KR101854706B1 - Method and recording medium for automatic composition using artificial neural network - Google Patents

Abstract

The present invention relates to an automatic music composition method using an artificial neural network and a recording medium therefor that can automatically generate a music of a new melody different from a learning music and that in particular, can output a music having a natural and high musical perfection of an actual composer level by processing the music so as to suit to musicality or music theory. According to the present invention, the automatic music composition method using an artificial neural network comprises: a step of generating time series data by converting a plurality of notes and beats constituting a music to be learned by the artificial neural network (hereinafter referred to as a first learning music) into a numeric form; a step of learning the artificial neural network using the time series data; a step of outputting a new music by the artificial neural network; and a step of post-processing a beat to correct an exceeding bar with a complete bar if there is the bar (hereinafter referred to as an exceeding bar) which exceeds a beat compared with the complete bar in the new music.

Description

TECHNICAL FIELD [0001] The present invention relates to an automatic composing method using an artificial neural network,

The present invention relates to an automatic composing method, and more particularly, to an automatic composing method using an artificial neural network capable of automatically creating a new song by learning a song using an artificial neural network and newly adding an initial melody to the learned neural network, And relates to the recording medium.

Attempts to automatically compose music using a computer have been around for a long time. These studies have been classified as algorithmic compositions, and most of the methods developed in the field of artificial intelligence have been used, as can be seen from the purpose of composition.

The first method is to create an algorithm that performs automatic composing based on grammar, rule based, case based. It is based on grammar that it is applied to composition by simulating writing through grammar in language. The rule is based on creating rules that generate melodies and creating songs based on them. Based on the case, the existing song is databaseed by each measure, and the composition is composed by using the measure in the existing song.

As a second method, there is an optimization technique. The optimization technique that is mainly used in composition is evolutionary algorithm (Evolutionary Algorithms). In particular, the genetic programming that expresses an object as a tree is a very useful technique in composition because it is composed hierarchically from the sound to the syllable, and from the syllable to the syllable. However, the biggest problem with the optimization method is that it is difficult to evaluate the object. The evolution of a short piece expressed as an object must be properly assessed to make it more musically appropriate to express a better piece.

However, it is very difficult for a computer to evaluate a song by itself. So the average person uses a lot of methods to evaluate each object, but it is very tedious and hard work for people to evaluate very many objects from tens to hundreds of generations. Also, evaluation of music is very subjective and it is difficult to present a consistent standard. In order to mitigate the difficulty of the evaluation of fitness, some researches are performed to evaluate the music through learning function of artificial neural network. However, it is very difficult to ensure that the evaluation of the music of the learned neural network is performed properly.

A third way is by machine learning. Machine learning is an artificial neural network that is representative of a computer learning data given to itself. The artificial neural network learns the given data and can output the same or similar data as the learned data.

In the artificial neural network method, a song is output by learning an existing song, and thus a song having a feeling similar to that of a song written relatively by the composer is outputted. However, because there are many repetitive parts of the song, it is difficult to learn the song and it takes a very long time. In addition, the method of learning and outputting one song is problematic in that the outputted music is very similar to the existing music, and it is difficult to determine how to learn by some criteria when learning a plurality of songs.

Another example of machine learning is using a Markov chain. In Korea, there is an example of performing composition using Hyper Network, which is a type of machine learning tool.

However, such conventional methods are not enough to generate a music having a natural feel at the composer level as a method of simulating a part of performing a composition or exploring the possibilities.

In addition, there is almost no way to automatically generate and compose the entire music including the remaining melody, beat, and harmony just by generating a part of the melody or beat.

It is an object of the present invention to provide an automatic composition method using an artificial neural network capable of automatically generating a complete music piece including a new melody, a beat and a harmony using an artificial neural network, And to provide the recording medium.

It is still another object of the present invention to provide an automatic composition method using an artificial neural network capable of post-processing the beat and composition of a song output by the artificial neural network according to the music theory and further completing the beat- And to provide the recording medium.

It is another object of the present invention to provide an automatic composition method and an automatic music composition method using the artificial neural network that can prevent a case where a sound suddenly bounces on a generated song due to improper comma learning and distort the overall melody flow.

Another object of the present invention is to provide an automatic composing method using an artificial neural network capable of solving the phenomenon that the characteristic or melody of an output music is biased to one side when an artificial neural network learns a plurality of songs, and a recording medium therefor.

According to another aspect of the present invention, there is provided an automatic composition method using an artificial neural network, comprising: converting a plurality of notes and a beat constituting a song to be learned by an artificial neural network (hereinafter referred to as a first learning song) Generating time series data based on the time series data; Learning the artificial neural network using the time series data; Outputting a new music piece by the artificial neural network; And a beat-after-processing step of correcting, when there is a section exceeding a beat exceeding a beat provided in the new music (hereinafter, referred to as an 'excess bar') with a bar having the excess bar.

The step of post-processing may be performed by any one of a first process, a second process, and a third process.

Specifically, in the first process, a first step of subtracting a beat (hereinafter, referred to as 'd') exceeding the excess node from the provided one of the notes of each note constituting the excess node .

In the second process, 'd ₁ ' (here, d ₁ <d) is subtracted from any one of the notes of each note constituting the excess node, and in the other one of the notes, 'dd ₁ ' And subtracting &lt; / RTI &gt;

The third process includes a first c step of adding 'd ₂ ' beats from one beat of each note constituting the excess node, and subtracting 'd + d ₂ ' beats from another beat And a control unit.

Preferably, a composition post-processing step for correcting the composition of the new tune may be further included. The post-composition post-processing step includes: making the composition of the new tune into a C major having a first treble; And converting the second treasurer's tone to another treasure tone by adding the difference between the first treble tone and the second treble tone to all notes of the C major tone.

According to the automatic composition method and the recording medium using the artificial neural network according to the present invention, it is possible to automatically generate an entire music piece having a new melody different from the learning music.

Especially, the rhythm, harmony, composition, comma and equipments of the automatically generated music are processed in accordance with musical ability or music theory, so that it is possible to create a music which is natural to the actual composer level and has excellent musical perfection.

Furthermore, when the artificial neural network learns a large number of songs, it is possible to solve the problem that the characteristic or the melody of the output music is biased to any one music.

1 is a flowchart showing a processing flow of an automatic composing method using an artificial neural network according to the present invention.
2 is an example of a first learning song of the present invention.
FIG. 3 shows an example of a learning pattern of an artificial neural network according to the present invention, and thus, a learning data generation process and a learning process thereof.
4 shows an example of the comma processing problem of the present invention.
5 is a flowchart showing a process flow of a post-beating process according to the present invention.
6 is an example of a first process of the post-tone process according to the present invention.
7 is an example of a second process of the post-tone post-processing according to the present invention.
8 is an example of a third process of the post-tone process according to the present invention.
9 (a), 9 (b) and 9 (c) illustrate post-composition processing according to the present invention.
FIG. 10 shows an example of a problem of initial melody processing according to the present invention.
11 is a diagram illustrating an average artificial neural network generation principle according to the present invention.
12 is an example of a song that is automatically composed using an artificial neural network according to the present invention.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises" or "having" and the like refer to the presence of stated features, integers, steps, operations, elements, components, or combinations thereof, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Also, in this specification, the terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

In the following, preferred embodiments, advantages and features of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating a process flow of an automatic composition method using an artificial neural network according to the present invention. Referring to FIG. 1, an automatic composition method using an artificial neural network according to the present invention includes a step of digitizing a specific melody (step S10), a step of generating a time series data (step S20), an artificial neural network learning step (Step S30), a lead-in time series data input step (step S40), and a new song output step (step S50).

The automatic composing method using the artificial neural network according to the present invention is characterized in that learning is performed through a plurality of learning data S21 generated using the time series data of step 2 (S20) in the artificial neural network learning of step 3 (S30) (S22), and further performs a learning step (S30) by additionally inputting the division information of the melody to be learned (S22), and postprocessing the beat and harmony of the new song output through the step 5 (Step 6 (S60)).

Furthermore, the automatic composing method using the artificial neural network of the present invention may further include a step of post-processing the composition of the new tune after completion of the beat and harmony (Step S70).

As an alternative embodiment, the automatic composing method using the artificial neural network of the present invention may omit the learning step of the artificial neural network. In this case, the number of steps of the specific melody (step 1 (S10) 4 (S40)) may be omitted.

In other words, instead of learning the neural network, instead of learning the neural network, instead of randomly putting the initial melody, or randomly using the neural network, instead of learning the neural network, instead of outputting the rest of the composition by composing the initial melody into the neural network after learning the existing music on the neural network It is a method to generate a beat and process from step 'S60'. This way, you can create a new song without learning the existing song. However, in this case, since the existing music is not learned, more sophisticated melody / beat processing and more sophisticated melody / beat enhancement are performed to produce meaningful results.

Hereinafter, preferred embodiments according to the respective steps will be described in more detail.

&Lt; Step 1 (S10)

Step 1 (S10) of the present invention includes a step of converting a plurality of notes, beats, and commas constituting a corresponding song (hereinafter, referred to as 'first learning song') into numerals when a music to be learned by the artificial neural network is selected to be.

Here, the music to be learned by artificial neural network is an existing music, and its genre is not particularly limited. For example, if the melody of the automatically composed music desired by the user desires a feeling similar to the 'A' song corresponding to the song, the artificial neural network may learn the song 'A'. Furthermore, if the melody of the automatically composed music desired by the user desires to have a feeling of well-combined feeling of various songs (for example, a song 'A' and a pop song 'B'), the song 'A' and the pop song 'B' This can be the song to learn.

The number type conversion of the first learning song will be described in detail as follows. In order to learn the melody of the predetermined music piece in the artificial neural network, the notes (i.e., the pitch of the note) constituting the melody of the first learning music (i.e., the pitch of the note) and the comma are expressed by numbers.

First, note (note pitch) can be represented by numbers for all 1 to 7 octaves. In the present invention, however, it is assumed that only 2 to 4 octaves are used. In the above case, there are a total of twelve notes such as' degrees, degrees #, #, #, #, #, #, #, #, #, Quot ;, and &quot; 12 &quot;, respectively.

For example, two octaves of 'DO' are '01', 'D0', '02', 'L' is '03', 'L #' is '04' '06', 'Pa #' is '07', 'Sol' is '08', 'Sol #' is '09', 'La' is '10' Can be expressed as '12'.

In the present invention, since three octaves are used, the notes of each octave can be converted into a number from '01' to '36'.

On the other hand, there are not only notes but also commas in the music. Therefore, the comma must also be expressed in numbers. If a note has a comma, it is configured to use a number that is not used in the note to separate the note from the comma.

For example, in the present invention, a note is configured to express a note using a number from '01' to '36'. A comma is converted using a number greater than '36', for example, '37' So that notes and commas can be distinguished. Therefore, the artificial neural network can perform the learning by separating the number of '36' when the learning of step 3 is performed, by separating it with a comma.

Next, the beat (negative length) is defined as "1/16, 2/16, 4/16, 8/16, 16/16" based on the 16th note, multiplied by 16 Quot; 1, 2, 4, 8, 16 ". That is, when the sixteenth note is used as a reference, the beat (i.e., the length of the note) includes a first note, a second note, a quarter note, an eighth note, and a sixteenth note.

(2 + 1) = 3 beats. "When a point is attached to the fourth beat," (4 + 2) = 6 beats " "(8 + 4) = 12 beats". (4 + 2 + 1) = 7 beats "when there are two points on four beats and" (8 + 4 + 2) = 14 beats "when two beats on eight beats do. However, since most popular songs only use half a beat with a single dot, we usually use 8 beats of "1, 2, 3, 4, 6, 8, 12, If you use only 8 beats in this way, you can represent each beat with a number from 1 to 8.

Meanwhile, the time series data of the present invention can distinguish a note from a comma through a number expressed differently. When the beat information represented by a numeral is added to the corresponding note or comma, the beat note of the corresponding note, the comma beat of the corresponding comma Information becomes available. Thus, there is no need for another separate number to distinguish the note and comma beats.

However, it is also possible to configure it to separate the note and comma beats by the following method. In this case, the note rhythm further expresses a first digit (e.g., '1') in place of the '10', and a comma rhythm represents a second digit (e.g., 0 ').

For example, '♪' is the number '1' of the note rhythm, so it can be converted to the number '14', and ' In case of beat, it can be converted to '02'.

In the above case, the artificial neural network distinguishes the note and comma beats from each other by a numeral (e.g., a first numeral or a second numeral) in which tens of the numeral expressing the beat are expressed during the learning in step 3, .

&Lt; Step 2 (S20)

The step 2 (S20) of the present invention is a step of generating time series data in which numbered notes, a comma, and a beat are arranged according to the time series sequence of the first learning songs through the conversion process of the step 1 described above.

Thus, when the process of step 2 is completed, one piece of time series data (hereinafter referred to as 'note time series data') in which notes of the piece to be learned is converted into numbers, and one piece of time series data Time series data &quot;) is obtained.

Meanwhile, when the process of step 2 is completed, time series data (hereinafter, referred to as 'time series data') obtained by converting the commas of the songs to be learned into numbers is obtained. The time series data of this comma is configured to be included in the time series data , Or time-series data separated from the note / beat time series data.

As described above, the comma of the present invention is represented by a number different from the number representing the note or by a number larger than the maximum number of notes. However, when an artificial neural network learns a note and a comma together, a note space may be distorted, which is caused by a comma number expressed by a large number of notes compared to a note.

In order to solve this problem, the comma time series data of the present invention is separated from the note / beat time series data, and the artificial neural network is composed of artificial neural networks for notes learning notes and comma artificial neural networks for learning comma .

This will be described in more detail as follows. Suppose that a note is represented by a number between 1 and 36, and a comma is represented by a number greater than a note, for example, '50' so that the note and the comma are learned together in the artificial neural network.

In the above case, the artificial neural network that has learned the note space of '1 to 36' in the learning tune is learned by a large number of spaces of '50' due to the occurrence of the comma. Such learning may distort the musical notes output from the automatic musical composition by distorting to the note space. For example, if a note is progressed to "10, 14, 10, ..." and a comma is generated, it will be "10, 14, 10, 50, The space behind the time series data value is quickly mapped to a large value.

When the learned artificial neural network generates a note space similar to "10, 14, 10" in the composition of a new song, for example, "12, 15, 9", the next note space is mapped to a larger number, . In this case, notes such as "12, 15, 9, 34" are generated, resulting in sudden large notes that cause distortion in the overall melody flow. For example, referring to the example of FIG. 4, it can be seen that a song that normally proceeds as shown in a red circle suddenly produces a high sound.

To alleviate this problem, a comma may be displayed with a number one greater than the largest number of notes (e.g., 36) (e.g., 37). In this case, distortion of the melody note space is reduced, but the artificial neural network outputs almost all of the range from 1 to 36 in the composition of a new song, so that another problem may occur that the comma is hardly produced.

As a method to solve all of these problems, it is possible to separate artificial neural networks learning notes and artificial neural networks learning comma positions.

In this case, one artificial neural network (hereinafter, referred to as 'first artificial neural network') learns notes using note time series data, and another artificial neural network (hereinafter referred to as 'comma artificial neural network' The position of the comma is learned using the time series data having the comma information.

If the comma artificial neural network learned in the process of generating the new music by the first artificial neural network outputs a comma, it is processed as if there is a comma at the output position. If the note output by the first artificial neural network exists at the comma output position, the note is configured to replace the note with a comma.

For example, in the case of the time series data used for learning the artificial neural network for the comma, the comma information represented by a larger value (for example, '50') than the number representing the note may be included. In this case, If the neural network outputs a value greater than the maximum value of the note (for example, '36'), the position is treated as having a comma, and the output of the first artificial neural network at this time is ignored.

In the above example, '50' is input as a numeral indicating a comma. If this value is made larger, a comma may occur more frequently. That is, the frequency with which the artificial neural network outputs a comma depends on the size of the number indicating the comma, and thus the occurrence frequency of the comma is adjustable through the size of the number indicating the comma.

As described above, when the artificial neural network is divided into the first artificial neural network and the comma artificial neural network, the first artificial neural network is configured to perform learning using time series data from which the comma information is removed.

That is, the first artificial neural network can remove a comma in a plurality of notes and a comma constituting a song to be learned, and put the same value as a note immediately preceding the comma position in a position where the comma is omitted. If there is no preceding note, you can insert the same number as the note immediately after it.

In this way, time series data composed only of notes is constructed to learn the notes to the first artificial neural network, and the position of the comma is learned to the comma artificial neural network using the time series data composed of notes and comma.

As described above, when a music piece is composed separately from an artificial neural network that learns only the position of a comma, it is possible to prevent a sudden jumping sound from distorting the overall melody flow.

Meanwhile, the time series data of the notes and the beat may be separated from each other. In this case, the artificial neural network may be divided into artificial neural networks for notes and artificial neural networks for beats.

The artificial neural network for notes uses the time series data of notes to perform the learning and then generates the notes of the new music (that is, the pitch of the sound). The beat artificial neural network performs learning using time series data, (I.e., the negative length).

For example, in the above description, the note time series data and the beat time series data are configured to be respectively learned by two artificial neural networks. However, it is needless to say that the artificial neural network may be configured to learn notes and time series data when necessary.

2 is an example of the first learning song of the present invention, and is a part of the score of the song 'Roly-Poly' of 'Singer Tiara'. This song is within the note range from the lowest note '2 octaves' to the highest note '4 octaves less'. When there is a synthesized note based on the main score of 'Roly-Poly', it is numbered as one representative sound. A time series data conversion example of the two songs will be described as follows. The notes can first be converted into musical time series data of [15 18 18 18 18 17 15 18 18 18 18 17], and the tempo can be converted into beat time series data of [2 2 2 4 4 2 2 2 2 4 14] . When converted to including a traversal, a total of 360 time series data is generated including notes and commas.

&Lt; Step 2-1 (S21) >

The artificial neural network performs learning based on the note, beat and comma time series data generated in step 2 (S20). Before such learning, the time series data includes a plurality of learning data Lt; / RTI &gt;

The pattern for generating learning data is as follows. The time series data input / output learning pattern is as follows. First time series data from 'first' to 'N' is input to the artificial neural network and 'N + 1' th time series data is output to generate first learning data.

Then, the second learning data is generated by inputting the time series data from the second time to the (N + 1) th time into the artificial neural network and outputting the (N + 2) th time series data.

Then, the third learning data is generated by inputting the time series data from the 'third' to the 'N + 2' th to the artificial neural network and outputting 'N + 3' th time series data.

Then, by repeating the above-mentioned pattern for generating the n-th learning data by inputting the time series data from the n-th to the N + (n-1) th to the artificial neural network and outputting the N + A plurality of learning data is generated.

FIG. 3 is an example showing a learning pattern of the artificial neural network according to the present invention, and a process of generating and learning the learning data. Referring to FIG. 3, a method of generating a plurality of learning data through time series data will be described in detail

In the embodiment of FIG. 3, two artificial neural networks are used to generate learning data and to learn notes, commas, and beats based on the data. Specifically, the artificial neural network includes a feedforward neural network (FNN) Network).

The structure of the neural network is configured to have 10 input nodes, 20 intermediate nodes, and 1 output node in both neural networks. An input / output learning pattern is created using the time series data representing the tune to be learned, and a plurality of training data (TD: Training Data) can be generated through iteration.

More specifically, when the first to tenth time series data (# 01 to # 10) are input, the learning pattern is the first learning pattern in which the 11th time series data (# 11) is output, (TD # 1) is generated.

In the second learning pattern, the first time-series data of the first learning data generated by the first learning pattern is excluded, and the second to the 11th time-series data (# 02 to # 11) (# 12) is output, and second learning data (TD # 02) is generated.

When a learning pattern of this type is repeated, for example, a total of 90 learning data is generated by using a neural network having 10 input nodes in a song having 100 time series data.

In the above description, the feedforward artificial neural network is used as an artificial neural network for automatic composition (that is, a first artificial neural network, a comma artificial neural network, a beat artificial neural network, etc.). The omnidirectional feedback artificial neural network is constructed to regenerate learning data as described above and to learn learning data constructed as described above.

Needless to say, it is also possible to use Recurrent Neural Networks that directly learn time series data instead of such forward feedback artificial neural networks. Using the circular neural network, it is possible to directly learn the time series data without having to construct the learning data regressively. Especially, when LSTM (Long Short-Term Memory) loop neural network, which is widely used in deep learning recently, is used, it is advantageous to include the characteristics of time series data in the front part even in the case of long learning songs. In learning the existing music at the stage of automatic composition, it is possible to learn and compose by using such a circular neural network, and it is possible to apply other processes without using the circular neural network.

&Lt; Step 3 (S30) >

Step 3 (S30) of the present invention is a step of learning an artificial neural network using time series data. Specifically, when the above-described step 2-1 (S21) is performed, learning data TD # 1, TD # 2, .., TD #N are generated in each learning pattern step. Each learning data is applied to an artificial neural network to perform learning.

In the case of FIG. 3, the learning is performed using an error back propagation algorithm, and the maximum learning number is set not to exceed 1,000,000 times. However, the learning can be terminated when the termination condition such as no more learning before that or the total learning error value falls below a specific value is satisfied.

Artificial neural networks can learn learning data provided by input / output given adequate structure and learning rate and sufficient learning time. However, if there is data that can not be learned in the learning data, there may arise a problem that proper learning can not be performed even if a sufficient time learning process is performed (for example, the maximum number of learning times is reached).

Here, the data that can not be learned is a case in which learning data giving different outputs for the same input is given. In this case, the neural network can not determine which of the two outputs to learn to output, but can learn to shift to one side or back and forth between the two outputs, so that even if a lot of learning time is given, .

This is because the time series learning data obtained by transforming the music differently from the general learning data is caused by the occurrence of the time series data repeated in the characteristics of the music. In other words, there are many melodies that are generally repeated in a song, and in particular, a repeat melody is repeated several times over a repeat. The reason why such a repetitive melody is a problem is that a lot of learning data for generating different outputs for the same input side data in the neural network learning data occur.

In such a case, the neural network needs to learn various other outputs for the same input, so it can not converge to a specific value and the learning moves back and forth. As a result, even if the number of learning is performed, Occurs.

&Lt; Step 2-2 (S22) >

Step 2-2 (S22) of the present invention is for solving the problem of occurrence of a lot of repeated time series data due to the characteristics of the music as described above. In learning of the artificial neural network, time series data (Hereinafter, referred to as &quot; segment classification information &quot;) for identifying a segment of a song.

In other words, 'segment classification information' means information to be put in the same way as notes or beats in order to distinguish other notes or beats following the same note or beat at the learning of the artificial neural network.

The node classification information may be expressed by a different number for each node, and the artificial neural network may be further input to the time series data constituting the learning data at the time of learning each learning data.

By additionally adding the nodal information of the time series data currently being learned to the time series data consisting of only the notes and the beats, it is possible to learn other outputs using the nodal classification information even in the learning data having the same note or beat.

In this way, at least one or more clause classification information must be provided to distinguish the nodes. However, because of the increased input to the nodal segment, the learning time of one learning data as a whole increases. Therefore, do not increase the number of section information.

Although it is desirable to input more than 40 nodal information for ensuring the nodal classification, the number of nodal information may vary depending on the repeatability of the nodal and the degree of similarity between the nodal information.

Since the artificial neural network learns based on the nodal information to be input through step 2-2, the input can be classified even if the time series data of the music to be learned is the same. It is not necessary to have a specific value for the input of the nodal classification information, but it is only necessary to input different inputs (that is, different numbers for each node) for each node. For example, 40 numbers can be randomly generated and used as input for sorting information.

As a result, it is possible to prevent the occurrence of the same input / output learning data by inserting new node classification information around the learning data that the artificial neural network learns. This segmentation information also has the effect of specifying the position of the segment for the corresponding learning song. In other words, when two songs are learned, the two songs are equally divided, so that two songs can be learned by each word.

On the other hand, the artificial neural network outputs melody according to the learned music, so it is difficult to generate a repetitive melody composition in the process of generating the music. However, in composing a song, the melody repetition adds to the interest of the song and further functions to capture the composition of the song. Therefore, a method is needed to allow the artificial neural network to repeat the progress of the melody when generating a new song, which can be implemented through the above-described segment classification information.

That is, the above-mentioned segmentation effect can be used to generate repeated segments when outputting a new song, which can control the repeatability of the output song by appropriately arranging the segmentation information.

&Lt; Step 4 (S40), Step 5 (S50) >

Step 4 (S40) of the present invention is a step of composing a lead-in portion of a new piece of music and inputting the time-series data thereof into the artificial neural network.

When a person composes a new piece of music (that is, a piece of music to be composed by the artificial neural network) (hereinafter, referred to as an "introduced piece") to convert the time series data into time series data and inputs it into the artificial neural network, The artificial neural network outputs a new song based on the time series data of the input introduction song (S50). Here, the time series data of the introduced tune refers to the notes and the tempo constituting the introduced tune are converted into numerical form.

If the input tune is the beginning of a new tune, the artificial neural network outputs the new tune following the input tune series data as a start melody (S50).

If the artificial neural network is sufficiently learned, put 10 initial notes, a comma, and a beat in the same way as the learned song, and output the remaining notes, comma, and beat in the same way as the original learned song. Compose 10 new initials, comma, and beat differently from the learning song and insert it into the beginning, automatically generating the remaining notes, commas, and beats. Thus, a new melody can be obtained, and a new composition thus produced has a similar melody feeling to the learned song, but generates another melody.

However, since a music piece to be newly generated by the artificial neural network is based on the learning of the neural network, it is possible to generate a result that is incompatible with the music theory (for example, beat, Mars, etc.). Therefore, the melody produced by the neural network needs to be further processed in accordance with the music theory.

&Lt; Step 6 (S60)

Step 6 (S60) of the present invention is a step of post-processing the new music created through step 5 (S50) so as to conform to the music theory, and more specifically, divided into a post-beating process and a post-conversion process.

In the post-processing step according to the present invention, the first thing to be post-processed is the beat, and the beat must be modified to produce a measure for each measure. The post-beat processing is an operation that corrects the beat when the note of the note output by the artificial neural network fails to form a predetermined measure.

5 is a flow chart showing a process flow of the post-tone post-processing according to the present invention. A detailed algorithm of the post-tone post-processing method will be described below with reference to Fig.

The beat-post-processing step of the present invention is a step of correcting a beat exceeding a beat (hereinafter referred to as an &quot; excess beat &quot; ), A second process (S63), and a third process (S65).

The first processing (S61) of the beat-post-processing step is a step of subtracting a beat (hereinafter, referred to as 'd') exceeding the excess node from a beat provided in any beat of each note constituting the excess node Step 1a.

According to a preferred embodiment, the first step is a step (step &lt; RTI ID = 0.0 &gt; a &quot; step &quot;) & d ') is subtracted from the smallest beat (hereinafter, referred to as' first beat').

Then, it is determined whether or not a beat subtracted by 'd' from the first beat exists in the beat group, and if there is a result of the discrimination, Quot; d &quot; in the next small pitch (hereinafter referred to as 'second pitch').

In step S62, it is determined whether or not the beats subtracted by 'd' from the second beat are present in the beat group. If there is a result of the discrimination, The same process is performed for the next smaller beat compared to the second beat.

6 is an example of the first process of the post-tone processing according to the present invention. In Fig. 6, the uppermost score exceeds four beats compared to 16 nights, which is a node with one bar exceeding 20 bars. In this case, when the excess beat ('d' = 4) is subtracted from the smallest beat to the largest beat in accordance with the first processing method, it is determined whether correction is possible with the prepared bar.

In the example of Figure 6, even though the excess beat (d '= 4) is subtracted from the smallest beat (i.e., the two-beat note' ♪ ') and the next small beat (ie, the four- It can not be corrected. However, if '4' is subtracted from the largest beat (ie, the eight beat note), it is possible to perform the beat after-beat process satisfying the beat group at the same time while maintaining the order of the beats before correction (S66).

If it is not possible to make a measure even if the first processing method described above is performed on the beats of all notes constituting the subject to be subjected to the after-beat processing, the second processing to be described later is performed.

For reference, the 'used beat group' refers to the type of beat (that is, the length of a note) that can be used by the artificial neural network when automatic composing by the artificial neural network of the present invention, Can be set by the user.

For example, if the sixteenth note is set as a reference, the used tempo group can be designated as a group of "1,2,3,4,6,8,12,16" beats (hereinafter referred to as "first group") .

For example, if the song is 4/4, the used tempo group is the same as the first group, and the sum of the timings of the subject-to-be-processed is 18, subtract 2 from the smallest If the smallest beat obtained by subtracting the 'two beats' is present in the first group, the after-beat process for the corresponding beat is completed. If not present, , And if the beat satisfying the first group is absent even if the 'two beats' subtraction is performed on all the beats, the second process to be described later is performed.

Subtracts the second process (S63) is as long as one of the beat of each musical note constituting the excess bars in any of the time 'd _1' (wherein, d ₁ <d) time after the beat treatment step, and another one of the time Quot; dd ₁ &quot; Here, 'd' refers to a beat exceeding an exceeding node in terms of a provisional node.

And, if said 'd _1' and dd ₁ 'to a minus as beat time used to determine (S64) whether or not present in the time group, if the determination result exists to complete the post-processing time for the excess bars, and not present A third process to be described later is performed.

Fig. 7 shows an example of the second process of the post-tone post-processing according to the present invention. In FIG. 7, the uppermost score of the score exceeds three beats compared to 16 nights, which is a bar with one bar exceeding 19 bars. In the case of Fig. 7, it is impossible to perform the post-processing after the first processing. Therefore, the number of times exceeding 16 is subtracted from the beats of the two notes according to the second processing method.

In the example of FIG. 7, since the sum of the note pitches "3, 8, 8" in the excess section is 19, which is larger than 16 by 3 (d), it is necessary to subtract 3 from 3, but subtract from 3, Method (that is, the first process), there is no case where the measure group satisfies the use time group and does not become a measure. For reference, if the beat is set on the basis of the 16th note as in the example of FIG. 7, the beat group may be designated as [1,2,3,4,6,8,12,16] beats.

Therefore, in the above case, the second process of subtracting from the two beats and finding a case of satisfying the condition is performed (S63). In other words, if you subtract '1' (d ₁ ) from the triplet note to make it a two-beat note, subtract '2' (dd ₁ ) from the eight- It is possible to perform the post-beating processing that satisfies the used beat group while maintaining the order of the beats before correction (S66).

If it is not possible to make a measure even if the second processing method described above is performed on the beats of all notes constituting the subject to be subjected to the after-beat processing, the third processing to be described later is performed.

The third process (S65) of the post-tone processing step is to add a 'd ₂ ' beat number to any one of the beat of each note constituting the excess node, and to add 'd + d ₂ ' . Here, 'd' refers to a beat exceeding an exceeding node in terms of a provisional node.

'D ₂ ' is a value that satisfies the time signature group when a beat of 'd ₂ ' is added to an arbitrary beat of an excess measure, and 'd + d ₂ ' + d ₂ 'It is composed of the values that can satisfy the beat group when the beat is taken.

According to a preferred embodiment, the second processing and the third processing are configured to create a measure equipped with a 'beat size order' of the target to be processed (i.e., excess). Here, the 'beat sequence size' is configured to process the 'beat sequence size' even if the size of the modified beat is the same.

For example, if the notes of the notes constituting the target segment are "1, 2, 3, 6, 8" and the segments after the beat are "1,1,3,3,8" In this case, it is also deemed that the order of 'beat size' is satisfied.

8 is an example of a third process of the post-tone post-processing according to the present invention. In Fig. 8, the score at the top exceeds one beat in comparison with 16 nights, which is a node equipped with an excess of 17 nights. In the case of FIG. 8, even after any one of the first process and the second process method, the post-tone post-process is impossible and the post-tone post-process is performed according to the third process method.

In the example shown in FIG. 8, the sum of the musical notes "1, 8, 8" in the excess section is "17", and the sum of the excess section "d" However, with such a first processing or a second processing method, it is not possible to create a node satisfying the used time signature group.

In such a case, a third processing method of adding to one beat and subtracting it from the other beats can be solved. That is, the first addition of "1 (d _2), the beat notes made of two beat note, the time increment (d ₂ = 1) the value" 2 "obtained by adding the excess of the time (d = 1) by 8 beats notes As a result, when the processing is performed in the subtracting manner, it is corrected to a set of bars consisting of "2, 6, 8" beats, thereby enabling the beat processing to satisfy the beat group at the same time while maintaining the order of the beats before correction (S66).

According to any one of the first, second, and third processes described above, it is possible to perform a post-beating process on an excess bar in any case.

The post-processing step of the present invention includes a first step of generating a candidate for harmony by dividing the melody of a new song by each of the nodes, and a second step of creating a candidate for each harmony candidate, A second step of selecting the first step; And a third step of modifying the melody of the new song on the basis of the selected harmony.

Specifically, the first step comprises: arranging the notes in each of the bars from a low note; and determining a chord for each bar by calculating the interval between the remaining notes in succession from the lowest note to the root note And generating the harmony candidate group by storing chords determined for each of the nodes.

The second step may include the steps of: comparing the root candidate differences generated by the first step with the root candidate differences of the nodes; searching for a code corresponding to the development of the specific combination of the specific code; If there is a combination, storing it and determining the final harmony.

If there is a code combination that does not coincide with the search result in the second step, an error tolerance range is set based on the specific code combination, and the code combination within the error tolerance range is stored.

The third step further includes modifying the code combination within the error tolerance into the final harmonic.

The specific combination of codes in the second step may be a Money Chord.

&Lt; Step 7 (S70)

Step 7 (S70) of the present invention is a step for post-processing the composition of the new tune to further enhance the musical completeness of the tune generated in accordance with the processes of steps 'S10' to 'S60'.

The composition refers to the character of the scales that are arranged by regularly arranging the whole tone and semitone around the megamitus. One song usually has one composition, and the composition of the song determines the atmosphere of the song. For example, in the case of the composition C, "Do, Lem, Mi, Paw, Sol, La, Si and Do" are used. They are arranged in a whole tone order and have a bright atmosphere.

However, a composition composed using a learned artificial neural network does not constitute a specific composition. In order to solve this problem, the automatic composition method of the present invention may further include a composition post-processing step.

The composition post-processing step includes a first step of making the composition of the song (hereinafter referred to as a 'new song') produced in accordance with the process of steps 'S10' to 'S60' into C major, And a second step of converting the new song back to the composition desired by the user.

Specifically, in the first step of the composition post-processing step, when there is a sound (hereinafter, referred to as 'negative sound') not used in the C major in the new music, the negative sound is sounded just above or below the negative sound And converts the sound into a sound used in the C major.

That is, the height of each tone of a new song is collectively raised or lowered based on the C major so that the new song has C (degrees) as its dominant tone. In this process, Wave #, sol #, la #, etc. ", the corresponding note is changed to the note used in the C major.

For example, if 're #' is present in the process of making the composition of the new song C major, the corresponding note is changed to 'Do', which is the immediately lower note, or 'Not', which is the note immediately above.

If there are a plurality of negative sounds, all of the negative sounds may be changed collectively to the sound immediately above the excluded sound, or all of them may be changed collectively to a negative sound, or a mixture thereof.

When the first step is completed, a second step of converting the composition of the new C-major tune into a desired composition is performed using the difference of the root sounds.

Specifically, the second step of post-composition processing is to add the difference between the first summing tone and the second summing tone to all the tones of the new tune transformed into the C major, thereby converting the second trumping tone to another trunk having the trumpet do. Here, the 'first tune' refers to the dominant tone of C major, that is, C (degrees).

For example, if it is desired to make the composition of the first new song C-major converted by the first step into the G major, only the difference between the root note C (degrees) and G (sol) So that only the notes used in the G major can be left and the regularity of the major can be maintained.

9 (a), 9 (b) and 9 (c) illustrate an example of post-composition processing according to the present invention, in which a new song generated in the process of steps 'S10' to 'S60' The post-treatment process.

9A shows a melody before a post-composition process for a new song generated according to the process of steps 'S10' to 'S60', and has various sounds used in various compositions. In other words, it has the sounds of "par #, lah #, lah, lah #, par, sol, par". There is no wave # in the descending order because it is "re #, par, sol, sol #, la #, do, les, re #". "Wave, wave #" is the sound used in the down-counting. Therefore, composing tune can not be seen as a single composition, so it is necessary to adjust the tones so as to form a composition through post-composition processing.

In this case, first, the melody of FIG. 9 (a) is made C major by the first step. That is, when "par #, l #, l #" corresponding to the excluded notes in the "par #, l #, l, l, l #, par, It constitutes the C major consisting of the scales of "Sol, Si, Le, Les, Mi, Pa, Sol, Wave".

Subsequently, for each tone of the C major transformed by the first step, the E major transform is completed by adding the difference between the root major of the C major and the E major axis. That is, the C tonic of the C major is C (degrees), and the root of the E major is A (minor), so that the root difference corresponds to "4". Thus, if the root difference '4' is added to all the tones of FIG. 9 (b), it is possible to add "Tone, Ree #, Wave #, Wave, Sol #, La, And converted into the E-major composition composed of notes used in the E-major.

In the above description, if the user desires to make the composition of a new music to be a major desired one (hereinafter, referred to as a "desired major"), the music composition is first converted to the C major. However, It can be configured to change immediately.

That is, a certain tone is added or subtracted to each tone of a new song at a time to form a composition of a desired tone. In this case, a tone not used in the desired tone may exist.

If there is a sound that is not used in the hope major (i.e., a negative sound), the negative sound is changed to a sound immediately above or below the sound to be used in the desired sound. In other words, when a new song is converted to have G (sol) as masquerade, the height of each tone of the new song is raised or lowered collectively based on the G major tone. If a tone not used in G major exists, Change the note to a note used in the G major.

For example, if 're #' is present in the process of making the composition of the new song into the G major, the corresponding note is changed to 'Do', which is the immediately lower note, or 'Not', which is the note immediately above.

According to the steps 1 (S10) to 6 (S60) as described above, a song automatically composed by the artificial neural network is finally generated. Artificial neural network that has learned one song completely makes the rest melody the same as learned music by putting the same initial melody as the learning song. However, if you add a learned melody and a song that you learned, it will output a different melody similar to the feeling and feel of the song you learned. Ultimately, in order to compose a new song, a new initial melody (ie, introduction song) must be created by the user. However, since the initial melody created by the user can not predict the melody feeling of the artificial neural network following the initial melody, the melody of the artificial neural network may not match with the rest of the melody produced by the artificial neural network, .

Fig. 10 shows an example thereof. In FIG. 10, the portion in the red box is the introduction song portion that the user has composed and the remaining portion is the artificial neural network output portion. The output part of the artificial neural network outputs a song with a certain feeling and atmosphere, whereas the part of the introduction song that the user inserts is different from the part output by the artificial neural network. Since the space mapped to the artificial neural network can not be anticipated in advance, it is very difficult to create an intro song so that the artificial neural network has a similar feel to the song.

However, the initial melody (introduction song) is an indicator that indicates the starting point in the melody space mapped to the artificial neural network, and plays a role in determining the position in the time series space to output the music. Therefore, the initial melody is only used to map the position of the space to output a new song, and it is not necessary to use it for the composition result.

Therefore, it is desirable to construct an initial melody randomly so that the initial melody is not used in the result of composition. In other words, it is preferable to use the artificial neural network to exclude the introduced tune from the completed composition when the composition is finally completed. According to this, all of the compositions composed of only the output of the artificial neural network can solve the problem that the initial melody part and the rest do not match.

In order for an artificial neural network to compose musical pieces, many songs must be learned. At this time, there may be a method of learning a plurality of songs one by one. However, according to this learning method, the artificial neural network is influenced by the music to be learned at a later time, and the characteristic or melody of the generated music may be biased to one side (that is, a later learning music).

The present invention proposes a method of using weights of a plurality of artificial neural networks that have learned different songs as follows in order to solve the problem of eliminating generated songs that may occur when learning a plurality of songs in order.

FIG. 11 is a view illustrating an average artificial neural network generation principle according to the present invention. Referring to FIG. 11, first, a different music is learned in N artificial neural networks (here, N? 2). That is, the first artificial neural network is learned using the first time series data obtained by converting a plurality of notes and beats constituting a music to be learned by the first artificial neural network (hereinafter, referred to as 'first learning music' And the second artificial neural network is learned using the second time series data obtained by converting a plurality of notes and beats forming a song to be learned by the second artificial neural network (hereinafter, referred to as a 'second learning song' Nth artificial neural network using the Nth time series data obtained by converting a plurality of notes and beats composing a piece of music to be learned by the Nth artificial neural network (hereinafter referred to as an &quot; Nth learning song &quot; .

In the automatic composition method using an artificial neural network, the generation of the first through Nth time series data and the learning method using the first through Nth time series data are performed in the steps 2 (S20), 2-1 (S21), 2-2 S22) and step 3 (S30).

The weight and bias of the hidden layer (hereinafter referred to as 'first hidden layer') and the output layer (hereinafter, 'first output layer') of the learned first artificial neural network, the hidden layer of the learned second artificial neural network (Hereinafter referred to as an Nth hidden layer) and an output layer (hereinafter referred to as an Nth hidden layer) of the learned Nth artificial neural network (hereinafter referred to as an Nth hidden layer) and a weight of the output layer Output layer ') and the bias of the output layer.

Here, the weight is a number related to the connection between a neuron or two neurons, and is used to assign an output to determine the activation value of the neuron. And bias is a value that is added to the activation of neurons and plays an important role in learning. The bias can be used to move the activation function from side to side. And a hidden layer refers to an array of neurons between an input layer and an output layer.

(Hereinafter referred to as a 'hidden layer average weight') and a bias average value (hereinafter referred to as 'hidden layer average') of the first hidden layer, the second hidden layer, and the Nth hidden layer are obtained by obtaining both the weight and the bias of each artificial neural network. Bias').

At the same time, a weighted average value (hereinafter referred to as 'output layer average weight') and a bias average value (hereinafter, 'output layer average bias') of the first output layer, the second output layer and the Nth output layer are also calculated.

Thereafter, in the case of the hidden layer, another neural network composed of the neurons having the hidden layer average weight and the hidden layer average bias and composed of neurons having the output layer average weight and the output layer average bias in the case of the output layer, (Hereinafter referred to as 'average artificial neural network').

Then, the new artificial neural network is used to generate a new song according to the above-described steps 4 (S40) to 7 (S70), and after the beat, harmony, and composition of the new song are processed, new song composition is completed .

As described above, according to the artificial neural network, it is possible to solve the problem that each song can not be learned without being concentrated in one song when learning a plurality of songs, and the problem that the characteristics of the songs to be produced vary according to the order of the songs. As a result, the automatic composition method using an average artificial neural network can be especially useful for learning a lot of songs.

In this paper, we propose an average artificial neural network in order to solve the problems of generating songs that can occur when learning a number of songs in order. In this case, instead of using the average artificial neural network, we use MSE (Mean Square Errors) You can also use the method.

That is, a method of limiting the degree of learning of a music piece to be learned later in learning a plurality of tunes in turn, is configured to use Mean Square Errors (MSE) as a learning parameter for determining to what extent the artificial neural network learns a specific music .

The MSE represents the error difference between the data of the song to be learned and the output of the artificial neural network. Therefore, by setting the specific value of the learning permalizer as the learning termination condition and proceeding to the learning, it is possible to adjust to what extent the specific music is to be learned, thereby limiting the degree of learning of the music to be learned later.

More specifically, suppose that the artificial neural network learns the A song and the B song but learns the A song first and then the B song. In this case, the music A proceeds until it is fully learned. That is, it is configured to learn until the MSE of the A song becomes '0'.

If the A song is fully learned (ie, MSE = 0), then the B song is learned, where the B song is configured to stop learning before it is fully learned (ie, MSE> 0).

In other words, the MSE of the B song is configured to have a value larger than '0'. Specifically, the MSE of the B song is obtained by applying the B song to the artificial neural network that has learned the A song and obtains the initial MSE. Is configured to set a specific ratio R1 (where 0 &lt; R1 &lt; 1) to the MSE of B songs.

For example, assume that the specific rate is set to 0.1 (i.e., 10%) and that the initial MSE is calculated to be '247' when the B song is input to the artificial neural network learning the A music. Applying a specific rate of '0.1 (ie, 10%)' to the initial MSE calculated in this case, the MSE of the B song is determined to be '24.7 ', and this value serves as a learning termination condition for the B song.

That is, the learning of B song stops after MSE = 24.7. As a result, the B song can not learn until the MSE becomes '0', and the learning can be terminated after the learning progresses to some extent.

Therefore, by adjusting the value of the specific ratio, the MSE of the B song can be adjusted, thereby limiting the degree of learning of the B song, thereby solving the problem that the generated song is biased toward a song to be learned later.

According to the method using the MSE, the degree of learning of each music can be limited to some extent, but the result is different depending on the order of the songs to be learned. In the case of learning a lot of songs, It still happens. Therefore, it is desirable to use the average neural network when learning a lot of songs.

12 is an example of a song that is automatically composed using the artificial neural network according to the present invention, and is a result of learning two songs of Iirma's piano piece "All myself you" and "River flows in you".

As can be seen from FIG. 12, no artifacts were generated by applying the post-tone processing of the present invention at all, and no artifacts were generated in the middle by using a comma artificial neural network.

The initial melody, that is, the introduction song was selected randomly, and the method of excluding this introduction song from the composition result outputted by the artificial neural network was applied. That is, the introduction song was not used as the composition result.

The automatic composing method using the artificial neural network according to the present invention as described and illustrated above can be read by an electric / electronic apparatus in which a program for executing each of the above-described functions or steps S10 to S70 is recorded in an electric / May be provided in the form of a recording medium.

While the preferred embodiments of the present invention have been described and illustrated above using specific terms, such terms are used only for the purpose of clarifying the invention, and it is to be understood that the embodiment It will be obvious that various changes and modifications can be made without departing from the spirit and scope of the invention. Such modified embodiments should not be understood individually from the spirit and scope of the present invention, but should be regarded as being within the scope of the claims of the present invention.

Claims (4)

delete A method for automatically generating a song using an artificial neural network,
Generating time series data by converting a plurality of notes and beats constituting a music to be learned by the artificial neural network (hereinafter, referred to as a 'first learning song') into numerical form; Learning the artificial neural network using the time series data; Outputting a new music piece by the artificial neural network; And a beat-after-processing step of correcting, when there is a section exceeding a beat exceeding a beat provided in the new music (hereinafter, referred to as an &quot; over-section &quot;),
Wherein the time-post processing step is performed by any one of a first process, a second process, and a third process,
The first process includes a first step of subtracting a beat (hereinafter, referred to as 'd') exceeding the excess node from the provided node with any one of the beats of each note constituting the excess node ,
The second process is a process of subtracting 'd ₁ ' (here, d ₁ <d) beats from any one of the beats of each note constituting the excess node, and subtracting 'dd ₁ ' And subtracting &lt; RTI ID = 0.0 &gt;
The third process includes a first c step of adding 'd ₂ ' beats from any one of the beats of each note constituting the excess node, and subtracting 'd + d ₂ ' beats from another beat, / RTI &gt;
The first step is configured to subtract 'd' from the smallest one of the notes of each note constituting the excess node (hereinafter referred to as 'first beat'),
The first process may include:
('First beat-d' beat) obtained by subtracting 'd' from the artificial neural network in the first beat group (hereinafter, referred to as 'use beat group') designated as beats available for composition A step 2a of discriminating whether or not it exists;
If there is a result of the discrimination in step 2a, the process of post-beat processing for the excess clause is completed. If it is not present, the second beat is compared with the first beat in the second beat Step 3a;
(4a) of determining whether a beat ('second beat-d' beat) subtracted by 'd' is present in the used beat group in step 3a) And
And if the result of the determination in step (4a) is yes, repeating the beat back process for the excess node.
A method for automatically generating a song using an artificial neural network,
Generating time series data by converting a plurality of notes and beats constituting a music to be learned by the artificial neural network (hereinafter, referred to as a 'first learning song') into numerical form; Learning the artificial neural network using the time series data; Outputting a new music piece by the artificial neural network; And a beat-after-processing step of correcting, when there is a section exceeding a beat exceeding a beat provided in the new music (hereinafter, referred to as an &quot; over-section &quot;),
Wherein the time-post processing step is performed by any one of a first process, a second process, and a third process,
The first process includes a first step of subtracting a beat (hereinafter, referred to as 'd') exceeding the excess node from the provided node with any one of the beats of each note constituting the excess node ,
The second process is a process of subtracting 'd ₁ ' (here, d ₁ <d) beats from any one of the beats of each note constituting the excess node, and subtracting 'dd ₁ ' And subtracting &lt; RTI ID = 0.0 &gt;
The third process includes a first c step of adding 'd ₂ ' beats from any one of the beats of each note constituting the excess node, and subtracting 'd + d ₂ ' beats from another beat, / RTI &gt;
The 'd ₂ ' and the 'd + d ₂ ' of the third process are generated in such a way that the artificial neural network may include a beats group, Wherein the method comprises the steps of:
A method for automatically generating a song using an artificial neural network,
Generating time series data by converting a plurality of notes and beats constituting a music to be learned by the artificial neural network (hereinafter, referred to as a 'first learning song') into numerical form; Learning the artificial neural network using the time series data; Outputting a new music piece by the artificial neural network; A step of post-processing to correct, when there is a section exceeding a time interval (hereinafter referred to as &quot; exceeding section &quot; And a composition post-processing step of correcting the composition of the new song,
Wherein the step of post-processing includes performing at least a first process,
The first processing may include: an i-th step of subtracting a beat (hereinafter, referred to as 'd') exceeding the excess node from the provided node at any one of the beat of each note constituting the excess node; And a step 2a of discriminating whether or not a beat obtained by subtracting the 'd' of the step 1a from the artificial neural network in the beat group (hereinafter, referred to as 'use beat group'&Lt; / RTI &gt;
The post-composition treatment step
A first step of making the composition of the new tune a C major having a first tune; And a second step of adding the difference between the first summing tone and the second summing tone to all of the notes of the C major to thereby convert the second summing tone to another major having the summing tone. Automatic composing method.

Images