JP2000066669A - Music creating device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To economically create a music not attracted with attention by an audience such as background music, without being restricted by copyright and the related. SOLUTION: A sound source device SS accumulates digitally sounds corresponding to musical instruments assigned by program numbers as forms of phonemes, the sounds are read out to get frequencies assigned by a pitch generator PG, and analog audio signals are generated with wave forms corresponding to musical instruments classified by musical instrument kinds indicated by the program numbers from a controller CNT-P, with amplitudes of musical instrument sound volumes based on signals from a controller CNT-V, and with action applied to clear sounds of percussion instruments in particular based on output trigger signals of a sigmonide function generator T in the final layer of the second neural network. Atmosphere constituting music with many elements is generated with very economical constitution free from copyright.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a music generating apparatus, and more particularly to a music generating apparatus for generating music that does not attract the attention of an audience.

[0002]

2. Description of the Related Art In recent years, with the expansion of use of environmental music, demand for music that does not particularly attract the attention of an audience has been increasing in workplaces and public places. However, even for music that does not aim for the positive emotions of the audience (it does not originally have an appealing meaning), considerable cost and labor are required to create this music software. Even if the user purchases and uses this software, there is a current burden on the user side as with normal music, such as the necessity of clearing the copyright again for public use and application to paid environments.

On the other hand, music software such as environmental music, which does not originally have an appealing meaning, does not necessarily have to rely on intentional music composition work at the stage of creation, and even in the case of music generated purely electronically, it can be used in a limited area. Is quite wide. Therefore, if it is possible to create infinite electronic music automatically without repetitive playback of content using music package media, the man-hours for package selection, purchase or rental costs, regular replacement man-hours, etc. can be reduced. It is no longer necessary to obtain permission for each case, and the range of applications can be dramatically expanded.

[0004]

Originally, music software such as environmental music which does not originally have an appealing meaning is often presented for the purpose of relieving the tension of the viewer. In general, there are many things that present slow movements, such as the sound of waves and wind, and strings made by repeating type codes (chords). In order to automatically generate environmental music electronically with this atmosphere, it is necessary that the form (melody line, chord, volume, rhythm) changes slowly, that the presented content is not simply repeated, Need to be electronically synthesized. In particular, among the above three requirements, the requirement is important. When the repetition period is detected by the viewer, interest in the music is rapidly reduced, and the presentation effect is rapidly reduced.

[0005] Under the above restrictions, tape,
The general method of presenting the slowly moving content stored on the hard disk with a so-called effect device by altering it is difficult to achieve, and the copyright problem of the original content is not cleared. Therefore, there is a need for a technology for creating purely electronic music without a composition process and a periodic presentation process.

[0006] The present invention has been made in view of the above points,
It is an object of the present invention to provide a music creating apparatus capable of creating music, such as environmental music, which does not particularly attract the attention of the audience, in a very economical manner and without being restricted by copyright.

[0007]

In order to achieve the above object, the present invention provides a first feedback type in which a chord number constituting music is newly generated from a chord number immediately before and a chord number several tempo past. A neural network, a pitch generator for converting an output code number of the first feedback type neural network into a plurality of real frequencies, a trigger signal constituting a rhythm for generating music, a performance trigger signal immediately before and a trigger signal several tempo past When it is detected that the number of cyclic chains of code numbers generated in the second feedback-type neural network newly generated from the first feedback-type neural network exceeds the first threshold value, different probabilities are obtained. A code generated from a function is converted to a first feedback neural network. A first control device provided in a feedback loop of the first feedback type neural network, which is input to a shift register of a first stage of the network, and a cyclic chain number of a trigger signal generated in the second feedback type neural network. Is detected to exceed a second threshold value, a trigger signal generated from another probability function is input to a first-stage shift register of a second feedback type neural network. A second controller provided in a feedback loop of the neural network;
A tone generator which receives each output signal of the pitch generator and the second feedback type neural network, generates and outputs an analog audio signal having a specified rhythm at a specified frequency, and first and second feedback type neural networks And a synchronization control device that operates in synchronization with each other.

According to the present invention, of the first and second feedback neural networks, each of which generates a current element from a history of a finite number of past elements, the first feedback neural network detects a code change, Since the rhythm change is generated by the feedback-type neural network of No. 2, even though there is no unreasonable leap, the current element is determined without reference to the previous contents, The determination process is not uniform, and a chord change and a rhythm change can be generated.

Further, according to the present invention, when the first and second control devices detect that the number of cyclic chains of occurrence of elements in the feedback type neural network exceeds the first and second threshold values, By changing the input conditions of the first and second feedback neural networks by inputting a code number and a trigger signal generated from another probability function, a circular chain can be broken.

Further, according to the present invention, the first and second feedback neural networks have learned weighting coefficient groups which have been learned in advance using existing music, respectively. Characterized by further comprising a switch for switching the learned weighting coefficient group in the feedback type neural network.

Further, according to the present invention, at least one of a third control device for outputting a program signal for determining a musical instrument type to a tone generator and a fourth control device for outputting a signal for determining a musical instrument volume to the tone generator is provided. Further, it has a configuration. According to the present invention, the third control device can give a change in musical instrument type, and the fourth control device can give a change in volume.

[0012]

Next, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a music creating apparatus according to the present invention. In the figure,
Each bit storage section C1 to Ci of the shift register, DA converters DA1 to DAi, weight coefficient group W1, second layer sigmonide function generators D1 to Dj, weight coefficient group W2, last layer sigmonide function generators E1 to Ek And the code generator CG includes a first two-layer neural network (N
N) 10. This is a circuit for newly generating a chord constituting the music from the immediately preceding performance chord number and a chord number several tempo earlier. In the feedback loop of the first NN 10, a code control device CNT-C is provided.

The output of the code generator CG is a newly generated code of k bits, and is fed back to the code control unit CNT-C which supplies a shift clock to the bit storage units C1 to Ci constituting the shift register.
The newly generated code of k bits which is the output of the first NN 10 having this feedback loop is also input to the pitch generator PG.

Each bit storage unit R of the shift register
1 to Rl, weighting coefficient group W3, second-layer sigmonide function generators S1 to Sm, weighting coefficient group W4, and last-layer sigmonide function generator T also constitute a second two-layer neural network (NN) 20. are doing. This is a circuit for newly generating a trigger constituting a rhythm in music generation from the immediately preceding performance trigger and a trigger a few tempos past.

The output of the sigmonide function generator T of the last layer of the second NN 10 is a digital signal indicating the presence or absence of a trigger-on signal, and the shift clock is supplied to the bit storage units R1 to R1 constituting the shift register. The trigger controller CNT-R that supplies the feedback. That is, the trigger control device CNT-R is provided in the feedback loop of the second NN.

The switch SW switches the contents of the NN-learned weighting coefficient groups W1 to W4. Furthermore,
In this embodiment, there are provided a synchronization control device CLK, a control device CNT-P for a program signal for determining a musical instrument type, a control device CNT-V for determining a musical instrument volume, and a sound source device SS.

The bit storage units C1 to Ci constituting the shift register synchronize with the timing clock based on the output pulse of the code control unit CNT-C to which the timing clock generated by the synchronization control unit CLK is input. The storage data is shifted from the bit storage unit C1 in the Ci direction. The data output from each of the bit storage units C1 to Ci of the shift register is stored in the DA converter DA1 to DAi.
Are separately converted from digital to analog by
The analog values are divided into j values, and pass through the weighting coefficient group W1 on the way to the sigmoid function generators D1 to Dj in the second layer. At this point, each signal is independently multiplied by a weighting coefficient by i × j attenuators in W1, and input to the sigmoid function generators D1 to Dj in the second layer.

Second-layer sigmoid function generators D1 to Dj
Are output in the next-stage weighting coefficient group W2.
After being multiplied by j × k weighting coefficients, they are input to the sigmoid function generators E1 to Ek in the final layer. These final layer sigmoid function generators E1 to E
The k output signals, which are analog values of k sigmoid functions, are binarized by a code generator CG and converted to k-bit digital values.

It is assumed that learning of the weighting coefficient groups W1 and W2 has been completed using existing music in advance by a known method such as the back provacation method. The recursive input to the first-stage bit storage unit C1 of the bit storage units C1 to Ci constituting the shift register is a k-bit newly generated code number which is the output of the code generator CG. , Code control device CNT
The circular chain is verified through -C.

First NN1 with feedback loop
Values generated through zero have a tendency to cycle. In this embodiment, this is a musical chord, which is relatively less of a problem for the generated music, but too much of this cyclic chain results in boring music. For this reason, when the number of chains exceeds a certain threshold value, the code control device CNT-C interrupts the circulation by inputting another code generated from another probability function to the first-stage bit storage unit C1. Work.

FIG. 2 shows the control units CNT-C and CN of FIG.
1 shows a circuit diagram of an example of TR. That is, both the control device CNT-C and a control device CNT-R described later have the configuration shown in FIG. To explain each block in the case where FIG. 2 is the control device CNT-C, SD is an output code number of the code generator CG input to the control device CNT-C of FIG. FIG. 2 does not show the data widths handled by the components of this figure including the internal shift register SR. The input SD is shown in FIG.
, Is a k-bit code number. Therefore, FIG.
When used as C, a data width of k bits is handled.

SC in FIG. 2 is a shift clock input from the outside, and is output from the synchronous control device CLK in FIG. SR is an internal shift register, which is 16 bits in this example. The data transfer direction is SR1
→ The direction is SR16. In addition, CMP1 to CMPn are comparators, and are 8 bits from the first bit of the internal shift register SR.
The outputs of the bit storage units SR1 to SR8 of the bit,
Bit storage units SR2 to SR of the 9th to 9th bits
Compare each 8-bit value such as the output of 9 and if they are equal,
An ON signal (coincidence signal) is sent to the detector DET.

The detector DET includes comparators CMP1 to CMP
n, counts the number of accumulated code chains by type, and when it exceeds a specific threshold, the gate EXC
Send the switching signal to. The gate EXC switches between outputting the input code number SD directly to the output terminal OUT and outputting the contents of the internal code generator GEN to OUT.
This is output from the code control device CNT-C in FIG. 1 and input to the first-stage bit storage section C1 of the shift register.

Since the output of the code generator CG is a code number, in order to electronically reproduce this as a real sound, a real frequency must be specified. The pitch generator PG converts the code specified by the code generator CG into a plurality of real frequencies and sends out the converted frequencies to the tone generator SS. The sound source device SS digitally accumulates a sound corresponding to an instrument specified by a program signal described later, for example, in the form of a phoneme, and reads out the sound so as to have a frequency specified by the pitch generator PG. Is transmitted from the output terminal OUT as an analog audio waveform.

Next, the configuration of the second NN 20 will be described. The shift register composed of the bit storage units R1 to R1 moves data in the direction of R1 → R1 by a shift clock generated by the synchronous control device CLK. The outputs of the bit storage units R1 to R1 are each divided into m pieces and go to the sigmoid function generators S1 to Sm of the second layer, but pass through the weighting coefficient group W3 on the way. At this point, each signal is independently multiplied by 1 × m weighting coefficients in the weighting coefficient group W3 and input to the second layer.

The output of the second layer is the weighting coefficient group W at the next stage.
At the point of time when the signal passes through No. 4, multiplication by m weighting coefficients is performed, and the result is input to the sigmoid function generator T of the final layer. These weighting coefficient groups W3 and W4 are
It is assumed that learning has been completed in advance using existing music by a known method such as a back propagation method. The output of T is a digital signal indicating the presence or absence of a trigger-on signal by binarizing the sigmoid function.

Bit storage unit R constituting shift register
The input to the bit storage unit R1 in the first stage among R1 to R1 is a newly generated trigger which is the output of the sigmoid function generator T, which once passes through the control unit CNT-R to generate a cyclic trigger chain. Is verified.

Second NN2 with feedback loop
Values generated through zero have a tendency to cycle. In this embodiment, this is a relatively trivial problem for the generated music, as it is a cyclic trigger, but too much of this cyclical chain results in boring music. For this reason, CNT-
When the number of chains exceeds a certain threshold value, R functions to terminate the circulation by inputting another trigger generated from another probability function to the first-stage bit storage unit R1. On the one hand, the output of the sigmoid function generator T is input to the sound source device SS. This is mainly applied to the sound of a percussion instrument specified by a program signal described later, and has an effect of clarifying the rhythm.

The NN learned weighting coefficient groups W1 to W1
It may not be desirable for W4 to simultaneously use all types of music as teacher data during learning. For this reason, if the coefficient group is changed according to the type of learning music such as jazz, classical music, or specific ethnic music, the generated music can be provided with a type. Therefore,
The switch SW switches between these coefficient groups.
Thereby, the type of the generated music can be selected.

The control device CNT-P is a control device for a program signal for determining the type of musical instrument, and has a probability function and a random number generation function therein. Thereby, the control device CNT-
P randomly generates a musical instrument type corresponding to the code signal of the pitch generator PG and a musical instrument type corresponding to the trigger signal of the sigmoid function generator T. At the same time, the selection probability and the change probability are defined. Therefore, there is a high probability that the selection and transition of musical instruments suitable for the atmosphere of the playing field are high, such as continuous music instruments under the control of the PG, and plucked or percussion instruments under the control of the T. Is set to The selection and the change are performed based on the output timing signal of the synchronous control device CLK. The output of the control device CNT-P is sent to the sound source device SS, and selects the instrument sound output by the sound source device SS.

The CNT-V is a control device for determining the volume of the musical instrument. Like the control device CNT-P, it has a probability function and a random number generation function inside. Thereby, each volume of the musical instrument group corresponding to the code signal of the pitch generator PG and each volume of the musical instrument group corresponding to the trigger signal of the sigmoid function generator T are respectively randomly determined. Since the probabilities are specified, the selection and transition of the volume and balance, which are suitable for the atmosphere of the music hall, are set to have a high probability. If this volume designation is particularly strongly assigned to one of the chords constituting the chord, the melody line can be clarified. The output of the control device CNT-V is sent to the sound source device SS, where the sound volume of the musical instrument output by the sound source device is selected.

Next, the operation of this embodiment will be described more specifically. In this embodiment, at the start of operation, the trigger control device CNT-R inputs a trigger signal to the first-stage bit storage unit R1 of the shift register, and the code control device CNT-C refers to the operation start time and the like. The code determined by the generated random number is input to the first-stage bit storage unit C1 of the shift register. In addition, the control device CNT-
P and CNT-V similarly input a musical instrument type group generated from random numbers and their volume value group to the sound source device SS.
Each of the above control devices CNT-R, CNT-C, CNT-P
And CNT-V are, for example, 8 from the synchronous control device CLK.
Upon receiving a timing signal in units of a minute note, they operate in synchronization with each other.

First, code generation will be described. Assuming that the code input from the code controller CNT-C is C (a major triad with C as the root), the code is stored in the first-stage bit storage unit C1 of the shift register including the bit storage units C1 to Ci. It is assumed that the newly generated k-bit code automatically generated and output from the generator CG is F. Hereinafter, similarly, as shown in FIG.
7, Cm, Fm, Fdim, C,. . . It is assumed that codes are generated in the order of.

Here, if the code generated after the last code C is again F and the following code is the same as the above-mentioned code sequence, this means that it is circulating. In that case, the circulation cycle is 6 units. Circular chords are not a problem per se, but the generated music is less variable. The code control unit CNT-C has a function of detecting this, replacing it with an appropriate internally generated code, and cutting off the circulation.

In FIG. 2, the code control unit CNT-C
In the case where the k-bit code number SD input to the internal shift registers SR1 to SR16 has a circulation cycle of 6 units as shown in FIG.
Fm, Cm, G7, F, C, Fdim, Fm, Cm, G
7, F, C, Fdim, Fm, Cm. Here, attention is focused on the comparator CMP6. One comparison value of the comparator CMP6 is a value of the shift registers SR1 to SR8, and the other comparison value is a value of the shift registers SR7 to SR14. Since the code (number) input to the shift register SR has a circulation cycle of 6 units (circulation code with a cycle of 6 units), they have the same result.

The comparator CMP6 outputs a coincidence signal indicating that a cyclic code having a period of 6 units has been found to the detector D.
Output to ET. Since the comparator CMP6 outputs a coincidence signal every time after the coincidence with the period, the duration of the cyclic code can be measured by counting the coincidence signal. When the count value exceeds a preset threshold value, the detector DET sends a switching signal to the gate EXC. As a result, the gate EXC replaces the external input code number SD with the internally generated code generated by the code generator GEN.

The internally generated code output from the gate EXC is input to the first-stage bit storage section C1 of the shift register composed of C1 to Ci in FIG.
The ten input conditions are disturbed, and the continuous generation of the cyclic code is broken. In this way, the k-bit newly generated code output from the code generator CG of the first NN 10 is supplied to the pitch generator PG, where it is designated to specify an actual frequency in order to reproduce it electronically as a real sound. After the code specified by the CG is converted into a plurality of real frequencies, it is transmitted to the sound source device SS.

Next, the operation of the second NN 20 that controls the rhythm will be described. It is assumed that the pattern of the rhythm trigger output from the sigmonide function generator T in the final layer is as shown in FIG. 4 based on the result of the shift register composed of R1 to R1. Here, the first bar and the third bar have the same pattern. If the second and fourth measures are circulating in the same pattern, a cyclic rhythm is obtained. It is very natural that a rhythm pattern has a cyclic form, but like a chord, it becomes monotonous if the same repeated rhythm pattern exists for a long time.

Therefore, the trigger control device CNT- of FIG.
R detects the cyclic rhythm and acts to break the regularity. That is, in FIG. 2 showing the configuration of the trigger control device CNT-R, the output trigger signal of the sigmonide function generator T of the last layer of the second NN 20 is input as the external input signal SD. Here, the shift register SR1
In SR16 to SR16, the signal situation when the trigger is set to 1 is 01101011 01101011 when divided into bars (the right end in this notation indicates the left end in FIG. 4).

Here, attention is paid to the comparator CMP8. One of the comparison values of the comparator CMP8 is the shift register SR1-S
The value of R8 is "0110 1011", and the other comparison value is the value of the shift registers SR9 to SR16. Therefore, "0110 1011" has the same result.

Then, the comparator CMP8 outputs a coincidence signal indicating that a cyclic rhythm of period 8 has been found to the detector DET.
Output to Since the comparator CMP8 outputs a coincidence signal every time after the coincidence with the cycle, the duration of the cyclic rhythm can be measured by counting the coincidence signal. When the count value exceeds a preset threshold value, the detector DET sends a switching signal to the gate EXC.

Thus, the gate EXC replaces the external input trigger signal SD with the internally generated trigger signal generated by the code generator GEN. Since this new internally generated trigger signal is input to the first-stage bit storage unit R1 of the shift register composed of R1 to Rl in FIG. 1, the output trigger signal of the sigmonide function generator T of the last layer of the second NN 20 is Regularity will be lost. In this manner, the trigger signal output from the sigmonide function generator T of the final layer of the second NN 20 is transmitted to the sound source device SS
Is input to

On the other hand, the control device CNT-P of FIG. 1 sends a program signal indicating the musical instrument type to the sound source device SS.
The program signal indicating the musical instrument type specifies a plurality of types of the musical instrument under the chord instruction and the musical instrument under the trigger instruction. This is based on random designation by random numbers, but the transition probability is specified. This will be described with reference to FIG. 5. In FIG. 5, P1 to Pm indicate program numbers (instrument type numbers), and R11 indicates P1 to P1 after a unit rhythm, that is, the continuation probability of the same program number. Similarly, Rpq indicates the transition probability from program number Pp to program number Pq (where p =
1,. . . , M, q = 1,. . . , M).

That is, at the start of the function, the control device CNT-P sequentially follows the probability table of FIG. 5 based on the program number specified by a certain probability of occurrence.
This is to change the program number. And
The transition targets independently driven by this algorithm are the required number of programs including the number of sounds specified by the chord signal and the number of sounds specified by the trigger signal.

The control device CNT-V of FIG. 1 sends a signal indicating the volume of the musical instrument to the sound source device SS. The control device CNT-V specifies a plurality of types of volume for each of the musical instrument under the chord command and the musical instrument under the trigger command. This is basically based on random designation by random numbers,
Transition probabilities are specified. This will be described with reference to FIG. 6. In FIG. 6, V1 to Vn are volume numbers, and S11
Represents the volume number V1 to the volume number V1 after the unit rhythm, that is, the continuation probability of the same volume. Similarly, Suv indicates a transition probability from the volume number Vu to the volume number Vv (where u = 1,..., N, v = 1,..., N).

That is, the control unit CNT-V sequentially shifts the volume number according to the volume number designated by a certain probability of occurrence according to the probability table of FIG. 6 at the start of the function. is there. The number of transition targets independently driven by this algorithm is the required number of programs including the number of sounds specified by the chord signal and the number of sounds specified by the trigger signal.

The sound source device SS digitally stores a sound corresponding to the musical instrument specified by the program number in the form of a phoneme, reads out the sound so as to have a frequency specified by the pitch generator PG, and Control device CNT
-P, a waveform corresponding to the musical instrument type indicated by the program number indicated by the program number, the amplitude of the musical instrument volume based on the signal from the control device CNT-V, and the output of the sigmonide function generator T in the final layer of the second NN 20 On the basis of the trigger signal, an analog audio signal that acts to clarify the sound of the percussion instrument in particular is generated and output to the output terminal OUT.

The control devices CNT-P and CNT-V
Without providing one or both of the above, it is possible to create music that basically does not attract the attention of the audience, such as environmental music.

[0049]

As described above, according to the present invention,
Despite the fact that there is no unreasonable leap, it is not unchanging, the current element is determined with reference to the immediately preceding content, and
The decision process is not uniform, and chord changes and rhythm changes can be generated by a feedback-type neural network circuit, making it possible to compose environment-constituting music with many elements in an extremely economical manner and to draw the attention of the audience Even if the music is not intended to be performed, the user of the device can be freed from the conventional adjustment of the copyright right due to the fact that the music was always provided by the author through composition, performance, and singing.

Further, according to the present invention, when the first and second control devices detect that the number of circulating chains of occurrence of elements in the feedback type neural network exceeds the first and second threshold values, Is monotonous because it breaks the circular chain by changing the input conditions of the first and second feedback neural networks by inputting a code number and a trigger signal generated from another probability function. You can get rid of boring music.

Further, according to the present invention, the first and second feedback-type neural networks each have a group of learned weighting coefficients which have been learned in advance using existing music. Since there is further provided a switch for switching the learned weighting coefficient group in the second feedback type neural network, the type of generated music can be selected.

[Brief description of the drawings]

FIG. 1 is a block diagram of one embodiment of the present invention.

FIG. 2 is a circuit diagram of an example of control devices CNT-C and CNT-R of FIG.

FIG. 3 is a diagram illustrating an example of a code string generated in a first NN in FIG. 1;

FIG. 4 is a diagram illustrating an example of a rhythm trigger pattern generated at a second NN in FIG. 1;

FIG. 5 is a diagram showing a probability table in the control device CNT-P of FIG. 1;

FIG. 6 is a diagram showing a probability table in the control device CNT-V of FIG. 1;

[Explanation of symbols]

C1 to Ci, R1 to Rl, SR shift register DA1 to DAi DA converter W1 to W4 Weighting coefficient group D1 to Dj, S1 to Sm Second layer sigmonide function generator E1 to Ek, T Last layer sigmonide function generator CG, GEN Code generator PG Pitch generator CNT-C First control device CNT-R Second control device CNT-P Third control device CNT-V Fourth control device CLK Synchronous control device SS Sound source device SW Switch DET detection circuit CMP1 to CMP8 Comparator EXC gate 10 First neural network circuit (NN) 20 Second neural network circuit (NN)

Claims (3)

[Claims] 1. A first feedback type neural network which newly generates code numbers constituting music from the immediately preceding performance code number and a code number several tempo past, and an output code of the first feedback type neural network. A pitch generator for converting a number into a plurality of real frequencies, a second feedback neural network for newly generating a trigger signal constituting a rhythm for generating music from the immediately preceding performance trigger signal and a trigger signal several tempo past, When it is detected that the number of cyclic chains of code numbers generated in the first feedback type neural network exceeds a first threshold value, the code generated from another probability function is converted to the first feedback type neural network. Input to the first-stage shift register of the neural network A first control device provided in a feedback loop of the first feedback type neural network, and the number of cyclic chains of a trigger signal generated in the second feedback type neural network sets a second threshold value. When it is detected that the signal has exceeded the threshold value, a trigger signal generated from another probability function is input to the first-stage shift register of the second feedback-type neural network, during a feedback loop of the second feedback-type neural network. A second control device provided, a sound source device for receiving the output signals of the pitch generator and the second feedback neural network, generating and outputting an analog audio signal of a specified rhythm at a specified frequency And the first and second feedback Music creation apparatus; and a synchronization control unit to operate in synchronization type neural network. 2. The first and second feedback neural networks each have a learned weighting coefficient group that has been learned in advance using existing music, and the first and second feedback neural networks. 2. The music creating apparatus according to claim 1, further comprising a switch for switching a group of learned weighting coefficients in the neural network. 3. The apparatus according to claim 2, further comprising at least one of a third control device for outputting a program signal for determining a musical instrument type to said sound source device and a fourth control device for outputting a signal for determining a musical instrument volume to said sound source device. 2. The method according to claim 1, wherein
A music creation device as described.