Musical piece simplification device, music piece simplification method, musical score editing device, musical score editing system, program, and information recording medium

Abstract

This musical piece abridging device (101) abridges a code string that represents a musical piece. Here, an abridging unit (111) executes abridgement wherein one of two adjacent note codes is absorbed by the other in a code string including note codes that represent tone pitches and tone lengths. A control unit (121) controls the abridging unit (111) to repeatedly execute abridgement until the note codes included in the code string, which represents a musical piece, become one. The abridgment is performed to convert a code string by means of a trained neural network (112), which has learnt conversion for replacing an absorbed code among two adjacent note codes with a placeholder code. The adjustment unit (113) removes the placeholder code from the converted code string.

Description

[Technical Field]

[0001] The present invention relates to a music piece simplification device, a music piece simplification method, a music score editing device, a music score editing system, a program, and an information recording medium for simplifying a code string representing a piece of music. [Background technology]

[0002] Since there are multiple possible interpretations of a single piece of music, this ambiguity poses an obstacle when implementing music theory on a computer.The inventor is researching and developing technology to automatically generate a time-span tree that represents a piece of music by combining the music theory GTTM (Generative Theory of Total Music) with statistical learning.

[0003] Figure 1 is an explanatory diagram showing the structure of a time span tree for a musical score. As shown in this figure, the time span tree is a binary tree whose leaves are notes in a piece of music arranged in chronological order, and shows the hierarchical structure of the musical score.

[0004] In this figure, the lower part shows a musical score consisting of two rests and 11 notes, and the upper part shows a time span tree. Between the root at the top of the tree and the leaves at the bottom of the tree, there are intermediate nodes.

[0005] An intermediate node has two children, which are either leaves or other intermediate nodes.

[0006] At intermediate nodes in a timespan tree, the edges (branches) from two children intersect like a T-junction: some edges continue to extend toward the root after intersecting at the intermediate node, while others terminate there.

[0007] If both children of an intermediate node are leaves, the note corresponding to the child of the former edge is interpreted as more important, or salient, than the note corresponding to the child of the latter edge. A time-span tree can be thought of as a regular binary tree with labels attached to each intermediate node indicating which of its children are salient.

[0008] The prominent note of an intermediate node is sometimes called the "head," representing the leaves and other intermediate nodes that are its children. The subtree consisting of an intermediate node and its children can be interpreted as being reduced to notes with the same pitch as the head and durations corresponding to the durations of the notes and rests contained in the subtree.

[0009] Once a time span tree for a song is obtained, various applications become possible, such as fusing two songs in a desired ratio to generate a new song that exhibits the characteristics of the two songs at that fusion ratio (Patent Document 1).

[0010] Here, in Patent Document 2, When analyzing the time span tree of a song, Based on the characteristics of each consecutive sound shown in the music data, the strength of the head, which is an important sound in each time span of the music, is calculated. The calculated strength value is stored in a storage means in association with the head; If the current processing target time span contains two or more heads, these heads are used as candidates for the head of the next hierarchy time span. These head strength values ​​are read from the storage means and compared to select the next hierarchical head from among the candidates. The technology is disclosed.

[0011] Furthermore, Non-Patent Document 1 proposes a technique for generating a time-span tree using a probabilistic context-free grammar (PCFG).

[0012] Meanwhile, translation technology using deep learning is beginning to produce results. In this field, a technology has been proposed that translates a code string representing a sentence in one language into a sentence in another language by repeatedly replacing part of the code string. This type of conversion is realized by neural networks called Seq2Seq (Non-Patent Document 2) and Transformer (Non-Patent Document 3), and various improvements have also been proposed.

[0013] Here, we consider a process of traversing intermediate nodes in a time span tree from leaf to root in depth-first, breadth-first, or order based on importance within the piece. This process can be thought of as a reduction process in which two adjacent notes in a piece are repeatedly absorbed into the more prominent other, and this reduction process can be interpreted as equivalent to a time span tree. [Prior art documents] [Patent documents]

[0014] [Patent Document 1] Patent No. 5051539 [Patent Document 2] Japanese Patent Application Laid-Open No. 2007-101780 [Non-patent literature]

[0015] [Non-Patent Document 1] Masatoshi Hamanaka, Keiji Hirata, and Satoshi Tojo, "σGTTM III: Learning based Time-span Tree Generator based on PCFG", Proceedings of the 11th International Symposium on CMMR 2015, pp.387-404, https: / / doi.org / 10.1007 / 978-3-319-46282-0_25, 2015 [Non-patent document 2] Ilya Sutskever, Oriol Vinyals, and Quoc V. Le, "Sequence to Sequence Learning with Neural Networks", Advances in Neural Information Processing Systems 27, pp.3104-3112, https: / / arxiv.org / abs / 1409.3215, 2014. [Non-patent document 3] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin, "Attention Is All You Need", 31st Conference on Neural Information Processing Systems (NIPS 2017), https: / / arxiv.org / abs / 1706.03762, 2017 Summary of the Invention [Problem to be solved by the invention]

[0016] Based on these background technologies, the inventors came up with a method for automatically and appropriately simplifying music using deep learning transformations in order to improve the accuracy of obtaining a time span tree from music.

[0017] In this method, a time span tree is generated by repeatedly performing musical reduction using a neural network.

[0018] According to the inventor's experiments, a certain degree of accuracy can be achieved by using a code string consisting of musical notes as a source, learning a transformation that targets a code string in which two adjacent notes are absorbed into the other, and repeatedly executing reduction using this transformation to generate a time span tree. However, there is a demand for improving the accuracy of automatically generated time span trees.

[0019] Therefore, there is a need for a technology that can obtain an appropriate time span tree by appropriately using neural networks to simplify music.

[0020] The present invention is intended to solve the above-mentioned problems and relates to a music simplification device, a music simplification method, a music score editing device, a music score editing system, a program, and an information recording medium for simplifying a code string representing a piece of music. [Means for solving the problem]

[0021] The music simplification device according to the present invention comprises a simplification unit and a control unit, the simplification unit performs a reduction in a code string including note symbols representing pitch and duration, in which one of two adjacent note symbols is absorbed into the other, the control unit controls the reduction unit to repeatedly perform the reduction until the number of note codes included in the code string representing the music piece becomes one; The reduction is transforming the code string using a neural network that has learned a transformation to replace one of the two adjacent note symbols that is to be absorbed with a placeholder symbol; Remove the placeholder code from the transformed code string. This is carried out by: [Effects of the Invention]

[0022] According to the present invention, it is possible to provide a music simplification device, a music simplification method, a music score editing device, a music score editing system, a program, and an information recording medium for simplifying a code string representing a piece of music. [Brief explanation of the drawings]

[0023] [Figure 1] FIG. 1 is an explanatory diagram showing the structure of a time span tree for a musical score. [Figure 2] 1 is an explanatory diagram showing a schematic configuration of a music simplifying device according to an embodiment of the present invention; [Figure 3] 1 is a flowchart showing a control flow of a music reduction process executed by a music reduction device according to an embodiment of the present invention. [Figure 4] 1 is an explanatory diagram showing a schematic configuration of a musical score editing device according to an embodiment of the present invention; [Figure 5] 1 is an explanatory diagram showing an example of a screen display of a musical score editing device according to an embodiment of the present invention; [Figure 6] 1 is an explanatory diagram showing an example of a screen display of a musical score editing device according to an embodiment of the present invention; [Figure 7] 1 is an explanatory diagram showing an example of a screen display of a musical score editing device according to an embodiment of the present invention; [Figure 8] 1 is an explanatory diagram showing an example of a screen display of a musical score editing device according to an embodiment of the present invention; [Figure 9] 1 is an explanatory diagram showing an example of a screen display of a musical score editing device according to an embodiment of the present invention; [Figure 10] 1 is an explanatory diagram showing a schematic configuration of a musical score editing system according to an embodiment of the present invention; [Figure 11] 1 is an explanatory diagram showing an example of a screen display when specifying a musical score to be edited in the musical score editing system according to an embodiment of the present invention; [Figure 12] 1 is an explanatory diagram showing an example of a screen display when specifying a musical score to be edited in the musical score editing system according to an embodiment of the present invention; [Figure 13] 1 is an explanatory diagram showing an example of a screen display when specifying a musical score to be edited in the musical score editing system according to an embodiment of the present invention; [Figure 14] FIG. 10 is an explanatory diagram showing an example of a screen display when simplifying measures in the musical score editing system according to the embodiment of the present invention. [Figure 15]FIG. 10 is an explanatory diagram showing an example of a screen display when simplifying measures in the musical score editing system according to the embodiment of the present invention. [Figure 16] FIG. 10 is an explanatory diagram showing an example of a screen display when simplifying measures in the musical score editing system according to the embodiment of the present invention. [Figure 17] FIG. 10 is an explanatory diagram showing an example of a screen display when simplifying measures in the musical score editing system according to the embodiment of the present invention. [Figure 18] FIG. 10 is an explanatory diagram showing an example of a screen display when simplifying measures in the musical score editing system according to the embodiment of the present invention. [Figure 19] FIG. 10 is an explanatory diagram showing an example of a screen display when expanding notes in the musical score editing system according to an embodiment of the present invention. [Figure 20] FIG. 10 is an explanatory diagram showing an example of a screen display when expanding notes in the musical score editing system according to an embodiment of the present invention. [Figure 21] FIG. 10 is an explanatory diagram showing an example of a screen display when expanding notes in the musical score editing system according to an embodiment of the present invention. [Figure 22] FIG. 10 is an explanatory diagram showing an example of a screen display when expanding notes in the musical score editing system according to an embodiment of the present invention. [Figure 23] FIG. 10 is an explanatory diagram showing an example of a screen display when expanding notes in the musical score editing system according to an embodiment of the present invention. [Figure 24] FIG. 10 is an explanatory diagram showing an example of a screen display when expanding notes in the musical score editing system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0024] The following describes embodiments of the present invention. Note that these embodiments are for illustrative purposes only and do not limit the scope of the present invention. Therefore, those skilled in the art can adopt embodiments in which each or all of the elements of the present embodiments are replaced with equivalents. Furthermore, elements described in each example can be omitted as appropriate depending on the application. In this way, all embodiments constructed in accordance with the principles of the present invention are included in the scope of the present invention.

[0025] (Configuration of the music simplification device) 2 is an explanatory diagram showing the general configuration of a music simplifying device according to an embodiment of the present invention. The following description will be made with reference to this diagram.

[0026] As shown in the figure, a music simplification device 101 according to this embodiment simplifies a code string representing a music piece, and includes a simplification unit 111 and a control unit 121. Here, the musical note code represents a pitch and a duration.

[0027] The simplification unit 111 performs a reduction in which one of two adjacent note symbols in a code string is absorbed into the other, thereby reducing the number of note symbols by one. The simplification unit 111 uses a neural network 112. This neural network 112 is trained using training data in which a source code string, which is a code string before a single reduction, is input, and a target code string, which is a code string after the reduction, is output. Therefore, in each training data, the source code string contains one more note symbol than the target code string.

[0028] The music simplification device 101 outputs the process of simplifying the code string representing the music as a processing result.

[0029] The generation unit 131, which is an optional component, generates a time span tree of the piece of music based on this reduction process.

[0030] The control unit 121 controls the simplification unit 111 to repeatedly perform the simplification until the number of musical note codes included in the code string is reduced to one.

[0031] In this figure, an optional element is also shown, namely, an adjustment unit 113. The adjustment unit 113 adjusts the output from the neural network 112 for input to the next neural network 112.

[0032] Each unit of the music reduction device 101 is typically realized by a computer that executes a program. The computer is connected to various output devices and input devices, and sends and receives information to and from these devices.

[0033] A program executed by a computer can be distributed or sold by a server to which the computer is connected for communication, or it can be recorded on a non-transitory information recording medium such as a CD-ROM (Compact Disk Read Only Memory), flash memory, or EEPROM (Electrically Erasable Programmable ROM), and then the information recording medium can be distributed, sold, etc.

[0034] The program is installed on a non-transitory information recording medium such as a hard disk, solid state drive, flash memory, EEPROM, etc., possessed by the computer. The music simplification device 101 of this embodiment is then realized by the computer. Generally, the computer's CPU (Central Processing Unit) reads the program from the information recording medium into RAM (Random Access Memory) under the control of the computer's OS (Operating System), and then interprets and executes the code contained in the program. However, in an architecture in which the information recording medium can be mapped within a memory space accessible by the CPU, explicit loading of the program into RAM may not be necessary. Various pieces of information required during program execution can be temporarily stored in RAM.

[0035] Furthermore, as mentioned above, it is desirable for computers to be equipped with a GPU (Graphics Processing Unit) to perform various image processing calculations at high speed. By using a GPU and libraries such as PyTorch and TensorFlow, it becomes possible to use learning functions such as deep learning in various artificial intelligence processes, as well as various conversion and translation functions, under the control of the CPU.

[0036] It should be noted that, instead of realizing the music simplification device 101 of this embodiment using a general-purpose computer, it is also possible to configure the music simplification device 101 of this embodiment using a dedicated electronic circuit. In this embodiment, the program can also be used as material for generating wiring diagrams, timing charts, etc. of the electronic circuit. In such an embodiment, an electronic circuit that meets the specifications defined in the program is configured using an FPGA (Field Programmable Gate Array) or ASIC (Application Specific Integrated Circuit), and the electronic circuit functions as a dedicated device that performs the functions defined in the program, thereby realizing the music simplification device 101 of this embodiment.

[0037] For ease of understanding, the following description will be given assuming that the music reduction device 101 is realized by a computer executing a program.

[0038] In this embodiment, for ease of understanding, consider the process in which the music reduction device 101 reduces a musical score expressed as follows: R2 c2 c2 c2 c2 c2 c2 c2 d6 c1 b1 c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16

[0039] This sheet music is The letters a, b, c, d, e, f, and g represent the pitch, followed by a number to represent the length. A rest symbol, which is the letter R followed by a number to indicate the length of the rest, This code string consists of 42 note symbols and 3 rest symbols.

[0040] Although not shown in this example, pitch can be expressed by adding "#" or "♭" after the letter to indicate a semitone above or below, or by expressing pitch using a numerical value such as frequency.

[0041] Furthermore, in this example, overtones (such as a note of a certain pitch and notes of pitches an integer octave above or below it) are collectively represented as a single pitch using a single alphabetic character, but it is also possible to adopt a coding system that represents notes of different octaves as different pitches.

[0042] When a human expert manually creates a time span tree from this score, they repeat the process of absorbing two adjacent notes into the other until there is only one note left. In the above code sequence, there are 42 note codes, so the reduction is performed 41 times.

[0043] Below, the state of the code string before and after reduction during the reduction process will be explained.

[0044] (Simplification method A) Below, we will explain the simplest method, reduction method A.

[0045] simplified(1) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 ((c1 b1)) c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 ( c2 ) c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 | c1 b1 → c2

[0046] Simplified(21) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 ((g4 c4)) c8 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 ( c8 ) c8 d4 e8 d8 c16 | g4 c4 → c8

[0047] In this example, the first reduction (1) and the 21st reduction (21) of reduction method A are extracted and shown.

[0048] A line beginning with "src:" indicates the source code string before reduction, and a line beginning with "tgt:" indicates the target code string after reduction.

[0049] Additionally, lines beginning with "|" show the two note symbols that were the subject of reduction and the single note symbol that is the result of the reduction, separated by "→". In reduction (1), two adjacent notes "c1 b1" are reduced to a single note "c2", and in reduction (21), two adjacent notes "g4 c4" are reduced to a single note "c8", and the sum of the durations of the two notes before reduction is carried over to the duration of the single note in the reduced word.

[0050] In this example, in the source code string, the parts to be reduced are enclosed in "((" and "))", and in the target code string, the results of the reduction are enclosed in "(" and ")", but these are notations for ease of understanding. Therefore, the neural network 112 does not process "((", "))", "(", ")".

[0051] In the simplification method A, the target code string output by the simplification (i) is a code string consisting of note codes and rest codes, and therefore can be used as it is as the source code string input to the simplification (i+1). In other words, the adjustment unit 113 can be omitted. Furthermore, reduction method A does not distinguish between the absorbing side (salient) and the absorbed side of two note symbols. For this reason, in an embodiment using reduction method A, it is difficult to automatically generate a time span tree in which intermediate nodes are assigned labels indicating which of the two children was salient. However, even reduction method A is useful for suggesting which two note symbols should be reduced when manually generating a time span tree.

[0052] (Simplification method B) In simplification method A, by taking the difference between the source code string and the target code string, it is possible to determine the position where the two note codes are simplified, but information regarding the simplified position is not explicitly stated in either the source code string or the target code string.

[0053] In the simplification method B, the performance of the music reduction device 101 is improved by specifying information about the position related to the simplification in the target code string.

[0054] Conversion (1) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 ((c1 b1)) c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 ( c2 p ) c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 | c1 b1 → c2 p

[0055] Conversion (21) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 ((g4 c4)) c8 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 ( p c8 ) c8 d4 e8 d8 c16 | g4 c4 → p c8

[0056] This example also shows the transformation (1) performed by the neural network 112 in the first reduction and the transformation (21) performed by the neural network 112 in the 21st reduction.

[0057] The difference between this example and simplification method A is that in the target code string, a placeholder code "p" is placed at the position where the absorbed note code was located, thereby making that position clear.

[0058] In transformation (1), "c1 b1" in the source code string is replaced with "c2 p", which clearly indicates that "b1" has been absorbed.

[0059] In the transformation (21), "g4 c4" in the source code string is replaced with "p c8", which clearly indicates that "g4" has been absorbed.

[0060] In reduction method B, the target code string output from the neural network 112 contains the placeholder code "p," so it cannot be used as is as the source code string to be input to the neural network 112 in the next reduction. Therefore, in order to use it as the source code string for the next reduction, it is necessary to remove the placeholder code "p" from the target code string. This removal process is performed by the adjustment unit 113.

[0061] (Simplification method C) In the simplification method B, the position of the absorbed note symbol in the source sequence is specified by a placeholder symbol in the target sequence. Therefore, the note symbol adjacent to or before the placeholder symbol is the absorbed note symbol.

[0062] In simplification method C, the performance of the music reduction device 101 is improved by providing different types of placeholder codes and clearly indicating whether the absorbing note code preceded the absorbed note code.

[0063] Conversion (1) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 ((c1 b1)) c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 ( c2 l ) c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 | c1 b1 → c2 l

[0064] Conversion (21) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 ((g4 c4)) c8 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 ( r c8 ) c8 d4 e8 d8 c16 | g4 c4 → r c8

[0065] This example also shows the transformation (1) performed by the neural network 112 in the first reduction and the transformation (21) performed by the neural network 112 in the 21st reduction.

[0066] The difference between this example and simplification method B is that in the target code sequence, the placeholder code "p" is replaced with the placeholder code "l" or "r", which indicates whether the absorbed note code precedes (before, to the left) or follows (after, to the right) the placeholder code.

[0067] In transformation (1), "c1 b1" in the source code string is replaced with "c2 l", which clearly shows that the preceding "c1" absorbed "b1" to become "c2".

[0068] In transformation (21), "g4 c4" in the source code string is replaced with "p c8", and it is clearly shown that the following "c4" absorbs "g4" and becomes "c8".

[0069] In reduction method C, as in reduction method B, the placeholder codes “l” and “r” are removed from the target code sequence output from the neural network 112 at a certain stage, thereby obtaining a source code sequence to be input to the neural network 112 at the next stage. This removal process is performed by the adjustment unit 113.

[0070] In addition, in all of the simplification methods A, B, and C, in conversion (1), the length of the note after absorption is "2", which is the sum of the length "1" of the absorbing note symbol "c1" and the length "1" of the absorbed note symbol "b1".

[0071] In conversion (21), the length of the absorbed note is also "8", which is the sum of the length of the absorbing "c4" ("4") and the length of the absorbed "g4" ("4").

[0072] this is, In a sequence of notes, two notes that are adjacent to each other are considered to be adjacent. In a reduction where two adjacent note symbols are absorbed into one another, the duration of the absorbing note becomes longer by the duration of the absorbed note. This means that...

[0073] The following shows the process of using simplification method C to reduce a code string representing a piece of music 41 times until only one note code remains.

[0074] Conversion (1) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 ((c1 b1)) c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 ( c2 l ) c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 | c1 b1 → c2 l

[0075] Conversion (2) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 ((c2 c2)) c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c2 c2 ( c4 l ) c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 | c2 c2 → c4 l

[0076] Transformation (3) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 ((c2 c2)) c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 ( c4 l ) c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e6 d2 d6 c2 c16 | c2 c2 → c4 l

[0077] Transformation (4) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 ((e6 d2)) d6 c2 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 ( e8 l ) d6 c2 c16 | e6 d2 → e8 l

[0078] Transformation(5) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 ((d6 c2)) c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 e2 e2 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 ( d8 l ) c16 | d6 c2 → d8 l

[0079] Transformation(6) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 ((e2 e2)) e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 e2 e2 ( e4 l ) e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | e2 e2 → e4 l

[0080] Transformation(7) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 ((e2 e2)) e4 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e2 e2 e2 ( e4 l ) e4 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | e2 e2 → e4 l

[0081] Transformation (8) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 ((R2 e2 e2)) e2 e4 e4 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 ( r R2 e4 ) e2 e4 e4 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | R2 e2 e2 → r R2 e4

[0082] Transformation (9) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 ((e4 e2)) e4 e4 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 ( e6 l ) e4 e4 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | e4 e2 → e6 l

[0083] Transformation (10) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e6 e4 e4 e4 e6 d2 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e6 e4 e4 e8 l d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | e6 d2 → e8 l

[0084] Transformation (11) src: R2 c2 c2 c2 c2 c2 c2 c2 d6 c2 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c2 c2 c2 c2 c2 c2 c2 d8 l c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | d6 c2 → d8 l

[0085] Transformation (12) src: R2 c2 c2 c2 c2 c2 c2 c2 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c2 c2 c2 c2 c2 c4 l d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | c2 c2 → c4 l

[0086] Transformation (13) src: R2 c2 c2 c2 ((c2 c2)) c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c2 c2 c2 ( c4 l ) c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | c2 c2 → c4 l

[0087] Transformation (14) src: ((R2 c2 c2)) c2 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: ( r R2 c4 ) c2 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | R2 c2 c2 → r R2 c4

[0088] Transformation (15) src: R2 ((c4 c2)) c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 ( c6 l ) c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 d6 c2 c4 g4 c4 g4 c4 d4 e8 d8 c16 | c4 c2 → c6 l

[0089] Transformation (16) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 ((d6 c2)) c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 c2 c2 ( d8 l ) c4 g4 c4 g4 c4 d4 e8 d8 c16 | d6 c2 → d8 l

[0090] Conversion (17) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 ((c2 c2)) d8 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 c4 ( c4 l ) d8 c4 g4 c4 g4 c4 d4 e8 d8 c16 | c2 c2 → c4 l

[0091] Conversion (18) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 ((c4 c4)) d8 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c2 c4 ( c8 l ) d8 c4 g4 c4 g4 c4 d4 e8 d8 c16 | c4 c4 → c8 l

[0092] Conversion (19) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 ((c2 c4)) c8 d8 c4 g4 c4 g4 c4 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 ( c6 l ) c8 d8 c4 g4 c4 g4 c4 d4 e8 d8 c16 | c2 c4 → c6 l

[0093] Transformation (20) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 g4 c4 ((g4 c4)) d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 g4 c4 ( r c8 ) d4 e8 d8 c16 | g4 c4 → r c8

[0094] Transformation (21) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 ((g4 c4)) c8 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 e4 e4 e8 d8 R2 c6 c8 d8 c4 ( r c8 ) c8 d4 e8 d8 c16 | g4 c4 → r c8

[0095] Transformation (22) src: R2 c6 c4 c4 d8 c4 c4 R2 e6 ((e4 e4)) e8 d8 R2 c6 c8 d8 c4 c8 c8 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 c4 c4 R2 e6 ( e8 l ) e8 d8 R2 c6 c8 d8 c4 c8 c8 d4 e8 d8 c16 | e4 e4 → e8 l

[0096] Transformation (23) src: R2 c6 c4 c4 d8 ((c4 c4)) R2 e6 e8 e8 d8 R2 c6 c8 d8 c4 c8 c8 d4 e8 d8 c16 tgt: R2 c6 c4 c4 d8 ( c8 l ) R2 e6 e8 e8 d8 R2 c6 c8 d8 c4 c8 c8 d4 e8 d8 c16 | c4 c4 → c8 l

[0097] Conversion (24) src: R2 c6 ((c4 c4)) d8 c8 R2 e6 e8 e8 d8 R2 c6 c8 d8 c4 c8 c8 d4 e8 d8 c16 tgt: R2 c6 ( c8 l ) d8 c8 R2 e6 e8 e8 d8 R2 c6 c8 d8 c4 c8 c8 d4 e8 d8 c16 | c4 c4 → c8 l

[0098] Conversion (25) src: R2 ((c6 c8)) d8 c8 R2 e6 e8 e8 d8 R2 c6 c8 d8 c4 c8 c8 d4 e8 d8 c16 tgt: R2 ( c14 l ) d8 c8 R2 e6 e8 e8 d8 R2 c6 c8 d8 c4 c8 c8 d4 e8 d8 c16 | c6 c8 → c14 l

[0099] Conversion (26) src: R2 c14 d8 c8 R2 e6 e8 e8 d8 R2 ((c6 c8)) d8 c4 c8 c8 d4 e8 d8 c16 tgt: R2 c14 d8 c8 R2 e6 e8 e8 d8 R2 ( c14 l ) d8 c4 c8 c8 d4 e8 d8 c16 | c6 c8 → c14 l

[0100] Conversion (27) src: R2 c14 d8 c8 R2 ((e6 e8)) e8 d8 R2 c14 d8 c4 c8 c8 d4 e8 d8 c16 tgt: R2 c14 d8 c8 R2 ( e14 l ) e8 d8 R2 c14 d8 c4 c8 c8 d4 e8 d8 c16 | e6 e8 → e14 l

[0101] conversion(28) src: R2 c14 ((d8 c8)) R2 e14 e8 d8 R2 c14 d8 c4 c8 c8 d4 e8 d8 c16 tgt: R2 c14 ( r c16 ) R2 e14 e8 d8 R2 c14 d8 c4 c8 c8 d4 e8 d8 c16 | d8 c8 → r c16

[0102] conversion(29) src: R2 c14 c16 R2 e14 ((e8 d8)) R2 c14 d8 c4 c8 c8 d4 e8 d8 c16 tgt: R2 c14 c16 R2 e14 ( r d16 ) R2 c14 d8 c4 c8 c8 d4 e8 d8 c16 | e8 d8 → r d16

[0103] conversion(30) src: R2 c14 c16 ((R2 e14 d16)) R2 c14 d8 c4 c8 c8 d4 e8 d8 c16 tgt: R2 c14 c16 ( r R2 d30 ) R2 c14 d8 c4 c8 c8 d4 e8 d8 c16 | R2 e14 d16 → r R2 d30

[0104] conversion(31) src: R2 c14 c16 R2 d30 R2 c14 ((d8 c4)) c8 c8 d4 e8 d8 c16 tgt: R2 c14 c16 R2 d30 R2 c14 ( r c12 ) c8 c8 d4 e8 d8 c16 | d8 c4 → r c12

[0105] conversion(32) src: R2 c14 c16 R2 d30 R2 c14 c12 ((c8 c8)) d4 e8 d8 c16 tgt: R2 c14 c16 R2 d30 R2 c14 c12 ( c16 l ) d4 e8 d8 c16 | c8 c8 → c16 l

[0106] conversion(33) src: R2 c14 c16 R2 d30 R2 c14 c12 c16 ((d4 e8)) d8 c16 tgt: R2 c14 c16 R2 d30 R2 c14 c12 c16 ( d12 l ) d8 c16 | d4 e8 → d12 l

[0107] conversion(34) src: R2 c14 c16 R2 d30 R2 c14 c12 c16 ((d12 d8)) c16 tgt: R2 c14 c16 R2 d30 R2 c14 c12 c16 ( r d20 ) c16 | d12 d8 → r d20

[0108] conversion(35) src: R2 c14 c16 R2 d30 R2 ((c14 c12)) c16 d20 c16 tgt: R2 c14 c16 R2 d30 R2 ( c26 l ) c16 d20 c16 | c14 c12 → c26 l

[0109] conversion(36) src: R2 c14 c16 R2 d30 R2 c26 ((c16 d20)) c16 tgt: R2 c14 c16 R2 d30 R2 c26 ( r d36 ) c16 | c16 d20 → r d36

[0110] Transformation (37) src: R2 c14 c16 R2 d30 ((R2 c26 d36)) c16 tgt: R2 c14 c16 R2 d30 ( r R2 d62 ) c16 | R2 c26 d36 → r R2 d62

[0111] Transformation (38) src: R2 ((c14 c16)) R2 d30 R2 d62 c16 tgt: R2 ( c30 l ) R2 d30 R2 d62 c16 | c14 c16 → c30 l

[0112] Transformation (39) src: R2 ((c30 R2 d30)) R2 d62 c16 tgt: R2 ( c62 l ) R2 d62 c16 | c30 R2 d30 → c62 l

[0113] Transformation (40) src: R2 c62 ((R2 d62 c16)) tgt: R2 c62 ( r R2 c78 ) | R2 d62 c16 → r R2 c78

[0114] Transformation (41) src: ((R2 c62 R2 c78)) tgt: ( r R2 c142 ) | R2 c62 R2 c78 → r R2 c142

[0115] (Handling of rest symbols) GTTM does not explain how to handle rests. However, the inventor's experiments have revealed that when learning is performed using data from which rests have been removed, the characteristic that "strong beats and weak beats are repeated," a fundamental feature of musical rhythmic structure, is not learned. For this reason, in this embodiment, rests are also included in the learning process. Furthermore, in order to avoid the need for exceptional processing for music that begins with a rest, this embodiment considers the rest and the note that follows it as a set.

[0116] Therefore, when a note symbol to the right of a rest symbol is absorbed into a note symbol further to the right, as in transformations (8), (14), (30), (37), (40), and (41), the placeholder symbol "r" is placed before the rest symbol.

[0117] Also, when one of two note symbols sandwiching a rest symbol is absorbed into the other, as in transformations (39) and (41), the rest symbol is also absorbed, and the duration of the absorbed note symbol after reduction becomes the sum of the durations of the two note symbols and the rest duration of the rest symbol.

[0118] Furthermore, in a sequence of rest sign, note sign, rest sign, note sign, as in transformation (41), when two note signs and the intervening rest sign are absorbed into the last note sign, the placeholder sign "r" is placed before the first rest sign.

[0119] These can be generalized as follows:

[0120] That is, a rest symbol, a first note symbol, and a second note symbol are arranged in this order. In a transformation in which the first note symbol is absorbed into the second note symbol, A placeholder symbol to replace the first note symbol is placed in a position preceding the rest symbol.

[0121] Also, two note symbols with a rest symbol between them are considered adjacent, In the reduction of two note symbols with a rest symbol between them, one of them is absorbed into the other, The note length represented by the absorbing other note is increased by the sum of the note length represented by the absorbed note and the rest length represented by the rest symbol. The pause code is absorbed into the other code, The absorbed one and the absorbed pause code are replaced together with one placeholder code.

[0122] Furthermore, the first rest symbol, the first note symbol, the second rest symbol, and the second note symbol are arranged in this order, In the conversion where the first note symbol and the second rest symbol are absorbed into the second note symbol, A placeholder symbol that replaces the first note symbol and the second rest symbol is placed in a position preceding the first rest symbol.

[0123] (Neural Networks) In this embodiment, the neural network 112 is Seq2Seq disclosed in Non-Patent Document 2, Transformer disclosed in Non-Patent Document 3 In other words, any model that converts one code string into another code string can be employed.

[0124] Here, Seq2Seq is a model that has an LSTM encoder that compresses input information and an LSTM decoder that expands this to produce output information, and uses SoftMax as the activation function for output. By using LSTM instead of RNN, it is possible to learn even for code sequences that are relatively long.

[0125] On the other hand, Transformer does not use RNN or LSTM, but instead uses an attention mechanism. Transformer provides functions such as positional encoding, which encodes using information about the position of the code within the code sequence, self-attention, which utilizes the dependency relationship between the input code sequence, and multi-head attention, which utilizes the dependency relationship between the input code sequence and the output code sequence.

[0126] For the neural network 112, in each reduction, a pair of code sequences listed as a source code sequence and a target code sequence is taken as training data.

[0127] Here, in reduction method A, both the source code string and the target code string may contain note symbols and rest symbols, but do not contain placeholder symbols.

[0128] In reduction method B, both the source code string and the target code string may contain note and rest symbols. The source code string does not contain the placeholder symbol "l", but the target code string does contain the placeholder symbol "l".

[0129] In reduction method C, both the source code string and the target code string may contain note and rest symbols. The source code string does not contain either the placeholder symbol "r" or "l", but the target code string contains either the placeholder symbol "r" or "l".

[0130] The neural network 112, which has been trained using the training data, is able to output a corresponding target code string when a source code string is input.

[0131] (Music simplification processing) The following describes the music reduction process executed by the music reduction device 101. Fig. 3 is a flowchart showing the control flow of the music reduction process executed by the music reduction device according to the embodiment of the present invention.

[0132] When the music simplification device 101 receives a code string representing a musical score to be processed (step S301), the generation unit 131 records the received code string as the initial state of the reduction process (step S302).

[0133] Next, the control unit 121 performs control to repeat the following process, with the accepted code string as the code string to be processed (step S303).

[0134] That is, the control unit 121 controls the reduction unit 111 to cause the neural network 112 to process the source code string and execute the conversion (step S304).

[0135] That is, the current code string is given as a source input to the neural network 112, and the neural network 112 outputs a target code string in which one of two adjacent note codes is absorbed into the other. The output target code string is then set as the code string to be processed.

[0136] When the reduction methods B and C are adopted, the code string output from the neural network 112 will contain placeholder codes.

[0137] Next, the generating unit 131 additionally records the code string to be processed as an intermediate state of the reduction process (step S305).

[0138] Furthermore, the control unit 121 determines whether or not the code string to be processed contains one note code (step S306), and if it does (step S306; Yes), ends the repetition and proceeds to step S309.

[0139] On the other hand, if two or more note symbols remain (step S306; No), the control unit 121 controls the simplification unit 111 to cause the adjustment unit 113 to delete the placeholder symbol from the symbol string (step S307), and repeats the processing from step S303 onwards (step S308).

[0140] When the repetition ends, the generation unit 131 generates a time span tree based on the recorded reduction process (step S309), and ends this process.

[0141] As described above, when simplification method A is adopted, the deletion of placeholder codes by adjustment unit 113 can be omitted.

[0142] Furthermore, the processing by the generation unit 131 can be omitted entirely.

[0143] (Experimental results) We conducted an experiment to automatically generate a time-span tree using the following specifications: Reduction methods B and C. As mentioned above, reduction method A is not suitable for automatically generating a time-span tree, so it was not included in the experiment.

[0144] First, the Seq2Seq model used the TensorFlow backend of Keras, with a latent dimension of 128 for the LSTM and a batch size of 2048, and one-hot encoding of the input and output. Categorical cross-entropy was used as the loss function, and Adam was used as the optimizer.

[0145] For Transformer, we used the OpenNMT-py toolkit (ver. 2.0.0rc2) with default parameters.

[0146] For training, 270 songs were selected from the GTTM database consisting of 300 songs, and 7,362 pairs of source and target sequences representing one reduction were used. In this case, placeholder sequences were appropriately placed in the target sequences depending on reduction methods B and C.

[0147] From the remaining 30 songs, 849 pairs of source and target chords were prepared for evaluation.

[0148] After training 20,000 epochs for simplification method B, we checked the accuracy of each simplification step to see if the evaluation data matched the results of the trained neural network. We found that Seq2Seq achieved an accuracy of 90%, while Transformer achieved an accuracy of 99%, indicating that the latter had performed better. Therefore, in the following experiments, we decided to use only Transformer.

[0149] The training accuracy when Transformer was combined with simplification method C was 99.92%, which was an improvement over simplification method B.

[0150] Furthermore, F measure We investigated the performance of F meature It is calculated by Precision P (the rate at which the note symbols selected as the absorber were correct) and Recall R (the rate at which the note symbols that should be identified as the absorber were correctly identified). F measure = 2×P×R / (P+R) is the weighted harmonic mean calculated as

[0151] F of σGTTM III disclosed in Non-Patent Document 1 measure was 0.76, F when combining simplification method B and Transformar measure is 0.91, F when combining simplification method C and Transformar measure was 0.95.

[0152] Therefore, it can be seen that the performance when simplifying music pieces is significantly improved by the simplification methods B and C according to this embodiment compared to the conventional methods.

[0153] In addition, it is generally expected that Transformer has higher learning accuracy than Seq2Seq, and when combined with Transformer, simplification method C has performance F compared to simplification method B. measure It can be seen that the

[0154] (Music score editing device) By using the music reduction device 101, it is possible to obtain a time span tree from a code string that represents a piece of music, and then to obtain the original code string from the time span tree. Therefore, editing a code string and transforming a time span tree can be considered to be equivalent.

[0155] Furthermore, the code string represents the pitch and duration of the information expressed by the musical score of a piece of music, and can be considered to be the same as the musical score.

[0156] Therefore, when editing a musical score, it is possible to edit the code string or the time span tree and reconstruct the musical score based on that information.

[0157] In this embodiment, the overall note duration remains unchanged even while the code string is being simplified. Therefore, for various ornaments (such as dynamics, tempo, and emphasis) expressed in the score, as long as the start and end times of the ornaments are retained, the ornaments can be applied to the edited score even if the score is converted into a code string or a time span tree and then edited.

[0158] The following describes a musical score editing device that uses the above-described musical piece simplification device 101. Fig. 4 is an explanatory diagram showing the general configuration of a musical score editing device according to an embodiment of the present invention. The following description will be made with reference to this diagram.

[0159] The musical score editing device 501 shown in the figure comprises the musical piece simplification device 101, a display unit 502, a reception unit 503, a specification unit 504, and an editing unit 505, and edits musical scores 506 based on instructions from a user. It is also possible to use an inverse neural network 511 that has been trained by swapping the input and output of the neural network 112 used in the musical piece simplification device 101.

[0160] Here, the display unit 502 displays the musical score of the music piece on the screen.

[0161] On the other hand, the accepting unit 503 accepts an edit instruction input for selecting a part of the displayed musical score.

[0162] Furthermore, the specifying unit 504 specifies the edited content based on the selected part.

[0163] Then, the editing unit 505 edits the musical score according to the specified editing content.

[0164] After the editing is completed, the screen display on the display unit 502 is updated to show the edited score, and the above process is repeated.

[0165] Below, as examples of editing content, a measure reduction for reducing a measure selected by the user and a note expansion for expanding a note selected by the user will be described.

[0166] (measure simplified) The bar reduction process starts when a bar reduction input is received as an editing instruction input, which selects the first bar from the displayed musical score.

[0167] Then, the identification unit 504 identifies a reduction in the reductions repeatedly performed by the music simplification device 101 on the code string representing the music piece in which a first note symbol representing a first note included in the selected first measure is absorbed into an adjacent second note symbol.

[0168] Then, the identification unit 504: a second note represented by the second note symbol before absorbing the first note symbol; a third note represented by the second note symbol after absorbing the first note symbol; Identify.

[0169] That is, when a measure (measure 1) is specified by the user, the music reduction device 101 searches for the first reduction performed in which any note (first note) in that measure (measure 1) is absorbed into another nearby note (second note) and converted into a new note (third note) and placeholder from among the reductions performed by the music reduction device 101.

[0170] And, the editorial department 505, in the musical score, Delete the first selected note; Substitute the specified third note for the second note in the specified reduction.

[0171] By performing this process, the user can select a measure and edit it to reduce the number of notes that appear in that measure while maintaining the time length of that measure.

[0172] (note expansion) The note expansion starts when a note expansion input is received as an editing instruction input, which selects any note (first note) from the displayed musical score.

[0173] Then, the specifying unit 504 first inserts a placeholder code into the code string representing the music piece so that the placeholder code is adjacent to the first note code representing the selected first note.

[0174] There are two possible locations where the placeholder symbol can be inserted: before and after the first note symbol. Therefore, if you are using simplification method B, A string of notes with a placeholder symbol p inserted just before the first note symbol, A string of notes with a placeholder symbol p inserted immediately after the first note symbol, You can make two of these.

[0175] Also, when simplification method C is adopted, a placeholder symbol is inserted to indicate that the first note symbol absorbs other adjacent note symbols. A string of chords with a placeholder symbol r inserted just before the first note symbol, A string of symbols with a placeholder symbol l inserted immediately after the first note symbol, You can make two of these.

[0176] The identifying unit 504 then provides the newly created code string as an input to the inverse neural network 511 for conversion.

[0177] As described above, the neural network 112 is trained using training data in which the source code string (src) is used as input data and the target code string (tgt) is used as output data (correct answer data).

[0178] On the other hand, the inverse neural network 511 switches these and learns using training data in which the target code string (tgt) is used as input data and the source code string (src) is used as output data (correct answer data).

[0179] When the code string into which the placeholder code has been inserted is given as input to the inverse neural network 511, the note code adjacent to the placeholder code is expanded into two, one of which is placed in the position of the placeholder code and the other is placed in its original position.

[0180] Therefore, the specifying unit 504 specifies that the first note symbol and the placeholder symbol have been converted into the second note symbol and the third note symbol.

[0181] Then, the identification unit 504: a second note represented by the converted second note symbol; a third note represented by the converted third note symbol; Identify.

[0182] The editing unit 505 replaces the selected first note with the identified second and third notes in the musical score.

[0183] By executing this process, the user can select a note and edit it to expand it into two notes while maintaining the total note length.

[0184] In bar reduction and note expansion, multiple candidates for the reduction method and the expansion method may be obtained. In this case, the candidate with the highest likelihood may be adopted for reduction or expansion using note transition probabilities, etc. Alternatively, each candidate may be presented to the user to allow the user to select one.

[0185] For example, the user may select multiple candidates for measure reduction or note expansion by dividing the display unit 502 into sections and displaying them on the same screen of the display unit 502, and then the user may select the desired candidate from the displayed candidates using the cursor.

[0186] The number of candidates to be displayed on the display unit 502 may be preset to a predetermined number (for example, the top three) in descending order of likelihood.

[0187] Furthermore, depending on the type of user (composer, general user, etc.), purpose of use (composing, arranging, etc.), etc., a mode in which the maximum likelihood candidate is automatically adopted or a mode in which the user is allowed to select may be input from a selection section (not shown) displayed on the display unit 502, and the mode selection may be controlled by the control unit 121.

[0188] In this way, by providing the various functions described above, new value such as score editing through bar simplification and note expansion can be provided to a wide range of users in a user-friendly and convenient device for a variety of purposes.

[0189] An example of the operation of the musical score editing device 501 will be described below. In this example, measure reduction and note expansion are performed for the most likely candidates. Figures 5, 6, 7, 8, and 9 are explanatory diagrams showing examples of screen displays of a musical score editing device according to an embodiment of the present invention. The following description will be made with reference to these figures.

[0190] In the display example shown in FIG. 5, a musical score display window 702 is located at the bottom of a screen 701, and a tree display window 703 is located at the top.

[0191] The musical score display window 702 displays the musical score currently being edited.

[0192] The tree display window 703 displays a time span tree for the musical score.

[0193] The user operates a cursor 704 with a mouse or the like to specify the measure or note to be edited.

[0194] Now, let us assume that the user moves the cursor 704 to the last measure and right-clicks the mouse. Then, a context menu 705 is displayed as shown in FIG. 6. The context menu 705 contains the following options: a measure reduction object 706 for reducing the measure pointed to by the cursor 704; a note expansion object 707 that expands the note pointed to by the cursor 704; Other objects 708 that perform other operations is displayed.

[0195] In this figure, the cursor 704 is not pointing at a note, so the note expansion object 707 is unavailable. In this figure, available objects are surrounded by solid lines, and unavailable objects are surrounded by dotted lines.

[0196] In the subtree representing the last measure of the time span tree displayed in the tree display window 703, it can be seen that the first dotted quarter note first absorbs the second eighth note, and then the third quarter note.

[0197] Now, suppose the user moves cursor 704 to measure simplification object 706 and selects it by left-clicking or the like. Then, cursor 704 remains positioned where measure simplification object 706 was displayed, context menu 705 closes, and the first dotted quarter note absorbs the second eighth note to become a half note, as shown in Figure 7. In the time span tree, the absorbed branch is displayed with a dotted line.

[0198] As shown in the figure, a measure reduction menu 709 may be displayed on the musical score display window 702, which includes a pre-measure reduction playback object 710 and a post-measure reduction playback object 711. When the user moves the cursor 704 to the measure reduction menu 709 and selects the post-measure reduction playback object 711 by, for example, left-clicking, a melody based on the edited musical score is played, allowing the user to confirm the melody after reduction.

[0199] If the user wishes to compare the melodies before and after the measure reduction, the user can select the pre-measure reduction playback object 710 by, for example, left-clicking with the cursor 704, and the melody based on the score before the measure reduction will be played back, allowing the user to compare the melodies before and after the measure reduction by checking the melody before reduction.

[0200] When a selection instruction for the pre-bar reduction playback object 710 or the post-bar reduction playback object 711 is input to the reception unit 504, a melody signal is generated by the control unit 121 based on the musical score corresponding to the selection instruction, and is played back by a speaker (not shown) provided in the music reduction device 101, thereby playing back the melody.

[0201] After that, suppose the user moves the cursor 704 to a half note and right-clicks the mouse, etc. Then, as shown in Figure 8, a context menu 705 is displayed, and the note expansion object 707 becomes selectable.

[0202] When the user moves cursor 704 to note expansion object 707 and selects it by left-clicking, etc., the cursor 704 remains in the position where note expansion object 707 was displayed, the context menu 705 closes, and the half note is expanded into a dotted quarter note and an eighth note, as shown in Figure 9.

[0203] Here, a note expansion menu 712 including a pre-exploded note playback object 713 and a post-exploded note playback object 714 may be further displayed on the musical score display window 702 .

[0204] When the user moves the cursor 704 to the note expansion menu 712 and selects the object 714 to play after note expansion by, for example, left-clicking, the melody based on the musical score with the expanded notes is played, and the user can check the expanded melody.

[0205] If the user wishes to compare before and after the note development, the user can select the pre-development playback object 713 by, for example, left-clicking the cursor 704, and the melody based on the score before the note development will be played, and the user can compare the melody before and after the development by checking the melody before the development.

[0206] When a selection instruction for the pre-expansion playback object 713 or the post-expansion playback object 714 is input to the reception unit 504, a melody signal is generated by the control unit 121 based on the musical score corresponding to the selection instruction, and the melody signal is played back through a speaker (not shown) provided in the music reduction device 101.

[0207] The context menu 705, measure reduction menu 709, note expansion menu 712, etc. can be closed by clicking anywhere outside the menu. Also, a button for closing the menu may be provided within each menu.

[0208] In the above example, one computer is used to configure the note editing device 501. However, the functions of each part of the musical score editing device 501 may be distributed among multiple computers that can communicate with each other via a computer communication network. For example, as shown in Fig. 10, a cloud server 902 and an external device 903 such as a PC may be communicably connected via a computer communication network 904 such as the Internet, and the external device 903 and the cloud server 902 may cooperate to realize the functions of each part of the musical score editing device 501, and the musical score editing device 501 may function as a whole as a musical score editing system 901.

[0209] In this way, by providing music score editing as a cloud service and allowing users to edit music scores on a browser running on the external device 903, the options for usage are expanded and user convenience is further improved.

[0210] In addition, this cloud service may provide multiple applications for editing musical scores, such as a professional version for composers (with a function to automatically select the most likely option from multiple options for bar reduction and note development, or a function for the user to select the option) and a standard version for the general public, and allow users to select the desired application from the screen of an external device 903 connected to the cloud server 902.

[0211] It should be noted that a low-performance PC or tablet can also be used as the external device 903. Rendering and displaying multiple candidate scores in real time places a heavy load on the process, and a browser running on a low-performance external device 903 may not be able to effectively run the process due to memory limitations or speed considerations. In such cases, it is possible to configure the system so that the cloud server 902 first renders multiple candidate scores, and then the cloud server 902 sends the rendered score image data to the external device 903, which is the client, and the candidate scores are displayed on the external device 903.

[0212] In the above example, only one candidate is obtained for each measure reduction and note expansion. However, multiple candidates may be obtained depending on the method of measure reduction or note expansion. Therefore, the following describes an example of the operation of the above score editing system when multiple candidates are obtained.

[0213] 11, 12 and 13 are explanatory diagrams showing examples of screen displays when specifying a musical score to be edited in the musical score editing system according to an embodiment of the present invention. The following description will be made with reference to these diagrams.

[0214] When the cloud server 902 is accessed from the external device 903 via a browser, the sheet music editing starts as shown in FIG.

[0215] The user can select XML formatted musical score data prepared in the external device 903 or the like as an object to be edited, as shown in FIG.

[0216] The score to be edited will then be displayed in the browser, as shown in Figure 13.

[0217] As described above, multiple candidates may be obtained depending on the method of measure reduction. Figures 14, 15, 16, 17, and 18 are explanatory diagrams showing examples of screen displays when reducing measures in a musical score editing system according to an embodiment of the present invention. The following explanation will be given with reference to these figures.

[0218] The user can then move the arrow cursor on the score to be edited, which is displayed in the browser, as shown in Figure 13. When the user moves the arrow cursor close to the part of the score they want to simplify, the color of the notes in that part (the fourth bar in the displayed score) changes (to gray in this figure), as shown in Figure 14.

[0219] Then, as shown in FIG. 15, an inquiry about reduction candidates is made from external device 903 to cloud server 902, and nine reduction candidates are obtained as shown in FIG.

[0220] As shown in FIGS. 16 and 17, when the cursor is near a reduction candidate, the cursor changes shape to a pointing hand, and the reduction candidate can be selected.

[0221] When the user selects the fourth reduction candidate as shown in FIG. 17, the fourth measure selected in the musical score is replaced with the selected fourth reduction candidate based on the selected reduction candidate as shown in FIG. 18.

[0222] Next, we will explain the case where multiple candidates are obtained depending on the method of note expansion. Figures 19, 20, 21, 22, 23, and 24 are explanatory diagrams showing examples of screen displays when expanding notes in a musical score editing system according to an embodiment of the present invention. The following explanation will be made with reference to these figures.

[0223] As shown in FIG. 19, the user can move an arrow cursor within the musical score displayed on the browser screen.

[0224] When the user selects the part of the score where they want to expand the notes with the arrow cursor, the color of the notes in that part (the ninth bar in the displayed score) changes, as shown in FIG.

[0225] Furthermore, as shown in FIG. 21, an inquiry is made from external device 903 to cloud server 902 about deployment candidates.

[0226] As shown in FIG. 22, after nine expansion candidates are obtained, when the cursor is brought close to an expansion candidate, the shape of the cursor changes to a pointing hand, as shown in FIG. 23, and the expansion candidate can be selected.

[0227] Here, when the user selects the fourth development candidate as shown in Figure 23, the ninth bar selected in the musical score is replaced with the selected fourth development candidate based on the selected development candidate as shown in Figure 24.

[0228] In the above example, the color of the notes was changed so that the user could confirm the selected part of the score, but it would also be possible to change the background color of the part or surround it with a border, etc.

[0229] In this way, the musical score editing device 501 and the musical score editing system 901 can perform the above-mentioned editing processing, and can also be configured to apply the technology disclosed in Patent Document 1 to enable editing that merges the time span tree of the musical score of one piece of music with the time span tree of the musical score of another piece of music in a desired ratio.

[0230] (summary) As described above, the music reduction device according to this embodiment a reduction unit that performs reduction in a code string including note symbols representing pitch and duration, such that one of two adjacent note symbols is absorbed into the other; a control unit that controls the reduction unit to repeatedly perform the reduction until the number of note symbols included in the code string representing the music piece is reduced to one; A music reduction device comprising: transforming the code string using a neural network that has learned a transformation to replace one of the two adjacent note symbols that is to be absorbed with a placeholder symbol; Remove the placeholder code from the transformed code string. This is carried out by:

[0231] In addition, in the music simplifying device according to this embodiment, The placeholder code indicates whether the one to be absorbed or the other absorbing code preceded the other in the code sequence. It can be configured as follows.

[0232] In addition, in the music simplifying device according to this embodiment, In the code string, two adjacent note symbols are considered to be adjacent, In the reduction of two adjacent notes by absorbing one into the other, The sound length represented by the absorbing other part is lengthened by the sound length represented by the absorbed part. It can be configured as follows.

[0233] In addition, in the music simplifying device according to this embodiment, the code string includes a rest code representing a pause length, In the code string, two note symbols with a rest symbol between them are considered to be adjacent, In the reduction of two note symbols with a rest symbol between them, one of them is absorbed into the other, The note length represented by the absorbing other note is increased by the sum of the note length represented by the absorbed note and the rest length represented by the rest symbol. The pause code is absorbed into the other code, The absorbed one and the absorbed pause character are replaced together with one placeholder character. It can be configured as follows.

[0234] In addition, in the music simplifying device according to this embodiment, In the code sequence, A rest symbol, a first note symbol, and a second note symbol are arranged in this order. The first note symbol is absorbed into the second note symbol. In the conversion, a placeholder symbol that replaces the first note symbol is placed in a position preceding the rest symbol. It can be configured as follows.

[0235] In addition, in the music simplifying device according to this embodiment, In the code sequence, The first rest symbol, the first note symbol, the second rest symbol, and the second note symbol are arranged in this order, The first note symbol and the second rest symbol are absorbed into the second note symbol. In the conversion, a placeholder symbol that replaces the first note symbol and the second rest symbol is placed in a position preceding the first rest symbol. It can be configured as follows.

[0236] In addition, in the music simplifying device according to this embodiment, The neural network is realized by a deep learning model including a Transformer. It can be configured as follows.

[0237] In addition, in the music simplifying device according to this embodiment, The music reduction device generates a time span tree of the music from the process of the reduction that has been repeatedly executed. It can be configured as follows.

[0238] The musical score editing device according to this embodiment includes: The above music reduction device, a display unit that displays the musical score of the piece of music on a screen; a receiving unit for receiving an editing instruction input for selecting a part of the displayed musical score; a specifying unit that specifies editing content based on the selected portion; an editing unit that edits the musical score according to the specified editing content; The device is configured to include the following.

[0239] In addition, in the musical score editing device according to this embodiment, the edit instruction input is a measure reduction input for selecting a first measure from the displayed musical score, The specifying unit specifies a transformation in which a first note symbol expressing a first note included in the selected first measure is absorbed into an adjacent second note symbol in the transformation by the neural network, and in the specified transformation, a second note represented by the second note symbol before absorbing the first note symbol; a third note represented by the second note symbol after absorbing the first note symbol; Identify the The editorial department, in the musical score, Delete the first selected note, replacing the identified second note with the identified third note; It can be configured as follows.

[0240] In addition, in the musical score editing device of this embodiment, the neural network receives a code string as input and outputs one or more code strings; The specifying unit specifies the transformation by allowing the user to select one of the transformations for the one or more code sequences. It can be configured as follows.

[0241] In addition, in the musical score editing device of this embodiment, the edit instruction input is a note expansion input for selecting a first note from the displayed musical score; The identification unit inserting a placeholder symbol in the symbol string representing the music piece so as to be adjacent to a first note symbol representing the selected first note; transforming the inserted code string using an inverse neural network trained by swapping the input and output of the neural network; Identifying a transformation in which the first note symbol and the placeholder symbol are replaced with a second note symbol and a third note symbol. Thus, in the specified transformation, a second note represented by the replaced second note symbol; a third note represented by the replaced third note symbol; Identify the The editing unit replaces the selected first note with the identified second and third notes in the musical score. It can be configured as follows.

[0242] In addition, in the musical score editing device of this embodiment, The placeholder symbol indicates that the first note symbol is the other of the symbols to be absorbed in the symbol string. It can be configured as follows.

[0243] In addition, in the musical score editing device of this embodiment, the inverse neural network receives a code string as input and outputs one or more code strings; The specifying unit specifies the transformation by allowing the user to select one of the transformations for the one or more code sequences. It can be configured as follows.

[0244] In addition, in the musical score editing device of this embodiment, When the reception unit receives a pre-edit playback instruction input, the music piece related to the musical score before editing by the editing unit is played back; When the receiving unit receives the post-edit reproduction instruction input, the music piece related to the musical score after the editing by the editing unit is reproduced. It can be configured as follows.

[0245] The music editing system according to this embodiment is configured such that the above-described musical score editing device and an external device having the screen are communicably connected, the musical score is transmitted from the musical score editing device to the external device, and displayed on the screen; The editing instruction input is transmitted from the external device to the musical score editing device based on an operation performed on the screen. Configure it as follows.

[0246] The music reduction method according to this embodiment includes the steps of: The music simplification device performs the following on a code string representing a music piece, the code string including note codes representing pitches and durations: A control is performed to repeatedly execute a reduction in which one of two adjacent note symbols in the code string is absorbed into the other, until the number of note symbols included in the code string is reduced to one. A method for reducing a musical piece, the reduction comprising: transforming the code string using a neural network that has learned a transformation to replace one of the two adjacent note symbols that is to be absorbed with a placeholder symbol; Remove the placeholder code from the transformed code string. This is carried out by:

[0247] The program according to this embodiment executes the following steps: a reduction unit that performs reduction in a code string including note symbols representing pitch and duration, such that one of two adjacent note symbols is absorbed into the other; a control unit that controls the reduction unit to repeatedly perform the reduction until the number of note symbols included in the code string representing the music piece is reduced to one; The reduction is performed by: transforming the code string using a neural network that has learned a transformation to replace one of the two adjacent note symbols that is to be absorbed with a placeholder symbol; Remove the placeholder code from the transformed code string. This is carried out by:

[0248] The program can be distributed or sold by recording it on a non-transitory computer-readable information recording medium, or via a transitory transmission medium such as a computer communication network.

[0249] The present invention allows various embodiments and modifications without departing from the broad spirit and scope of the present invention. Furthermore, the above-described embodiments are intended to explain the present invention and do not limit the scope of the present invention. That is, the scope of the present invention is defined by the claims, not the embodiments. Various modifications made within the scope of the claims and the meaning of the invention equivalent thereto are considered to be within the scope of the present invention. This application claims priority based on patent application No. 2021-175236, filed in Japan on Wednesday, October 27, 2021, and the contents of that basic application are incorporated into this application to the extent permitted by the laws and regulations of the designated countries. [Industrial Applicability]

[0250] According to the present invention, it is possible to provide a music simplification device, a music simplification method, a music score editing device, a music score editing system, a program, and an information recording medium for simplifying a code string representing a piece of music. [Explanation of symbols]

[0251] 101 Music Reduction Device 111 Simplification part 112 Neural Networks 113 Adjustment section 121 Control Unit 131 Generation part 501 Music score editing device 502 Display section 503 Reception 504 Specific part 505 Editorial Department 701 screens 702 Music score display window 703 Tree View Window 704 Cursor 705 Context Menu 706 Bar Reduction Objects 707 Note Expansion Object 708 Other Objects 709 Bar Simplification Menu 710 bar reduction pre-playback object 711 Bar reduction playback object 712 Note Expansion Menu 713 Pre-expansion note playback object 714 Note expansion playback object 901 Music Score Editing System 902 Cloud Server 903 External device 904 Computer Communication Network

Claims

1. a reduction unit that performs reduction in a code string including note symbols representing pitch and duration, such that one of two adjacent note symbols is absorbed into the other; a control unit that controls the reduction unit to repeatedly perform the reduction until the number of note symbols included in the code string representing the music piece is reduced to one; A music reduction device comprising: transforming the code string using a neural network that has learned a transformation to replace one of the two adjacent note symbols that is to be absorbed with a placeholder symbol; Remove the placeholder code from the transformed code string. This is carried out by In the code string, two adjacent note symbols are considered to be adjacent. A music reduction device characterized by:

2. The placeholder code indicates whether the one to be absorbed or the other absorbing code preceded the other in the code sequence.

2. The music reduction device according to claim 1,

3. In a reduction in which one of two adjacent musical notation symbols is absorbed into the other, The sound length represented by the absorbing other part is lengthened by the sound length represented by the absorbed part.

3. The music reduction device according to claim 1 or 2.

4. the code string includes a rest code representing a pause length, In the code string, two note symbols with a rest symbol between them are considered to be adjacent, In the reduction of two note symbols with a rest symbol between them, one of them is absorbed into the other, The note length represented by the absorbing other note is increased by the sum of the note length represented by the absorbed note and the rest length represented by the rest symbol. The rest symbol is absorbed into the other symbol, The absorbed rest symbol and the absorbed rest symbol are replaced by one placeholder symbol.

4. The music reduction device according to claim 3.

5. In the code sequence, A rest symbol, a first note symbol, and a second note symbol are arranged in this order. The first note symbol is absorbed into the second note symbol. In the conversion, a placeholder symbol that replaces the first note symbol is placed in a position preceding the rest symbol.

5. The music reduction device according to claim 4.

6. In the code sequence, The first rest symbol, the first note symbol, the second rest symbol, and the second note symbol are arranged in this order, The first note symbol and the second rest symbol are absorbed into the second note symbol. In the conversion, a placeholder symbol that replaces the first note symbol and the second rest symbol is placed in a position preceding the first rest symbol.

5. The music reduction device according to claim 4.

7. The neural network is realized by a deep learning model including a Transformer.

2. The music reduction device according to claim 1,

8. The music reduction device generates a time span tree of the music from the process of the reduction that has been repeatedly executed.

2. The music reduction device according to claim 1,

9. The music reduction device according to claim 1; a display unit that displays the musical score of the piece of music on a screen; a receiving unit for receiving an editing instruction input for selecting a part of the displayed musical score; a specifying unit that specifies editing content based on the selected portion; an editing unit that edits the musical score according to the specified editing content; A musical score editing device comprising:

10. the edit instruction input is a measure reduction input for selecting a first measure from the displayed musical score, The specifying unit specifies a transformation in which a first note symbol expressing a first note included in the selected first measure is absorbed into an adjacent second note symbol in the transformation by the neural network, and in the specified transformation, a second note represented by the second note symbol before absorbing the first note symbol; a third note represented by the second note symbol after absorbing the first note symbol; Identify the The editorial department, in the musical score, Delete the first selected note, replacing the identified second note with the identified third note; 10. The musical score editing device according to claim 9.

11. the neural network receives a code string as input and outputs one or more code strings; The specifying unit specifies the transformation by allowing the user to select one of the transformations for the one or more code sequences.

11. The musical score editing device according to claim 10.

12. the edit instruction input is a note expansion input for selecting a first note from the displayed musical score; The identification unit inserting a placeholder symbol in the symbol string representing the music piece so as to be adjacent to a first note symbol representing the selected first note; transforming the inserted code string using an inverse neural network trained by swapping the input and output of the neural network; Identifying a transformation in which the first note symbol and the placeholder symbol are replaced with a second note symbol and a third note symbol. Thus, in the specified transformation, a second note represented by the replaced second note symbol; a third note represented by the replaced third note symbol; Identify the The editing unit replaces the selected first note with the identified second and third notes in the musical score.

10. The musical score editing device according to claim 9.

13. The placeholder symbol indicates that the first note symbol is the other of the symbols to be absorbed in the symbol string.

13. The musical score editing device according to claim 12.

14. the inverse neural network receives a code string and outputs one or more code strings; The specifying unit specifies the conversion by selecting one of the conversions for the one or more output code sequences.

13. The musical score editing device according to claim 12.

15. When the reception unit receives a pre-edit playback instruction input, the music piece related to the musical score before editing by the editing unit is played back; When the receiving unit receives the post-edit reproduction instruction input, the music piece related to the musical score after the editing by the editing unit is reproduced.

10. The musical score editing device according to claim 9.

16. A musical score editing system in which the musical score editing device according to claim 9 and an external device having the screen are communicably connected, the musical score is transmitted from the musical score editing device to the external device, and displayed on the screen; The editing instruction input is transmitted from the external device to the musical score editing device based on an operation performed on the screen. A music score editing system characterized by:

17. The music simplification device performs the following on a code string representing a music piece, the code string including note codes representing pitches and durations: A control is performed to repeatedly execute a reduction in which one of two adjacent note symbols in the code string is absorbed into the other, until the number of note symbols included in the code string is reduced to one. A method for reducing a musical piece, the reduction comprising: transforming the code string using a neural network that has learned a transformation to replace one of the two adjacent note symbols that is to be absorbed with a placeholder symbol; Remove the placeholder code from the transformed code string. This is carried out by In the code string, two adjacent note symbols are considered to be adjacent. A method for simplifying a musical piece, comprising:

18. Computer, a reduction unit that performs reduction in a code string including note symbols representing pitch and duration, such that one of two adjacent note symbols is absorbed into the other; a control unit that controls the reduction unit to repeatedly perform the reduction until the number of note symbols included in the code string representing the music piece is reduced to one; The reduction is performed by: transforming the code string using a neural network that has learned a transformation to replace one of the two adjacent note symbols that is to be absorbed with a placeholder symbol; Remove the placeholder code from the transformed code string. This is carried out by In the code string, two adjacent note symbols are considered to be adjacent. A program characterized by:

19. A non-transitory computer-readable information recording medium on which the program according to claim 18 is recorded.

Images