JPH0690632B2 - Noise generator - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to speech synthesis, and in particular, arithmetic processing determined based on data read from a memory that stores waveform data sampled at predetermined time intervals. The present invention relates to a noise generation circuit of a speech synthesizer for executing the.

[Conventional technology]

Conventionally, in order to synthesize a voice, the data obtained by periodically sampling the original sound is stored in a memory, the arithmetic processing determined based on the data read from the memory is executed, and finally the DA converter is used. A method of synthesizing analog signals is known.

[Problems to be solved by the invention]

In the above-described conventional voice synthesis method, the voice noise waveform has a higher frequency than the tone waveform and is a random waveform. Therefore, if the noise waveform is to be sampled faithfully, the sampling interval must be made very small. As a result, there is a drawback that the amount of sampling data becomes huge and the memory area assigned to the tone waveform is limited. Furthermore, voice noise is
As shown in the figure, the frequency spectrum is a continuous spectrum having a peak at a specific frequency depending on the type of voice. The peak frequency of this frequency spectrum is usually around 8 KHz for voice noise in the Sa line and around 4 KHz for voice noise in the line. In order to generate a synthetic noise waveform having a spectrum peak at a specific frequency and forming a continuous spectrum in this way, a very complicated pseudo voice noise generation circuit has been conventionally required.

[Means for solving problems]

The noise generation circuit of the present invention includes a memory in which a part of a waveform obtained by sampling an original sound of voice noise is stored, a polynomial counter for designating an address of the memory, and a random number for setting an initial value in the memory addressing polynomial counter. It has a generating polynominal counter, a matching circuit for detecting that the memory address matches a specific value, and a DA converter for converting the output of the memory into an analog signal.

An audio noise waveform with a spectral peak at a specific frequency has waveform components with a frequency component substantially equivalent to the peak frequency when viewed temporally, and frequency components higher and lower than the peak frequency are mixed. , Its amplitude is considerably smaller than that of the peak frequency component. In order to synthesize a voice noise waveform having such characteristics, repeating the waveforms of peak frequency components results in a single spectrum. Therefore, in the present invention, several waveforms are extracted from the original sound. However, although a line spectrum having a frequency component of the peak frequency of the original sound, which is repeated as it is, is obtained, it is not a continuous spectrum.
Therefore, when repeating several waveforms extracted from the original sound, it was possible to generate a waveform with a continuous frequency spectrum by randomly repeating the start position of the waveform.
For example, a polynomial counter can be used to randomly set the start position. In the present invention, the memory address is designated by the polynomial counter so that the noise waveform data extracted from the original sound is output after the start position is set. The start position can be set randomly by setting the data obtained by inverting the content of the random number generating polynomial counter as the initial value of the memory addressing polynomial counter.

〔Example〕

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a circuit configuration diagram of an embodiment of the present invention. Address input terminals 2 and 3 specify the upper address of memory 4,
The lower address of the memory 4 is designated by the polynomial counter 7. Four types of noise waveforms can be selected by switching the inputs of the address input terminals 2 and 3. The original sound noise waveform as shown in FIG.
Store 31 points of sampling data sampled for each c. The polynomial counters 7 and 9 in the circuit diagram of FIG. 1 are both 5-bit polynomial counters shown in FIG. 2, and the contents of these polynomial counters are the same as the memory addresses shown in FIG. Changes. When the output of the polynomial counter is inverted by the inverter, the data changes in the order written on the vertical axis in FIG. A 40 KHz clock is input to the clock input terminal 1 of the memory addressing polynomial counter 7, and the coincidence circuit 8 detects that the value of the polynomial counter 7 matches 0 or 31. When they match, the content of the random number generating polynomial counter 9 is moved by one by the match output of the matching circuit 8, and the value is inverted by the inverter and set in the memory addressing polynomial counter 7. At this time, the start address of the original sound noise sampling waveform stored in the memory 4 is set in the memory addressing polynomial counter 7. FIG. 4 shows a change in the position of the start address of the original sound noise sampling waveform corresponding to the movement of the polynomial counter for generating random numbers. The vertical axis shows the change of the random number generation polynomial counter inversion output from top to bottom, and the horizontal axis shows the memory address start position at that time. It can be seen that the memory address start position changes randomly.

When the start address is set in the memory address designation polynomial counter 7, the coincidence output of the coincidence circuit 8 is stopped, and the polynomial counter 7 sequentially designates the address of the memory 4 with the clock of 40 KHz. The original sound noise sampling waveform appears as the output of the memory 4. Repeat this operation to output the output of memory 4 to DA
When converted into an analog signal by the converter 5, the output terminal 6 has a sixth frequency spectrum close to the frequency spectrum of the original sound noise shown in FIG.
A signal having a continuous frequency spectrum shown in the figure is obtained. The cycle of the polynomial counter 9 for generating random numbers has practically no effect on the synthesized spectrum. The match detection circuit detects 0 data as the end address of the memory, but 31 data is also detected because the inverted output of the random number generation polynomial counter contains 31 data This is because if the nominal counter contains 31 data, the contents of the polynomial counter remain at 31 even if a clock is entered to prevent it from stopping.

〔The invention's effect〕

As described above, according to the present invention, a part of the data obtained by sampling the original sound waveform is extracted and the signal having a continuous frequency spectrum can be synthesized by repeating the data. In the case of voice noise, the frequency with which the frequency spectrum becomes maximum is as high as 8 KHz, so if the sampling interval is less than 50 μsec, data cannot be sampled well, so if the voice noise is sampled every 25 μsec, it will take 1 second. The number of sampling points is 40,000, and the memory needs 40,000 addresses.
Even if the combined data is output every 5 μsec, the memory capacity is only 32 addresses, and the effect of reducing the memory capacity is remarkable. In addition, noise having a different frequency spectrum can be synthesized by increasing the memory capacity as needed. In particular, since a polynomial counter is used to specify a memory address, when a semiconductor integrated circuit is used, the layout occupying area can be reduced and the chip size can be reduced as compared to the case where a binary counter is used because the circuit is simpler. effective. Furthermore, since a signal with a continuous frequency spectrum can be configured with a simple circuit using a periodic clock, it is suitable for digital LSI, and since the data sampled from the noise waveform of the original sound is used, it has a frequency spectrum close to the original noise. Synthesized noise is obtained, and high-quality speech noise can be synthesized. Note that the number of stages of the polynomial counter, the clock input frequency of the memory addressing polynomial counter, the number of output bits of the memory, the number of bits of the DA converter, etc. are each related to the sound quality of the synthetic noise,
Obviously, a suitable value may be selected.

[Brief description of drawings]

FIG. 1 is a circuit configuration diagram of an embodiment of the present invention, and FIG.
FIG. 3 is a circuit diagram of a concrete example of the polynomial counter shown in FIG. 3, FIG. 3 is a data waveform diagram obtained by sampling an original sound noise waveform to be stored in the memory 4 of FIG. 1, and FIG. 4 is a memory for an inverted output of a random number generating polynomial counter. FIG. 5 is a diagram showing the position of the start address, FIG. 5 is a frequency spectrum waveform diagram of original noise, and FIG. 6 is a frequency spectrum waveform diagram of synthetic noise. 1 ... Clock input terminal, 2.3 ... Memory upper address input terminal, 4 ... Memory, 5 ... DA converter, 6 ... Output terminal, 7 ... Memory addressing polynomial counter, 8 ... Match Detection circuit, 9 ... Random number generating polynomial counter, 10 ... Flip-flop, 11 ... Exclusive NOR.

Claims (1)

[Claims] 1. Storage means for storing at least a part of a waveform obtained by sampling a noise original sound of voice, and a first polynomial counter for periodically generating an address for reading the storage information from the storage means. , A second polynomial counter, and when the value of the first polynomial counter reaches a minimum value or a maximum value, the inverted value of the second polynomial counter is transferred to the first polynomial counter. A noise generation circuit characterized by being set as an initial value.