that it can match or be linked to the pitch of the formant sound synthesized by the sound source of the voiced part and its fluctuation. Alternatively, the formant center frequency of the noise sound can be made to follow the keyboard pitch by providing m-keyboard pitch information as this external pitch information value.

In the multiplier 5, a noise signal having right-sloping spectrum characteristics outputted from the digital filter 3 and a sine wave amplitude value sin outputted from the periodic waveform generation circuit 4 are combined.
Multiply by 2πf, t. As a result, a noise sound signal having a formant shape is obtained as the output of the multiplier 5. In this case, the attenuation characteristic of the digital filter 3 described above on the frequency axis directly corresponds to the band width of the formant shape. That is, the frequency characteristic curve of the digital filter 3 becomes symmetrical with respect to the formant center frequency, forming a firmant shape.

The envelope waveform generating circuit 6 generates envelope data ENV representing the amplitude of the noise sound at successive points in time at the timing of the clock pulse φ in accordance with envelope parameters set inside the device.

The multiplier 7 is for adding an envelope as a musical tone to the noise sound signal output from the multiplier 5, and multiplies the output of the multiplier 5 by the envelope waveform ENV output from the envelope waveform generation circuit 6. do. As a result, a formant-like noise sound whose amplitude changes over time according to the envelope waveform is obtained as the output of the multiplier 7.

In the apparatus shown in FIG. 3, a plurality of noise sounds can be synthesized in parallel by using the digital filter 3 or multiplier 7 in a time-division manner. The accumulator 8 is for accumulating these plurality of noise sounds in terms of data and mixing them acoustically.

The noise sound output from the accumulator 8 is converted into an analog signal by a D/A converter (not shown), and is produced as a noise fillant sound via a sound system consisting of an amplifier, a speaker system, etc. (not shown). .

Next, in the device shown in Fig. 3, the bandwidth BW is set to 0.00.
2. An example in which a noise sound is generated with the formant center frequency f0 set to 5 kHz will be explained.

Figure 4 is a waveform diagram (a to d) of a formant-like noise sound when a DC component of 1% of the total amplitude is added, and an enlarged view of its starting point (e). A diagram showing a three-dimensional representation of the amplitude of a noise sound in the frequency and time axis directions is shown.

FIG. 6 and FIG. 7 are diagrams corresponding to FIG. 4 and FIG. 5 when the DC component is 0.

Comparing Figures 4 and 5 with Figures 6 and 7, the formant center frequency f. It is clear that the temporal fluctuations of the component amplitudes are extremely reduced by just adding a small amount of the DC component.

[Scope of the Invention] Note that the present invention is not limited to the above-described embodiments, and can be implemented with appropriate modifications.

For example, the peak height of the formant spectrum is determined by the addition of DC as shown in Figure 2(d) and (e).
Since it is determined by the size of the component, it may be set appropriately in consideration of this so that a desired peak can be obtained.

In addition, in the above, a sine wave is used as the periodic waveform,
This explains an example in which a formant with only one peak is obtained, but by using a periodic waveform that includes multiple frequency components corresponding to each of multiple formant center frequencies, a formant with multiple peaks can be obtained. It is also possible to synthesize noise sounds.

[Effects of the Invention] As described above, according to the present invention, 1. It is possible to eliminate fluctuations in the amplitude of components near the formant center frequency and synthesize a formant-shaped noise sound having a sharp peak at the formant center frequency. .

2. When simulating speech or musical sounds, when a voiceless sound or attack sound (noise sound) changes to a voiced sound or stationary sound, the formant frequency of the unvoiced sound is linearly changed to match the formant frequency of the voiced sound, etc. It is possible to move in synchronization with etc.

3. Noise sounds with multiple formants can be realized by using periodic waveforms with arbitrary overtones.

4. The algorithm for realization is simple.

5. If a method that requires less calculation time for the noise spectrum processing means (for example, a -order II4 filter) is adopted, the entire calculation time will be reduced.

6. The formant center frequency can be easily controlled using each Yuzu controller (including external controllers).

Effects such as these are realized.

FIG. 3 is a block diagram of a noise sound generation device according to the present invention, FIGS. 4 and 5 are diagrams showing the waveform and amplitude spectrum of the noise sound synthesized in the device of FIG. 3, and FIG. For comparison, FIG. 7 is a diagram showing the waveform and amplitude spectrum of a noise sound when the noise sound is synthesized with the DC component set to 0 in the apparatus of FIG. 3.

1: White noise generation circuit 2: Adder (DC component addition means) 3: Digital filter 4: Periodic waveform generation circuit 5: Multiplier

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram showing the basic principle of this invention, FIG. 2 is a waveform diagram for explaining the basic operation of this invention,

Claims (2)

[Claims] (1) Noise signal generation means for generating a noise signal waveform having a downward-sloping spectral characteristic, DC component addition means for adding a DC component to the noise signal waveform, and periodic function generation means for generating a periodic waveform of a predetermined frequency. and a multiplier for multiplying the noise signal added with the DC component by the periodic waveform. (2) The noise signal generating device according to claim 1, wherein the noise signal generating means comprises a white noise generating means and a spectrum processing means for imparting a right-sloping spectral characteristic to the white noise.