TWI784434B - System and method for automatically composing music using approaches of generative adversarial network and adversarial inverse reinforcement learning algorithm - Google Patents

Abstract

The present invention discloses a system and a method for automatically composing music using approaches of generative adversarial network and adversarial inverse reinforcement learning algorithm, comprising: a music feature extracting unit, a first operation module, a second operation module and a third operation module. The music feature extracting unit performs a feature extracting process on each of reference music data stored in the reference music database, thereby extracting a plurality of music features. Particularly, the first operation module performs a long and short-term memory (LSTM) neural network algorithm with bi-axial architecture. The second operation module uses an adversarial inverse reinforcement learning (AIRL) algorithm to perform a second operation on the plurality of music features, thereby obtaining at least one reward function. According to foregoing features of the present invention, the music generated by the system and method of the present invention are melodic and like the music composed by humans.

Description

Automatic composition system and method using adversarial generative network and inverse reinforcement learning method

The invention relates to the technical field of an automatic composition system, in particular to an automatic composition system and method using an adversarial generation network and an adversarial inverse reinforcement learning method.

Music is an extremely important part of human life. It is not only used as a way to relax in daily life, but also according to ancient Roman literature, music has curative effects on the human body and can improve human mood. According to a special report, music has the following seven benefits for people: increasing learning efficiency, relieving stress, having analgesic effect, enhancing memory, improving insomnia, improving exercise efficiency, and promoting a happier mood.

However, in the traditional composing method, the composer must learn many years of musical instrument skills and knowledge of music theory and spend many days to complete a piece of music; Unique tunes, many automatic composition systems have been proposed. One of the automatic composition systems uses a supervised deep learning (Deep Supervised Learning) algorithm as a model. But the above-mentioned supervised deep learning algorithm reuses the same melody excessively, and the music it produces has the disadvantage of being unpleasant.

From the above description, it can be seen that it is necessary to improve and redesign the existing automatic composition system so that it can produce more melodious and popular music. In view of this, the inventor of this case made great efforts to research and create, and finally developed an automatic composition system and method using confrontational generative network and inverse reinforcement learning method of the present invention.

The main purpose of the present invention is to provide an automatic composition system and method using adversarial generative network and inverse reinforcement learning method, wherein the automatic composition system is applied to an electronic device, so that the electronic device is based on a plurality of references A music is generated from the music data, and includes: a music database, a music feature extraction unit, a first computing module, a second computing module and a third computing module. The music database is used to store a plurality of reference music data. In particular, the music feature extraction unit performs a feature extraction process on each of the reference music data, thereby extracting a plurality of music features. Next, the first operation module is used to perform a first operation on the music feature by using a deep learning algorithm, so as to obtain at least one music probability feature and at least one pre-training weight parameter. In addition, the second operation module uses a reinforcement learning algorithm to perform a second operation on the plurality of music features to obtain at least one note reward function. Furthermore, the third calculation module uses a deep reinforcement learning algorithm to perform an initialization setting through the pre-training weight parameters, and rewards the plurality of sets of music theory data stored in a music theory database, the at least one note The function and the at least one pre-training weight parameter perform a third operation, so as to obtain a plurality of polyphonic reference music data; the polyphonic reference music data is pleasant and popular music.

In order to achieve the above-mentioned main purpose of the present invention, the inventor of this case provides an embodiment of the automatic composition system using the confrontation generation network and the inverse reinforcement learning method, which is applied to an electronic device, so that the electronic device is based on A plurality of reference music data are used to generate a music; the automatic composition system includes: a music database for storing the plurality of reference music data; a music feature extraction unit, configured to perform a feature extraction process on each of the reference music data, thereby extracting a plurality of music features; A first operation module, for performing a first operation on the plurality of music features by using a deep enhanced learning algorithm, so as to obtain at least one music probability feature and at least one pre-training weight parameter; A second operation module, used for performing a second operation on the plurality of music features by using a reinforcement learning algorithm, so as to obtain at least one note reward function; A third calculation module is used to use a deep reinforcement learning algorithm to perform an initialization setting with the at least one pre-training weight parameter, and for the complex set of music theory data stored in a music theory database, the at least one A note reward function and the at least one pre-trained weight parameter perform a third operation, so as to obtain a plurality of complex music data.

And, in order to achieve the main purpose of the above-mentioned present invention, the inventor of the present case provides an embodiment of the automatic composition method using the confrontation generation network and the inverse reinforcement learning method at the same time, which is applied to an electronic device through the A processor of the electronic device implements, and includes the following steps: (1) providing a music feature extraction unit for performing a feature extraction process on a plurality of music data stored in a music database of the electronic device, thereby extracting a plurality of music features; (2) providing a first operation module to perform a first operation on the plurality of music features by using a deep learning algorithm, thereby obtaining at least one music probability feature and at least one pre-training weight parameter; (3) providing a second operation module to perform a second operation on the plurality of music features by using an enhanced learning algorithm, so as to obtain at least one note reward function; (4) Provide a third calculation module to use a deep reinforcement learning algorithm to initialize the at least one pre-training weight parameter, and to store the plurality of sets of music theory data stored in the music theory database, the at least one The note reward function and the at least one pre-trained weight parameter perform a third operation to obtain a plurality of complex music data.

In order to more clearly describe an automatic composition system and method using an adversarial generative network and an inverse reinforcement learning method proposed by the present invention, preferred embodiments of the present invention will be described in detail below with reference to the drawings.

…………….(1) FIG. 1 shows a perspective view of an electronic device applied with an automatic composition system using an adversarial generative network and an inverse reinforcement learning method of the present invention. Moreover, FIG. 2 shows a functional block diagram of an automatic composition system using an adversarial generative network and an inverse reinforcement learning method of the present invention. As shown in FIG. 1 , an automatic music composition system 1 using an adversarial generative network and an inverse reinforcement learning method of the present invention is applied to an electronic device 2 . In one embodiment, the automatic composition system 1 of the present invention is installed in an operating system (OS) of the electronic device 2, so that a processor of the electronic device 2 can execute the automatic composition system 1 of the present invention according to A plurality of reference music data are used to generate a music. As shown in Figure 2, the automatic composition system 1 of the present invention mainly includes: a music database 11, a music feature extraction unit 12, a first computing module 13, a second computing module 14, a third computing module Group 15 and a database 16 of music theory. Wherein, the music database is used to store a plurality of music data, and the music theory database 16 is used to store a plurality of music theory data. The music feature extraction unit 12 performs a feature extraction process on each of the reference music data, thereby extracting a plurality of music features. More specifically, the feature of the music is to record the pitch of the music in MIDI (Musical Instrument Digital Interface) value. Fig. 3 shows a schematic diagram of recording the pitch of a music piece. As shown in Figure 3, in the matrix on the right, the horizontal row (row) represents the chord played for each beat, and the vertical row (column) is the pitch recorded in MIDI value; among them, Figure 3 is 8 The chord of a beat; from the matrix in Figure 3, it can be known that the first note's performance beat is two and one-half beats. That is to say, the polyphonic (harmony) melody of the music can be recorded through this recording method. Then, the schematic matrix of the music feature extraction unit 12 is shown in (1):

…………….(1)

In the above formula (1), N is the MIDI value of each pitch, T is the beat number, p records whether there is performance, and a is the corresponding articulation. It is supplemented that the first operation module 13 uses a deep enhanced learning algorithm to perform a first operation on the plurality of music features, so as to obtain at least one music probability feature and at least one pre-training weights (weights) parameter. FIG. 4 shows a functional block diagram of the first computing module. As shown in FIG. 4 , the first computing module 13 includes: a first computing unit 131 , a second computing unit 132 , a third computing unit 133 and a fourth computing unit 134 . More specifically, the first calculation unit 131 is used to convert each of the music features into a note vector feature; wherein, the length of the note vector feature is 79, and is composed of the current pitch vector feature, the current pitch vector feature, the pitch vector feature of the front note, the pitch vector feature of the back note, and the beat vector feature. Next, the second computing unit 132 is configured to use a deep learning algorithm to perform a fourth computing along the time axis on the note vector features, so as to obtain at least one time parameter. Moreover, the third operation unit 133 is configured to use a deep learning algorithm to perform a fifth operation along the note axis on the note vector features, so as to obtain at least one note parameter. It is worth noting that the deep learning algorithm is a long short-term memory (Long short-term memory, LSTM) neural network algorithm. Wherein, the first computing module 13 uses the second computing unit 132 and the third computing unit 133 to perform the long-short-term memory neural network algorithm along the time axis and the note axis respectively. In other words, the present invention The first calculation module 13 adopts a biaxial structure (Bi-axial architecture) long short-term memory neural network algorithm.

As mentioned above, the fourth operation unit 134 uses a deep learning algorithm (non-recursive linear algorithm) to perform a fifth operation on the time parameter and the note parameter, so as to obtain at least one training weight parameter and at least one note The music probability feature composed of the probability feature and at least one articulation probability feature. It is supplemented that, in order to avoid the overfitting of the LSTM neural network algorithm, the first calculation module 13 of the present invention uses a dropout algorithm in the fourth calculation and the fifth calculation.

………….…..(4) According to the above, the second operation module 14 is used for performing a second operation on the plurality of music features by using a reinforcement learning algorithm, so as to obtain at least one note reward function. More specifically, the reinforcement learning algorithm is an Adversarial Inverse Reinforcement Learning (AIRL) algorithm. The adversarial inverse reinforcement learning algorithm is similar to the structure of the conventionally known Generative Adversarial Guided Cost Learning (GAN-GCL); in other words, through the second computing module 14 utilized The inverse reinforcement learning algorithm to obtain the maximum reward (Reward), and then obtain the note reward function. Wherein, it respectively trains a generator and a discriminator simultaneously. Wherein, the aforementioned discriminator can be deduced by using the following formulas (2), (3) and (4). r _t (s,a)=c* r mt(s,a)+(1-c)* r airl(s,a) ……………..(2) L(θ)=E[r( s,a)+ γ(maxa'Q(s',a';θ ^- )-Q(s,a;θ)) ² ] …………….(3)

……………..(4)

為實際分布p(τ)被以波爾曼分佈(Boltzmann distribution)所表示之，且其獎勵函數為能量函數。繼續地說明本發明之技術，該第三運算模組15用以利用一深度強化學習演算法藉由該至少一預訓練權重參數進行一初始化設定，且對該儲存於一樂理資料庫16之中的複數組樂理資料、該至少一音符獎勵函數以及該至少一預訓練權重參數執行一第三運算，從而獲取複數個複音樂曲資料。其中，本實施例之第三運算模組15所使用之深度強化學習演算法為一深度Q網絡學習(Deep Q Learning network, DQN)演算法結合所述雙軸式長短期記憶神經網路演算法。更詳細地說明，本發明人設定不同數值的獎勵折扣因子c(Reward discount factor)之模擬與實驗，得到以下表(1)： Value of c 0.25 0.5 0.75 樂理獎勵 32.55 34.27 34.15 AIRL獎勵 25.3 29.71 22.46 平均值 28.93 31.99 28.31 表(1) In the above formula (2) and the above formula (3), where c is a constant, s is the current state, a is the current action, s' is the next state, a' is the next action, r mt is the music theory reward function, r airl is the reward function of confrontational inverse reinforcement learning, θ ^- is the weight of the target Q network (Target-Q network), and γ is the discount factor for future rewards. In the above formula (4), where θ is the weights of the Q-network (Q-network), and q(τ) is the generator density. and,

The actual distribution p(τ) is represented by a Boltzmann distribution, and its reward function is an energy function. Continuing to illustrate the technology of the present invention, the third calculation module 15 is used to use a deep reinforcement learning algorithm to perform an initialization setting through the at least one pre-training weight parameter, and store it in a music theory database 16 A third operation is performed on the complex sets of music theory data, the at least one note reward function and the at least one pre-trained weight parameter, so as to obtain a plurality of complex music data. Wherein, the deep reinforcement learning algorithm used by the third computing module 15 of this embodiment is a Deep Q Learning network (DQN) algorithm combined with the biaxial long-short-term memory neural network algorithm. To explain in more detail, the inventors set different values of the reward discount factor c (Reward discount factor) for simulation and experimentation, and obtained the following table (1): Value of c 0.25 0.5 0.75 Music Theory Award 32.55 34.27 34.15 AIRL rewards 25.3 29.71 22.46 average value 28.93 31.99 28.31 Table 1)

Next, the inventor compares the complex music data produced by the present invention with the music produced by different algorithms, the comparison is the music data generated by the algorithm and the existing human music (that is, the reference music) data or master music data), the comparison results are shown in the following table (2): Biaxial LSTM music theory AIRL Music theory+AIRL Ratio polyphonicity 0.108861 0.084123 0.040719 0.020159 Ratio steps in scale 0.301088 0.294629 0.092172 0.171911 Ratio empty bars 0.002749 0.000134 0.051174 0.020323 Ratio unique chords 0.540401 0.192479 0.026132 0.078249 Ratio chords repeated 0.157354 0.138524 0.085684 0.031908 Ratio tonality 0.022615 0.085325 0.058328 0.006202 Ratio chords in repeated motifs 0.275816 0.222622 0.270690 0.264218 Table 2)

It can be seen from the above table (2) that among the difference comparison values of the music theory and the AIRL algorithm used in this embodiment, three values are the lowest. It is worth noting that the lower the value, the more similar to the music made by humans. In other words, the smaller the difference value of the aforementioned difference comparison, the closer the music created by the automatic composition system 1 of the present invention is to the music composed by humans. Further, the inventor conducted a user preference survey on different algorithm frameworks and human music, as shown in the following table (3): Biaxial LSTM music theory AIRL Music theory+AIRL human composition Total times preferred 20 66 30 69 75 Percentage preferred 19% 63% 29% 66% 72% Standard Deviation 4.02 4.91 4.62 4.82 4.57 table 3)

It can be known from the above table (3) that the degree of preference obtained by the music theory and the AIRL algorithm architecture adopted in this embodiment is only lower than that of the music composed by humans, and the difference is small. From the objective analysis of the above table (2) and the subjective analysis of the table (3), it can be concluded that the music produced by the automatic composition system of the present invention against the generative network and the inverse reinforcement learning method is the closest to the music made by humans. In other words, the repertoire produced by the present invention has the advantage of being pleasing to the ear and loved by humans.

In this way, the above-mentioned system has fully described an automatic composition system using the confrontation generation network and the inverse reinforcement learning method of the present invention. Next, the following will continue to describe an automatic composition system using the confrontation generation network and the inverse reinforcement learning method Automatic composition method. Continuing to refer to FIG. 1 and FIG. 2 , and refer to FIG. 5 and FIG. 6 at the same time, which show a first flowchart and a second flowchart of an automatic composition method using an adversarial generative network and an inverse reinforcement learning method of the present invention. The automatic composition method using the confrontation generative network and the inverse reinforcement learning method of the present invention is applied in an electronic device 2, so that the processor of the electronic device 2 can perform the automatic composition method 1 according to a plurality of A music is generated by referring to the music data. As shown in FIG. 5 and FIG. 6 , the automatic composition method of the present invention includes multiple execution steps. First, in step S1, the music feature extraction unit 12 performs a feature extraction process on the plurality of reference music data stored in the music database 11, thereby extracting a plurality of music features.

As shown in FIG. 5 and FIG. 6, the method flow is followed by step S2: the first operation module uses a deep learning algorithm to perform a first operation on the plurality of music features, thereby obtaining at least one music probability feature and at least one A pretrained weight parameter. It is worth noting that the method flow will execute step S3: the second operation module (AIRL) uses an enhanced learning algorithm to perform a second operation on the plurality of music features, thereby obtaining at least one note reward function (Reward function ). Further, in step S4, the third operation module uses a deep reinforcement learning algorithm to perform an initialization setting on the at least one pre-training weight parameter, and performs an initialization setting on the complex sets of music theory data stored in the music theory database , the at least one note reward function and the at least one pre-trained weight parameter perform a third operation, so as to obtain a plurality of polyphonic reference music data .

In more detail, step S2 includes the following steps: in step S21, a first computing unit 131 of the first computing module 13 converts each of the music features into a note vector feature. Next, step S22 is executed, a second computing unit 132 of the first computing module 13 uses a deep learning algorithm to perform a fourth computing along the time axis on the note vector features, so as to obtain at least one time parameter. As shown in FIG. 5 , the method flow is then executed in step S23. A third computing unit 133 of the first computing module 13 is used to perform a first step along the note axis on the note vector feature using a deep learning algorithm. Five operations, so as to obtain at least one note parameter. Finally, a fourth calculation unit 134 of the first calculation module 13 is used to perform a fifth calculation on the time parameter and the note parameter by using a deep learning algorithm (non-recursive linear algorithm), so as to obtain at least A training weight parameter and the music probability feature composed of at least one note probability feature and at least one articulation probability feature.

In this way, the above system has completely and clearly described an automatic composition system and method using the confrontation generation network and the inverse reinforcement learning method of the present invention; and, it can be seen from the above that the present invention has the following advantages:

(1) The present invention mainly uses a music database 11, a music feature extraction unit 12, a first computing module 13, a second computing module 14, and a third computing module 15 to form the use confrontation generation of the present invention An automatic music composition system for networks and inverse reinforcement learning1. In particular, the music feature extraction unit 12 performs a feature extraction process on each of the reference music data, thereby extracting a plurality of music features. Wherein, the feature of the music is to record the harmony of the music in the form of matrix. In addition, the second operation module 14 uses an inverse reinforcement learning algorithm to perform a second operation on the plurality of music features to obtain at least one note reward function. The third operation module 15 uses a deep reinforcement learning algorithm to perform an initialization setting by at least one pre-training weight parameter, and executes a method on the complex set of music theory data, the value of the note reward function and the pre-training weight parameter. The third operation is to obtain a plurality of polyphonic reference music materials. Through the above objective and subjective analysis, it can be known that the music produced by the automatic composition system 1 using the confrontation generative network and the inverse reinforcement learning method of the present invention is the closest to the music made by humans. Moreover, the music produced by the system of the present invention is not only pleasing to the ear but also more popular with human beings.

It must be emphasized that the above detailed description is a specific description of a feasible embodiment of the present invention, but the embodiment is not used to limit the patent scope of the present invention, any equivalent implementation or modification that does not depart from the technical spirit of the present invention, All should be included in the patent scope of this case.

<The present invention> 1: Automatic composition system 2: Electronic device 11: Song Database 12: Music Feature Extraction Unit 13: The first computing module 131: The first computing unit 132: Second computing unit 133: The third computing unit 134: The fourth computing unit 14: Second computing module 15: The third computing module 16: Music theory database S1~S4: steps S21~S24: Steps

＜Knowledge＞ none

FIG. 1 shows a perspective view of an electronic device applied with an automatic composition system using an adversarial generative network and an inverse reinforcement learning method according to the present invention; Fig. 2 shows the functional block diagram of the automatic composition system using the confrontation generation network and the inverse reinforcement learning method of the present invention; Figure 3 shows a schematic diagram of recording the pitch of a musical piece; Fig. 4 shows the functional block diagram of the first computing module; Fig. 5 shows the first flow chart of the automatic composition method using the confrontation generation network and the inverse reinforcement learning method of the present invention; and FIG. 6 shows the second flow chart of the automatic composition method using adversarial generative network and inverse reinforcement learning method of the present invention.

1: Automatic composition system using adversarial generative networks and inverse reinforcement learning 11: Song Database 12: Music Feature Extraction Unit 13: The first computing module 14: Second computing module 15: Second computing module 16: Music Theory Database

Claims (17)

An automatic composition system, which is applied in an electronic device, so that the electronic device generates a music according to a plurality of reference music data; the automatic composition system includes: a music database for storing the plurality of reference music data; A music feature extraction unit is used to perform a feature extraction process on each of the reference music data, thereby extracting a plurality of music features; a first computing module is used to use a deep enhanced learning algorithm to extract a plurality of music features. The music feature performs a first operation, thereby obtaining a music feature probability (probability) and at least one pre-training weight parameter that can be represented by a Bollman distribution (Boltzmann distribution); wherein, the plurality of music features are stored based on MIDI values. The pitch of the note, and the music feature probability includes at least one selected from the group consisting of the note probability and the articulation probability; a second operation module for using a reinforcement learning algorithm to The plurality of music features perform a second operation to obtain at least one note reward function; a third operation module is used to use a deep reinforcement learning algorithm to perform an initialization setting by using the at least one pre-training weight parameter, And a third operation is performed on the complex sets of music theory data stored in a music theory database, the at least one note reward function and the at least one pre-trained weight parameter, so as to obtain a plurality of complex music data. The automatic composition system according to claim 1, wherein the first computing module includes: A first calculation unit is used to convert each of the music features into a note vector feature; a second calculation unit is used to perform a fourth algorithm along the time axis on the note vector feature using a deep learning algorithm. operation, so as to obtain at least one time parameter; a third operation unit, for using a deep learning algorithm to perform a fifth operation along the note axis on the note vector feature, thereby obtaining at least one note parameter; a fourth The computing unit is configured to use a non-recursive deep learning algorithm to perform a fifth operation on the time parameter and the note parameter, so as to obtain at least one training weight parameter, the note probability, and the articulation probability. The automatic composition system according to claim 2, wherein the first calculation module uses a discarding algorithm in the fourth calculation and the fifth calculation, so as to prevent the deep learning algorithm from overfitting. The automatic composition system as described in claim 2, wherein the deep learning algorithm is a long short-term memory neural network algorithm. The automatic composition system as described in Claim 1, wherein the deep reinforcement learning algorithm is a deep Q network learning algorithm. The automatic composition system as described in claim 4, wherein the deep reinforcement learning algorithm is a deep Q network learning algorithm combined with the long short-term memory neural network algorithm method, and the long-short-term memory neural network algorithm is a biaxial long-short-term memory neural network algorithm. The automatic composition system as described in Claim 1, wherein the reinforcement learning algorithm is an adversarial inverse reinforcement learning algorithm. The automatic composition system as described in claim 2, wherein the length of the note vector feature is 79, and is composed of the current pitch vector feature, the current pitch vector feature, the pitch vector feature of the previous note, and the pitch of the back note Vector features, and beat vector features. An automatic composition method is applied to an electronic device to be realized by a processor of the electronic device, and includes the following steps: (1) providing a music feature extraction unit for storing in a music database of the electronic device A feature extraction process is performed on the plurality of music data, thereby extracting a plurality of music features; (2) providing a first computing module to perform a first calculation on the multiple music features using a deep learning algorithm , so as to obtain a music feature probability (probability) and at least one pre-training weight parameter that can be represented by Bollman distribution (Boltzmann distribution); wherein, the plurality of music features are the pitches of multiple notes stored based on MIDI values, and the The music feature probability includes at least one selected from the group consisting of note probability and articulation probability; (3) provide a second operation module to utilize a reinforcement learning algorithm to perform a second operation on the plurality of music features, thereby obtaining at least one note reward function; (4) provide a third operation module to utilize A deep reinforcement learning algorithm performs an initialization setting on the at least one pre-training weight parameter, and executes on the plurality of sets of music theory data stored in the music theory database, the at least one note reward function, and the at least one pre-training weight parameter A third operation to obtain a plurality of polyphonic music data. The automatic composition method as described in claim 9, wherein the first computing module includes a first computing unit, a second computing unit, a third computing unit and a fourth computing unit, and the steps ( 2) further comprising the following steps: (21) the first computing unit converts each of the music features into a note vector feature; (22) the second computing unit utilizes a deep learning algorithm to perform the following steps on the note vector feature A fourth operation along the time axis to obtain at least one time parameter; (23) the third operation unit uses a deep learning algorithm to perform a fifth operation on the note vector feature along the note axis to obtain at least A note parameter; (24) the fourth operation unit utilizes a non-recursive deep learning algorithm to perform a fifth operation on the time parameter and the note parameter, thereby obtaining at least one training weight parameter, the note probability and The articulation probability. The automatic composition method according to claim 10, wherein, when performing the first calculation, the first calculation module uses a dropout algorithm (dropout) to prevent a calculation result of the deep learning algorithm from being excessive fit. The automatic composition method as described in Claim 9, wherein the deep learning algorithm is a long short-term memory neural network. The automatic composition method as described in claim 9, wherein the deep reinforcement learning algorithm is a deep Q learning algorithm (Deep Q Learning network, DQN). The automatic composition method as described in claim 13, wherein the deep reinforcement learning algorithm is a deep Q learning algorithm (Deep Q Learning network, DQN) combined with the long-short-term memory neural network algorithm, and the long-short-term The memory neural network algorithm is a two-axis long-short-term memory neural network algorithm. The automatic composition method system as described in Claim 9, wherein the reinforcement learning algorithm is an inverse reinforcement learning algorithm. The automatic composition method as described in claim 10, wherein the length of the note vector feature is 79, which is composed of the current pitch vector feature, the current pitch vector feature, the pitch vector feature of the previous note, and the pitch vector of the back note features, and beat vector features. The method for automatically composing music according to Claim 9, wherein the feature of the plurality of pieces of music is storing the pitches of the plurality of notes based on MIDI values.

Images