JPS5860798A - Voice synthesization system - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a speech synthesis method, and particularly includes a memory for storing data of tone waveforms (voiced sound waveforms) and data of noise waveforms (unvoiced sound waveforms) sampled at predetermined time intervals; The present invention relates to an improvement of a speech synthesis method in a speech synthesis device that executes a predetermined arithmetic processing based on data read from this memory.

The sounds of musical instruments and human voices are composed of a fundamental frequency (pitch) that determines the pitch (scale), a tone waveform that determines the basic characteristics of the timbre, an envelope waveform that determines changes in the amplitude of the sound, and an envelope waveform that determines the volume. The determined maximum amplitude value, the noise frequency of the unvoiced part, the noise waveform, the noise envelope waveform, and the maximum amplitude value that determines the volume of the noise can be combined as elements.

As shown in Figure 1, a conventional electronic speech synthesizer divides the oscillation output of an oscillation circuit 1010, which is composed of a crystal oscillation circuit or a CR oscillation circuit, into a frequency value required for synthesizing speech. The frequency is divided by a frequency dividing circuit 102, and an envelope waveform generated from an envelope waveform generating circuit 104 is added to the output thereof and synthesized.

However, since the output of the frequency dividing circuit 102 is a square wave, there is a drawback that only a voice having a simple tone waveform, such as before bell imitation, can be synthesized. On the other hand, in order to synthesize voices with very complex tone waveforms and noise parts, it is possible to use multiple oscillators with different oscillation frequencies, and select and extract the output of each oscillator according to the desired voice. There are also systems that have a built-in system, or systems that add hardware mechanisms such as filters, but these have the disadvantage that the equipment is sloppy and increases the cost.

Moreover, it was troublesome to select which oscillator's output to synthesize, and because a large number of oscillators were required depending on the type of voice to be synthesized, it was impossible to integrate the output into a semiconductor integrated circuit.
There were many flaws in the production.

SUMMARY OF THE INVENTION An object of the present invention is to provide a method that eliminates the above drawbacks and allows complex speech to be synthesized at high speed with a simple circuit configuration.

The present invention comprises four ROMs (memories dedicated to U high) that can store a large number of digital values, four address counters that specify the addresses of these ROMs, and
and an audio signal output section having a digital-to-analog conversion circuit that converts a digital audio signal into an analog audio signal based on the output obtained from this circuit. The amplitude digital value at the sampling point of the tone waveform for one cycle, the amplitude digital value at the sampling point of the tone envelope waveform, the amplitude digital value at the sampling point of the noise waveform, and the amplitude digital value at the sampling point of the noise envelope waveform. The above four address counters are assigned to tone waveform, tone envelope waveform, noise waveform, and noise envelope waveform. Read each amplitude value of the tone envelope, multiply them, and read each amplitude value of the noise and noise envelope using the noise waveform counter and the noise envelope waveform counter, multiply these, and obtain the result of the multiplication. This is a voice synthesis method characterized in that a voice digital signal based on the violet is converted into a first-order analog signal and output.

Hereinafter, one embodiment of the present invention will be described in more detail with reference to the figures.

FIG. 2 is a block diagram of main parts of a speech synthesis device showing an embodiment of the present invention. As shown in the figure, this embodiment has four largely independent ROMs. 1st ROM1
5 stores the amplitude digital value of the sampling point of the noise waveform indicating unvoiced sound. The second ROM 16 stores the amplitude digital value of the sampling point of the noise envelope waveform.

The third ROM stores an amplitude digital value obtained by sampling one period of a tone waveform representing a voiced sound at a predetermined timing. A fourth ROM stores amplitude digital values of sampling points of the tone envelope waveform. 4,10
．． A programmable counter 11 divides the count trigger signal given from the control circuit 1 at a predetermined frequency division ratio to control the reading interval of sampling data. 3
is a preset counter whose initial value is preset by the control circuit 1 when the counter contents hover and flow, and is a counter that specifies an address for one noise waveform. 5 is a counter that specifies the address of one noise envelope waveform in tOM16. 9 is the RO
This is required by the counter that specifies the address for one tone waveform in M17. Reference numeral 12 designates a counter for specifying an address for a one-tone envelope waveform in the ROM 18. The type of noise waveform stored in the ROM 15 is selected according to the contents of the address latch 2. Father, same (ROM
The type of noise envelope waveform stored in 16 is selected by the contents of address latch 6. On the other hand, R
The tone waveforms #i stored in the OM 17 are selected according to the contents of the address latch 8, and the tone envelope waveforms 111 stored in the hoMIB are selected according to the contents of the address latch 13. The contents of these address latches are created and set by the main processor depending on the speech to be synthesized. Also, the latch 7 is the amplitude level of the noise output, that is, the sound! 14 is a latch that stores data that determines the amplitude level (sound intensity) of the tone output. By setting the contents of the latch 2 as the upper address and the contents of the counter 3 as the lower address, waveform data for one period of the noise waveform is read out from the ROM 15. A similar addressing method is applied to latch 6 and counter 5, latch 8 and counter 9, and latch ↑3 and counter 12, and data for one cycle each of the noise envelope, tone waveform, and tone envelope is stored in ROM 16, ROM 17, and ROM 18. is read from. 19 is a switching circuit that selects and outputs one of the three inputs. 20 is 1 out of 4 people
This is a switching circuit that selects and outputs two inputs. 21)
A latch 22 stores the positive and negative signs of the tone waveform from the ROM 17. Since each waveform is a pseudo sine waveform, these positive and negative signs are set so that positive and negative can be distinguished.

23 is a digital multiplication circuit. 24 is a latch that temporarily stores the multiplication results, and after the noise and tone waveform data and the respective envelope data are multiplied, the data is further multiplied by the amplitude value data (data of latch 7.14) indicating the volume to a predetermined value. It is prepared for creating audio digital data that indicates the strength of the sound. 25 is a latch that stores the amplitude of the noise output. 26 is a latch that records the amplitude of the tone output. 27 is a switching circuit for selecting one of the two inputs. 28 is a circuit that performs addition 14 or subtraction depending on the contents of the code latches 21 and 22. 29 is a DA converter for converting the obtained audio digital signal into an analog signal and transmitting the audio output to the speaker.

Next, a method of synthesizing speech using the circuit shown in FIG. 2 will be explained. Figures 3 and 5 are waveform diagrams showing examples of tone waveforms and noise waveforms for one period.The tone waveform and noise waveform are sampled into 32 divisions along the time axis t, and each sampling point is The amplitude value at is encoded as a binary digital value into 7 bits, and 1 bit of positive/negative sign is added to the resultant data, which is stored in the 1-LOM 17 as a total of 8 bits of sampling data. FIG. 4 and WJ6 are waveform diagrams showing examples of tone envelope waveforms and noise envelope waveforms. One envelope waveform is sampled by dividing it into 16 parts along the time axis t, and the resulting radial signal is encoded into 5 bits as a binary digital value.
It is stored in a predetermined address area of M16 and M18. In this example, counter 3.9 requires 5 bits and counters 5 and 12 require 4 bits.

At the start of speech synthesis, the control circuit 1 resets the latches 25 and 26, and sets the address latches 2, 6, 8, and 13 as shown in FIG.
The upper bits of the ROM address storing the waveforms shown in FIGS. 4, 5, and 6 are set, and data of the maximum amplitude value consisting of 3 pits of the binary code is set in the latch 7.14.

First, the case of outputting a tone waveform will be explained. If the count trigger signal frequency of the programmable counter 10.11 is set to IM Hz, the frequency division ratio of the programmable counter 100 of the tone waveform is selected to 100, and the frequency division ratio of the grammatical pull counter 11 of the tone envelope waveform is selected to 12800, the tone The counter 9 that selects the address that contains the amplitude value of the waveform advances one address every 0.1 milliseconds, and the counter 12 that selects the address that contains the amplitude value of the tone envelope waveform advances the address by 1 every 0.1 milliseconds. The address will be advanced by one each time. When the contents of the counter 9 that selects the address containing the tone waveform change, the switching circuit 19 is switched to input the tone waveform amplitude output of the ROM 17 to the incoming power circuit 23, and its positive/negative sign output is sent to the latch 22. input.

Next, the switching circuit 20 is switched, and the high-order 7 bits of the 0 multiplication result for inputting the tone envelope waveform amplitude output of the ROM 18 to the multiplication n1713J path 23 are stored in the launch 24. Next, the switching circuits 19 and 20 are switched so that the data of the latch 24 and the latch 14 are given to the multiplication circuit 23,
The top seven bits of the multiplication result are input into a latch 26 that stores tone output amplitude data. Thereafter, the switching circuit 27 is switched so that the contents of the latch 25 are applied to the addition/subtraction circuit 28. As a result, the previously calculated noise data and the currently calculated tone data are added or subtracted using the positive and negative signs given from the sign latches 21 and 22, and as a result, the noise and tone are synthesized. The calculation result is converted into an analog signal via an 8-bit DA converter 30 and output.

The above operation is executed every time the contents of the counter 9, which selects the address space containing the amplitude value of one tone waveform, changes. In this way, in the analog signal output of the FA converter 30, as shown in FIG. The period is 12.8
(milliseconds) of audio output.

Next, the case of outputting a noise waveform will be explained. Set the count trigger signal frequency of programmable counter 4 to IMHz.
If the noise envelope waveform program pull counter 40 frequency division ratio is set to 5000, the counter 5, which selects the address containing the amplitude value of the noise envelope waveform, will advance the address by one every 5 milliseconds.

Set the count trigger signal of preset counter 3 to 40KH.
2, and when the counter contents overflow, a random 5-bit binary number is set from the control circuit 1 to randomly change the repetition period of the noise waveform in accordance with the actual noise. When the contents of the counter 3 that selects the address containing the noise waveform change, the switching circuit 19 is switched to input the noise waveform amplitude output of )t OM2S to the multiplication circuit 23, and the positive and negative sign outputs are output to the latch 21.
Enter. Next, switch the switching circuit 20 and
The upper 7 bits of the 0 multiplication result of inputting the noise envelope waveform amplitude output to the multiplication circuit 23 are stored in the latch 24. Next, the switching circuits 21 and 22 are switched so that the contents of the latches 24 and 7 are given to the multiplication circuit 23, and the upper 7 bits of the multiplication result are put into the latch 25 which stores the amplitude of the noise output. After that, the latch 26 that stores the previously calculated tone data is connected to the addition/subtraction circuit 2.
The switching circuit 27 is switched so as to connect 8 to 8, and the code latch 21 and 220 code data are used to add or subtract the previous tone data and the current noise data and calculate the result to the DA converter 29. Convert it to an analog signal and output it.

If the above operation is performed every time the contents of the counter 3 that selects the address containing the amplitude value of the noise waveform changes,
As shown in Fig. 8, noise amplitude data is output every 25 microseconds to the analog signal output of the DA converter 29, and one noise envelope waveform is output as a noise output for 80 milliseconds (sampling period 5 milliseconds). is obtained.

In the above operation, the noise data read from the latch 25 when creating the tone data is zero, while the tone data read from the latch 26 when creating the noise data is zero, and the addition/subtraction circuit 28 simply adjusts the noise data according to the sign of the tone and noise. Create only each tone and noise data.

The case where the tone waveform and noise waveform are generated separately has been described above, but counter 3. ．． 1,5,9゜10.11
．． 12, simultaneously give the required data ' from the control circuit l,
Every time the contents of the counter 3 and the counter 9 change, the above noise creation and tone f'l = creation operations are performed, the tone output latch 26 and the noise output latch 25 store the mu results of the tone output and the noise output, and the code lamp 21H
22, the tone output and noise output are added together (1), and the tone waveform T and noise waveform N can be combined and output. In this case, the beginning is silent i4
A consonant that switches to a voiced sound midway through ° can be synthesized as shown in Figure 9. Moreover, in the vicinity S of the boundary between unvoiced and voiced sounds, both are synthesized to obtain clear audio data that is close to natural sounds (f), and no skipping or the like occurs.

In this way, this implementation? According to 1i, the tone and noise sampled from the original waveform of the actual voice and their envelope data are stored in the ROM 15.1i. Since an acoustic signal can be generated continuously by starting from t6°17.18 and performing an inexpensive multiplication process, complex sounds including voiced and unvoiced sounds can be synthesized at an open speed. Tone and noise and their envelope data are stored in ROM15, 16,
Since they are stored separately in 17 and 18, LOM address specification (from the control circuit 1) becomes extremely simple. In particular, even if the reading timings of tone data and envelope data overlap, there is an advantage that addressing can be performed independently without mutual influence, and processing speed can be increased. If you want to change the shape of the tone waveform while the waveform is being output, you can easily do so by setting the contents of the address latch 8 to the address containing the desired waveform. Furthermore, the fundamental frequency can be easily changed by changing the division ratio of the programmable counter 10. The envelope waveform can also be easily switched to another envelope waveform by changing the contents of the address latch 13. Furthermore, since such a configuration does not require digital filters or complicated arithmetic circuits, it is extremely suitable for integration, and voiced and unvoiced sounds can be freely mixed to synthesize speech that approximates natural sounds. Furthermore, by creating the order of addresses containing the tone waveform, its envelope waveform, noise waveform, and its envelope waveform as a program, it is possible to synthesize continuous speech under the program "rb11-".

In addition, in the above embodiment, ROMI 5,16°17
．． 18 storage bits, 29 DA converters, 3 counters,
4, 5, 9, 10, 11, 12, latch 2.6.7.8
．． The number of moxibustion bits of 13.14, 24.25°26 can be selected as appropriate depending on the sophistication of the speech to be synthesized.

[Brief explanation of the drawing]

FIG. 1 is a configuration diagram of a conventional electronic sound synthesizer, FIG. 2 is a configuration diagram showing an embodiment of the present invention, FIG. 3 is a waveform diagram showing tone waveforms and sampling points, and FIG. 4 is a tone diagram showing a tone waveform and sampling points. Figure 5 is a waveform diagram showing the envelope waveform and sampling points, Figure 5 is a waveform diagram showing the noise waveform and sampling points, Figure 6 is a waveform diagram showing the noise envelope waveform and sampling points, Figure 7 is the waveform diagram of the DA converter. A waveform diagram illustrating the first part of the tone signal output, Figure 8 is a waveform diagram illustrating the first part of the noise signal output of the DA converter, and Figure 9 shows the noise envelope and tone envelope on the same time axis. This is the audio waveform diagram when superimposed on and synthesized.0 1...Control circuit, 2...Noise waveform address latch, 3...Noise waveform preset address counter, 4. ...Programmable counter, 5...Noise envelope binary or polynomial address counter, 6...Noise envelope address latch, 7...Noise maximum amplitude value latch, 8・・・・・・Tone waveform address latch, 9・・・・・・Tone waveform binary or variable input address counter, 10.11・・・Programmable counter, 12・・・Tone envelope binary or polynomial address counter, ]
3...Tone envelope address latch, 1
4...Tone maximum amplitude value latch, 15, 16,
17, 18...ROM, 19°20...
・Switching circuit, 21... Noise code latch, 22
...Tone code latch, 23... power!
1L circuit, 24... Multiplication result latch, 25.
...Noise output latch, 26...Tone output latch, 27...Switching circuit, 28...
...Adjustment true circuit, 29...l) A transformer.

Claims (1)

[Claims] storage means for independently storing waveform data of voiced sounds and their envelope data, and waveform data of unvoiced sounds and their envelope data, respectively; and four reading means for independently reading each of the data from the storage means of the third generation. , comprising means for multiplying the waveform data and its envelope data to create digital signals of voiced and unvoiced sounds, and means for converting the created digital signal into an analog signal,
2. A voice synthesis method, characterized in that the waveform data and the envelope data are read out simultaneously using two of the two readout means, and these are multiplied by the generation means to obtain voice output data.