JPS63311395A - Melody analyzing machine and automatically composing machine - Google Patents

Abstract

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

Description

[Detailed description of the invention] [Technical field of invention] The present invention relates to a melody analyzer and an automatic composer.

[Background] Regarding the quality of automatic music composers, one of the important factors to consider is the ability to produce music that is rich in musicality rather than purely mechanical. The question is whether the composing machine has its potential.

For example, Japanese Patent Application No. 58-125603 (Special Publication No. 60-4
No. 0027), individual pitch data is randomly sampled from a series of pitch data (for example, 12-tone scale data), and if the sampled data satisfies a limited condition,
An automatic composing machine has been disclosed that adopts it as a melody note, does not adopt it as a melody note unless conditions are met, samples it again, and repeats the condition check. Therefore, the melody generation process of this automatic music composer is basically a trial-and-error process. When pitch data is randomly sampled, a completely disordered sequence of pitches is created. This chaotic sequence of pitches cannot form a melody at all (there is a possibility that a good melody can be created by astronomical contingencies), so in order to bring some order to this disorder, we decided to perform a conditional test. A type of filtering (selection) is performed.In this case, the degree of selection is an important factor.If the selection is too strict, the generated melody will have a single pattern, and if it is too loose, it will be a single pattern. Anarchy will prevail.

The above-mentioned automatic composing machine is suitable for composing elusive melodies of a single piece rather than melodies with which one person is familiar, and is mainly effective as a music composition device for listening training and performance practice (familiar). (In general, it is difficult to transcribe and perform innovative pieces that do not have a unique score.) In this sense, he does not have the abilities listed at the beginning.

In view of these points, the applicant has filed an application for an automatic production machine that can develop and grow a motif based on a motif input by the user, and control the consistency and diversity of a song. (patent application,
Named ``Automatic Composer'', filed April 8, 1985 and May 20, 1988), its basic features are a means to evaluate and analyze motifs given by users, and evaluation results. Plan the outline and style of the melody to be generated based on motif characteristic parameters.

Disclosed are associated melody control parameter generation means, and melody generation/execution means for specifically generating a melody according to planned control parameters and chord progression information.

The automatic composing machines related to these applications are based on the overall characteristics of the piece, 4.
He controls the atmosphere and style, has a plan in each composition, and is able to create songs that are rich in diversity and hierarchy. This is one of the major advantages. The second feature is that a motif is input by the user and a composition is created based on that motif. This feature effectively deepens the user's satisfaction and increases their sense of participation in the composition.

The third feature is that an approach is adopted in which harmonic sounds and non-harmonic sounds are treated separately.

That is, the motif evaluation means (motif feature parameter extraction means) includes at1@ which extracts harmonic feature parameters regarding non-harmonic sounds and harmonic sounds included in the motif.
The melody generation means (melody control parameter generation means and melody generation execution means) also include the extracted harmonic characteristic parameters (for example, the type of harmonic 1 dispersion chord, various non-harmonic ff Contains a function to generate melodies in a format that reflects and reflects the number ゛). like this,
The harmonic/inharmonious tones distinction feature helps to make the generated songs less mechanical and more musical.

Furthermore, related to the third feature, the harmonic feature parameter extracting means can be used as a harmonic analyzer of the melody. Through the melody analyzer, students can easily learn the features of non-harmonic sounds and the functions of harmonic sounds in melody lines.

In the above-mentioned application, a means for classifying and extracting non-harmonic sounds is shown as being related to melody analysis. The logic of this non-harmonic sound classification extraction means is to convert the pitch sequence of a motif (melody) into a pitch progression sequence.

This interval progression is the condition (pattern) for various non-harmonic tones.
If there is a match, it is estimated that the corresponding note in the motif (melody) is the non-harmonic note.The conversion from a pitch sequence to an interval progression sequence is based on the pitch of an acrobatics. It is executed by means of calculating the difference.

The present invention is characterized in that non-harmonic sounds are extracted or classified using a new logic different from the above-mentioned logic.

[Object of Part II] That is, an object of the present invention is to provide a melody analyzer that automatically performs harmonic evaluation of a melody using a novel approach. Another objective is to improve automatic composers of the type described above, and in particular to provide automatic composers that are improved in terms of motif analysis.

[Summary of the Invention] In order to achieve the above object, the melody analyzer according to the present invention is provided with a melody code input means in addition to the melody input means, and based on the input melody chord, the melody analyzer analyzes non-harmonic sounds included in the melody. The main point is that each non-harmonic sound is classified based on the non-harmonic sound information and the melody sound sequence that are the result of this identification.

In general, the automatic music composer according to the present invention is characterized by using the functions of the above-mentioned melody analyzer in the area of motif analysis, specifically.

In addition to the motif input means, a motif code input means is provided, and non-harmonic sounds included in the motif are identified or extracted based on the input motif code.
In this automatic composer, the harmonic feature parameters of the motif are obtained from the extraction results of such non-harmonic sounds, and the harmonic and non-harmonic sounds are created in a format that reflects these harmonic feature parameters. The structure is such that a melody containing the song is generated.

[Operation and development of the invention] A person who intends to use the melody analyzer according to the present invention to learn harmonic techniques of melodies inputs a melody from an input device, and also inputs the background of the melody. Enter the code that appears. On the other hand, the melody analyzer assumes that the constituent tones of the input chord are harmonic tones, and estimates that the sounds in the input melody other than the chord constituent tones are out-of-phase or IS tones. Furthermore, the melody analyzer performs a pattern analysis on the melody note sequence based on the estimated non-harmonic sound information (information indicating which number and number II of the melody note sequence are non-harmonic sounds). Estimate which sound is which type of non-harmonic sound. As a result, the non-phasic -5 tones are classified.

After that, the classified non-harmonic sound information and melody can be outputted through a suitable output means. Any configuration or output format can be used for the output means.For example, the melody can be displayed on a staff notation on the screen of a display device such as a CRT, and non-harmonic sounds classified into one category can be identified and displayed, or the sound source can be The melody is output as 51 through the sound system. You can also output the code.

Here, students can learn whether the chords they have imagined match the melody, as well as the melodic characteristics and functions of various non-harmonic tones in the melody line, sums, and the functions of 41 m. By using a melody analyzer, students will gradually deepen their pattern recognition and learn what kind of chords go with what kind of melody.
You will also acquire the ability to categorize melody lines.
``Ear for electronic pictures'' is also developed. Moreover, in order to acquire such sound and electrographic techniques, the ability to play musical instruments such as the piano, which is required in conventional music education settings, is not required.

On the other hand, when the present invention is applied to an automatic music composer, the following
Benefits such as: In other words, for users, especially those with musical background, it is important for users to select their favorite chords from among several promising chord candidates in relation to the musical world they wish to express, even if the melody line is the same. Can be done. On the other hand, an automatic music composer uses a non-harmonic sound extracting section included in its motif feature parameter extracting means to identify or extract non-harmonic sounds included in a motif in the same manner as described above. Then, from this extraction result, harmonic feature parameters (for example, number of harmonic tones,
A pitch pattern of a dispersion chord with one non-harmonic tone is obtained. This information is passed to the melody generation means of the automatic composer, and a melody based on this information is generated. If the user does not like the resulting melody, they can enter the next most likely chord as the motif chord. This time, the extracted harmonic feature parameters will be different from those of 半, and these parameters will be reflected in the generated melody, so naturally the melody that is higher than IQ will also be different from that of 半. Become something. By following such a process, it is expected that the user's desired melody can be reproduced more effectively.

When the present invention is applied to an automatic music composition machine, it is possible to "classify" non-harmonic sounds depending on the configuration of the melody generation means of the composition machine.
There may be cases where there is no way to do so. However, in the embodiment described later, the melody generating means includes means for adding various non-harmonic sounds, and in relation to this type-specific non-harmonic sound adding means, the motif analyzing side also has the following: A means is provided for classifying and adding non-harmonic sounds of the motif. The logic may be similar to that described for the melody analyzer, and this combination of classification extraction and classification addition of non-harmonic sounds is effective for more finely controlling the generated melody.

[Example J Examples of the present invention will be described below.

This embodiment is an example in which the principle of this invention is applied to an automatic musical composition machine. The characteristic part of the embodiment related to this invention lies in the functional part of extracting motif characteristic parameters from a motif among the overall functions of the automatic musical composition machine. The views that best represent these parts are FIGS. 9-10, 11A, 11B, and a portion of FIG. 2.

The basic parts of the automatic music composer conform to the type of music composer disclosed in the above-mentioned patent application. For details on the basic parts, see Part 1.
2 to 49.

Overall configuration> The overall circuit configuration of the automatic music composer is shown in Figure 1.In Figure 1, l is an input device, 2 is a chord uI tone generation memory, 3 is a chord progression memory, 4 is a root note data memory, and 5 is a scale. , 6 is a motif memory, 7 is a parameter C memory, 8 is a passage identification data memory, 9 is a CPU, and 10 is a work memory.

11 is a parameter C memory, 12 is a learning data memory,
13 is a melody data memory, 14 is a monitor, 15 is a CRT, and 16 is a musical staff shilling.

17 is a musical tone forming circuit, 18 is a sound system, and 19 is an external storage device.

The motif memory 6 stores information on a motif (input melody) input from the input wave fil. Motif information consists of a string of pitch and length (total pitch) data. When automatically composing music, the CPU 9 extracts parameters characterizing the motif (motif characteristic parameters) from this motif information.

The chord progression memory 3 stores chord progression information expressed as a string of chord names.

The chord progression information may be inputted sequentially by the user specifying chords from the input wave 211, or may be inputted by the user in response to a rough specification (for example, specification of the music format).
U9 may automatically generate chord progressions. Automatic generation of chord progressions is possible, for example, by concatenating 11 chord patterns (frequently used chord patterns) or by concatenating allowed chords. For example, a Markov chain model can be used as the connection logic. However, it is not important to the invention whether the chord progression for the entire surface is specified directly by the user or automatically generated by the machine.

However, regarding the motif code, the code specified by the user is input from the input type ff1l, which is an important point. In other words, the chord progression memory 3 stores at least the chord progression of the melody.

Automatic linking of the motif chord and the chord or chord progression of the melody that follows the motif can be realized using the Markov chain model described above.

The chord structure note memory 2 stores various chord structure notes (chord member pitch data), and in this example, the chord structure is determined from the contents (chord name) of each address in the chord progression memory 3. A storage area for specific chord constituent sound data on the sound memory 2 is designated. During automatic composition, the CPU 9 advances the address in the chord progression memory 3 at each chord change timing (for example, every measure), and calculates the address in the chord constituent note memory 2 from the chord name that is the content. , reads each pitch data making up the chord.

The root note data memory 4 stores the root note data of the chord, and the ff scale weight data memory 5 stores weight data indicating the degree of effectiveness of each pitch that makes up the note scale. When automatically composing music, a scale is selected by an appropriate method, and the weight data of this scale is read out. The root note data memory 4 is used to shift the weight data of the read scale to the root note.

- Power, Parameter C The memory 7 stores data (parameter B) for controlling consistency and diversity in the flow of the melody.

In addition, in order to bring a higher level of hierarchy to the song, musician 1
A separate data memory 8 is used. This memory 7 and memory 8 have a user area, ie.

A read/write area is provided to store parameters specified by the user. During automatic composition, the CPU 9 uses parameter B, musical style identification data (data after decoding)
, create a parameter C (melody plan information) that depends on the motif feature parameters and the music progression section variable (eg, bar number). Parameter C has the property of controlling or characterizing the generated melody. The generated parameter C is stored in the parameter C memory 11.

The work memory 10 stores intermediate data (for example, melody data being processed) generated by the CPU 9 in the process of automatically composing music.

The melody data memory 13 stores melody data constituting a completed song.

The completed song can be outputted to the monitor 14, for example, and can be previewed through the musical tone forming circuit 17 and the sound system 18, if necessary. Additionally, a copy of the musical score can be obtained from the staff notation printer 16.

The user may wish to partially modify a piece of music through the monitor 14. In this embodiment, in such a case, the user may request modification through the CRT 15 and the input type 221. Modifications can be made in an interactive manner. The corrected data is stored in the learning data memory 12 as knowledge. For later automatic composition.

The CPU 9 uses this knowledge to generate a melody.

External storage 2119 is used as a backup copy of completed songs, learned knowledge, other copies, or as a resource for an alternative automatic composition program.

Automatic music composer 1> Next, the overall function of the 11-move music composer will be explained with reference to FIG.

As shown in the figure, the main *f#, a motif feature parameter extraction means FIO that evaluates and extracts its features from motif data, and information given by this means FIO (inclusively indicated by PA) a melody control information generating means F20 that generates information for controlling the melody (inclusively indicated by PC); and a melody control information generating means F20 that specifically generates a melody (single melody) according to the information PC given from this means F20. There is a generation execution means F30.

As will become clearer from the explanation that follows, the features of the present invention are 1.
The information sent to the means FIO includes chord information (chord data) assumed by the user for the motif, and within the means FIO, this chord information is utilized to calculate the harmonic characteristic parameters of the motif ( It has an internal structure that extracts the number of harmonic tones, the number of harmonic tones, the type of dispersive chords, etc.).

Chord information is input device in Figure 1! This is the information entered through.

On the other hand, the melody control information generation means F20 has the ability to associate and plan a melody based on the motif characteristic parameter PA, and the melody generation execution means F30 decodes this plan (expressed by the PC) and executes the melody according to the plan. It has rules for generating melodies. In a broad sense, 1 means F2
0 and F30 constitute a melody generating means.

On the other hand, from the aspect of inference, motif feature parameter extraction1
The means FIO and the melody generation execution means F30 basically have an inverse relationship, and the motif feature parameter extraction means FIO derives the four-character parameter PA, which is the essence, from a specific motif (input melody). In contrast, the melody generation execution means F30 has an inference ability to derive a specific melody from the essence of the melody feature parameter PC. The planning space of the melody control information generating means F20 is vast, and F
A. The degree to which IPc is reflected can be freely and variously controlled.

The parameter B and musician identification data shown in this section F20 may be data created by the user via the input device Ml. If the parameter B and musician identification data used change, the control parameter (pc) output by the melody control information generating means F20 also changes.

Generally, a song changes as it progresses. However, if you look at the limited time, certain characteristics are fixed.The automatic music composer takes this point into consideration.

For example, each function in the motif feature parameter extraction means FIO extracts FA using a certain interval as a unit. Similarly, the melody control information generation means F20 passes the PC value assigned to each section to the melody generation execution means F30, and the melody generation execution means F30 generates a melody for that section using the PC value. Of course, this does not mean that all PCs change in the same section (the same applies to FA).

Depending on the type of PC, the intervals in which values can be fixed are generally not equal, so in principle.

It is also possible to have different interval units for each type of PC and generate values in different intervals for each type. However, this complicates the process. Therefore, in the embodiment, an interval of class J4, which is far fewer than the types of PCs, is used, preferably an interval of a unit that can be common to all PCs (an interval of the greatest common divisor). The same applies to the motif feature parameter extraction 111 PIFIG.

In the configurations shown in FIGS. 9 to 49, as the most simplified example, each function FIO, F20, and F30 has a common section. "1 bar" is selected as this section. Of course, the present automatic music composer is not limited to this, and each aI function can have two or more different rK ranges. For example, the interval for the pitch sequence and the interval for the note length sequence do not always have to be the same, and one-move fiF
20 assigns a PC value to each measure, whereas chord progressions can also be made to progress in sections that do not have measures as units.

Let us now discuss the details of the main functions FIO, F20, and F30 in FIG.

In addition to the motif related to the user input, chord information (chord data) specified by the user for the motif is sent to the FIO sexy code as input information. The code of this motif is stored in the motif pitch pattern extraction means Fil inside the section FIO, especially the non-harmonic sound extraction means F.
Used in il-IA. That is, one means Fl
1-IA, the Uke infers that a motif note with the same note name as a constituent note of the chord taken is a harmonic tone, and a motif note that does not match the chord constituent notes is inferred to be a non-harmonic note. therefore,
The logic configuration of the non-harmonic sound extraction means Fll-IA is very simple, and once chord information and a motif are given, it is possible to uniquely distinguish between harmonic and non-harmonic sounds for each sound of the motif. . Nevertheless, if the user specifies different chord information, a different interpretation will be made, and it is up to one composer, J! ! It is related to the musical world that the composer or performer is trying to express, and it cannot be determined uniquely in general. It makes inferences based on R, and if the chord designation is changed, it is also reflected in the generated melody.

In section FIG, the non-harmonic sound classification means Fll-
IB analyzes the a column of the motif data based on the identification results of harmonic 2'f/non-harmonic sounds by means Fl 1-IA, and determines to which type of non-harmonic η each non-harmonic sound belongs. We are in the process of deciding what to do. In other words, the motif is interpreted harmonically at a more detailed level than the level of harmonic tones versus relative harmonic tones. The meaning of this is also related to the melody generation section (F20, F30) of the automatic music composer, and the melody generation section has various non-harmonic sound characteristic style parameters (included in the PC). There is a part (inside F20) in which non-harmonic sounds are generated, and a part (F31-2A) in which non-harmonic sounds are added before, after, or between harmonic sounds according to these parameters. Fine-grained control over the characteristic patterns of harmonic/non-harmonic sounds.

The dispersion chord pattern (pitch) extraction means Fl 1-2 extracts a pitch pattern obtained by removing non-harmonic sounds from the motif, that is, a dispersion chord J! ! This is the part that extracts LLI (described later).

On the other hand, the motif length (rhythm) pattern extraction means F12
is a function that extracts features related to the tfIi sequence of motifs. Feature mini-pattern extraction means F, which is a component thereof
12-1 extracts a characteristic small tone length sequence included in the motif, and extracts a dispersion chord (rhythm) pattern extraction means F12-
2 is.

The tone lengths of non-harmonic tones included in the motif are absorbed into the tone lengths of harmonic tones to form a tone length sequence of only harmonic tones.

Among the above elements, Fil-IB, Fll-2, F12
-2 operates in response to the result of the non-harmonic sound extraction means Fil-IA. These means are functions for extracting harmonic feature parameters in the configuration shown in FIG. More generally, the harmonic characteristic parameter is a parameter that characterizes the mode and style of the harmonic 7h tone and non-harmonic tone in a motif or melody line.

In the block of the melody control information generating means F20, there is shown a progress dependent parameter generating means F22 which generates a parameter depending on the section number indicated by the interval counter F23. These parameters include those that vary regularly as one song progresses, and these parameters are generated by the regularly varying parameter generating means F indicated by F22-1.
The random number generating means F24 acts on the parameter 1 generated by the zero means F22 generated by the random number generating means F22-1, and can introduce a random number or fluctuation controlled by the parameter into the parameter.

The above elements F22 and F23 are shown to show that parameters with the above-mentioned properties are included in the PC, and even in a non-arithmetic type (for example, a database of parameter C), the parameters C ( This is shown to clarify that it is possible to generate melody control information). Actually, in this embodiment, the parameter C is generated using a calculation type, and the part that executes this calculation is F.
It is a calculation means indicated by 21. The calculation means F21 calculates the motif characteristic parameter FA from the -r stage FIO, the information on the parameter B indicated by 11, the small t1:1 number indicated by the measure counter F23-1, and the musical style identification data from the musician identification data generation means F25. is received as input, and the parameter C is calculated using these inputs as variables. The musician a-specific data generating means F25 has a passage counter F25-1, which is used to select information regarding a specific passage from the musical style identification data (2). The information regarding the passage includes the type of passage (repetitive type, expansion type).The musical style identification data generating means F25 uses the minor mIR indicated by the measure counter F23-1.
Read the number, check the positional relationship between the measure number and the musical passage,
The decoded musical style identification data is passed to the operator stage F21, which decodes the relevant musical style identification data based on the test result. The role of the musical style identification data generating means F25 is to create a higher level of hierarchy in a piece of music.

As you can see from the explanation of I-L, is it about melody control? 41
There are various PCs output from the generation stage F20, and some PCs are constant over a relatively long period of time, and some PCs are constant over a relatively long period of time, and some
C fluctuates in a cycle suitable for a unit interval (here, one measure), and a certain PC is influenced by the musician and changes to a different value at a particular interval or point. This is how it goes. Value of parameter B, value of musical style identification data, operator stage F
Depending on the type of function used by 21, even in the same FA, various PCs may occur.

In other words, the parameters PC include local stylistic features of a song, stylistic features over a broader range, stylistic features that appear regularly, hierarchical styles in the flow of the melody, and the flow of the melody. It is a parametric expression of the structural features of non-harmonic sounds, the features of regular or quasi-regular changes in the flow of the melody, etc.

The melody generation execution means F30 has as its main elements a pitch sequence generation means F31 and a tone length sequence generation means F32. IIF high-row usury means F31 is chord progression data■
Current chord and melody control information generation means F from 3
Is it the PC itself from 20, or the random a generation means F31-4?
Dispersed chord generating means F31-1 generates a pitch sequence of a dispersive chord using a PC in which fluctuations are introduced. The non-harmonic sound adding means F31-2 adds a non-harmonic sound according to addition rules. Each non-harmonic sound classification adding means F31-2A is an example of the structure of the means F31-2. F31-3 is pitch control means, which has a function of controlling the use of melody note candidates obtained in the generation process of the means F31-1 or F31-2. That is, the used pitch control means F31-3 is the Nokoto scale generation means F31-3A.
generates data that weights each pitch of the scale,
The effective pitch testing means F31-3B is caused to test the validity of the candidate pitch based on the result. The melody note that passed the test is the means F31-1. Sent back to F31-2,
It is used here as the official melody note.

The ff Nagatsuki generation stage F32 includes an optimal combining means F32-1, an optimal dividing means F32-2, and a feature pattern incorporating thousand FIF32.
-3, and generates a tone length sequence according to the PC plan. Optimal connector stage F32-1 and optimal dividing means F32-2
In this embodiment, is used to form a tone length sequence of the unidispersion a-tone. The optimum combination means F32-1 is an initial tone length sequence (for example, FIO's dispersion chord (rhythm) pattern extraction means F1).
2-2), the note lengths are combined using the minimum number of combinations until the target number of notes of the dispersed chord (given by the PC) is reached. Similarly, optimal dividing means F32-2
divides the initial tone length sequence by the minimum number of divisions until the number of distributed chords PC is reached. In addition, one vehicle means F32-1.
F32-2 executes division and combination using a pulse scale. The pulse scale may be given by the means F20 as a type of PC (melody planning information), but in the specific example described later, the pulse scale is included in the division and combination rules of the five means. On the other hand, the feature pattern incorporating means F32-3 operates according to the related PC,
In this example of incorporating feature mini-patterns into the note length sequence of a melody, the feature mini-patterns are injected in the final process of melody generation.

The i1J1 means F33 is for controlling activation of each element of the melody generation execution means F30 and data transfer between elements. The melody data that is the execution result of the melody generation execution means F30 is stored in the melody data memory 13 (first
(Figure).

Details> Figures 3 to 49 show details regarding various data, extraction of motif characteristic parameter A, generation of control parameter C, and specific generation of melody. A detailed explanation will be omitted except for the characteristic parts of I11. Those skilled in the art will be able to easily understand the contents to an implementable level from clear descriptions such as flowcharts.

Let me briefly mention some of them to help you understand.

A list of main variables is shown in FIG. 3, and a list of parameters C is shown in FIG. The symbols for each parameter shown in these lists match the symbols used in the flowcharts from FIG. 9 to FIG. 49. In these flows, pitch data is assigned as shown in FIG. 5b.

Continuous integer values are given to consecutive pitches on the chromatic scale. Also, regarding the note length, one measure is 16
The unit is the length of the tone when divided into equal parts. Therefore 16
The pitch length of a diacritic note is 1", and the pitch length of an eighth note is 2" (note length data M shown in the motif data column in Figure 6).
(Please refer to Di and pitch data MRi).

Figure 8 shows the melody obtained when the motif data, chord composition note data, chord progression data, musical style identification data, scale weight data in Figure 6, and parameter C data in Figure 7 are input conditions. This is an example.

The entire process of generating a melody from a motif begins with the evaluation of the motif, that is, the extraction of motif characteristic parameters (PA).

After the FA is determined, the control parameters (PC) of the unit section (small wJ) of the melody are calculated, and after the PC is determined, the specific melody is generated.

The processing relationship for extracting the motif feature parameter A from the motif is shown in FIGS. 9 to 20.

Here, Fig. 9, Fig. 1θ, Fig. 11A, Fig. 11
Figure B shows the first classification of non-harmonic sound extraction that is related to the features of the present invention, and these will be explained in detail.

Left 1j So L Abdominal Voice FIG. 9 illustrates a flowchart for extracting incoherent tones. The basic logic is as follows. That is, among the motif notes, a note having an h name that does not match the constituent notes of the chord is considered a non-harmonic note. In other words, the logic is that all the constituent notes of the chord are harmonic tones.

To be more specific, in the section from 9-1 to 9-4, the pitch sequence (MDi) of the motif data in the motif memory 6 is
7- Area prepared on the memory IO (HD i
). In other words, all the pitch data of the motif data is assigned to HDi. By the way,
In this example, each note of the motif is expressed by a pair of pitch data and duration data. However, since there is no concept of pitch for rests, this is expressed by inserting a special value into the pitch data.

From 9-5 to 9-15, for the a note of i#il in the array (HDi), whether it is a harmonic tone or a non-harmonic tone is identified by comparing it with chord constituent notes.The identification results are as follows. ,
In the case of a "non-harmonic sound", a unique value (-20 in the figure) is stored in HDi. j indicates the number of chord constituent notes. For convenience of explanation, this example considers only the case where the chord has four independent voices or three independent voices including two voices with an octave difference. Furthermore, in order to comply with this, the chord aI & tone memory 3 (FIG. 1) has four pieces of data as constituent tone data of each chord.

In this flow, the pitch data of the chord constituent notes is KD
It is indicated by j. The subscript j indicates that the chord component note is at the height jJt from the bottom.

As shown in 9-7, when the starting motif note is a rest, it is not a non-harmonic note, and the process moves on to checking the next motif note. 9-8 and 9-9 are calculating the note names of the motif and chord constituent notes, respectively (eliminating the concept of octave numbers), and 9-10 is checking whether the note names match or do not match. If there is a match, the motif note of interest is a "harmonic note" and the process moves on to checking the next note. If there is a mismatch, use the following code structure, J
In order to proceed to the comparison with &9, j is incremented at 9-12, and steps 9-7 and subsequent steps are repeated.

In 9-11, j≧4 holds true when the motif note we are looking at is compared with all the constituent notes of the chord, but none of them match; in other words, it is a "non-harmonic note".
This is the case. Therefore, in 9-13, the identification value of the non-harmonic tone is substituted into tfD7-taHDi of the motif note. Nos. 9-14 are the total number of notes in the motif. Therefore, when I≧No holds true, the identification of whether or not all notes of the motif are non-harmonic notes has been completed. Therefore, the non-harmonic sound extraction process ends.

Classification of non-harmonic sounds is performed by pattern analysis of the a sequence of motifs based on the extraction results of non-harmonic sounds as described above.

There are several possible classification logics. FIG. 1O is a flow example based on the first classification logic, and FIGS. 11A and 118 are flow examples based on the second and third classification logic, respectively.

This will be explained starting from Fig. 1θ. In step 10-1, the motif note number l is initialized to 'l' and becomes primary.

Increment i until HDi≧O (10-
1, I 0-2), I (D i≧0 indicates that the i-th note is a harmonic tone. As inferred from the increment of five in 10-4, from the beginning of the motif Check to distinguish the non-harmonic sounds up to the first harmonic sound from others.0 In the flow shown in the figure, the HDi of this non-harmonic sound remains minus 20 (Sound, HDi = -
20).

1O-5 is a check to see if the starting note is a chord tone or a rest, and if it is true, the HD
There is no need to change j, so proceed to the next note processing.If not established, 2. The note is a non-harmonious f().In 10-6, it is checked whether the note one position below the starting note and the note one note after are at the same height. is H
Execute Di=-30 (embroidery sound, tO-a). In other words, we conclude that the first note we are focusing on is an embroidery note. If this is not the case, it is assumed that the pitch does not match, and HDi-40 (passing tone, 1O-7) is executed. In other words, we conclude that it is a passing sound.

If it is a non-harmonic note and it is the last note, check 1.
1=No holds true for 0-9.

This non-harmonic tone is treated as a consensus and HDi--60 is executed (
10-10),,,10-11? When i<No, the non-harmonic sound classification for all the notes has not been completed, so the process returns to i=L+1(10-4) and moves on to the analysis of the first-order notes.

In addition, the name given to the four types of non-harmonic sound identification values of HDi written as a result of the processing in FIG. 10, that is, "倚a"
, "passing sound", "embroidery blindness".

"I n J" is for convenience of explanation. This also applies to Figures 11A and 118.

:jThe classification of non-harmonic sounds shown in the SllA diagram is almost the same as in Figure 1θ. The difference is that the processing shown in IIA-6 and 11A-7 is added.・The calculation shown in 11A-6,
That is, a+ = (NDt-+ -MDi)
Di-N[l+-+) is a multiplication of pitch differences formed between three consecutive tones.

Therefore, when al is positive, the pitches of the three tones are monotonically changing (rising or falling). Whether or not this holds true is checked with IIA-7, and if it holds, HDi
Put minus 40 in.

It is concluded that it is a passing sound.

FIG. 11B is a flowchart of non-harmonic sound classification according to the third logic. As can be seen from the comparison with the non-harmonic sound classification shown in FIG. 10, (a) the knowledge (check items) regarding transition sounds has been increased, and (b) the conditions for concluding that they are strident sounds have been slightly changed.

In other words, the three tones (MDt+, MDi, NDt-+) including the end of a certain non-harmonic tone (HDi) have monotonous pitch changes (see 11B-6, IIB-8, IIB-9). ), and the pitch difference from the previous note is a semitone or completely blind (
IIB-10), the non-harmonic tone (HDi
) is concluded to be a passing sound (HDi=-40). When this condition for transitional sounds does not hold true, the non-harmonic sound is concluded to be a soaring sound. To explain in more detail, non-phase 7i1 that does not meet any of the embroidery sound conditions shown in 11B-6, the transition sound conditions from 11B-8 to IIB-11, and the appropriate sound conditions shown in IIB-12. It is concluded that R is a wheezing sound (HDi exits with minus 20).

The explanation will move on to FIG. 12 and subsequent figures.

In the rhythm evaluation shown in Figure 12, the a-length ratio is 3:1, 1:3.
We are investigating how much each mini-pattern of 1:1.2:1:1.1:2:1.1:1:2 is included in the motif that is the fff' valence target, and the results are compared to each pattern. The counter HR1 is set to HI3.

The parameter extraction shown in FIG. 13 is a process that is performed after the first classification of motif tones and P sounds extracted as described above, and various parameters A (FA) are extracted here.

The details of 13-1 are illustrated in FIG.

In this flow, we are looking for a joint wall for the dispersed chords of the motif. For example, when (LLi) = (3, 2, 110, 1), the data of the joint wall is set in the array (LLi), it means the following. The first u 3 s means that the first harmonic sound in the motif is the third highest from the bottom among all the harmonic sounds included in the motif.The first order "2" means that the harmonic sound that appears at 2mUJ is , it is the second highest height from the bottom among all the harmonic tones of the motif, and so on.

However, '0' indicates that the note is a rest (see example N in Figure 14).

The details of 13-2 are illustrated in FIG.

In the flow shown in FIG. 15, the motif tone length pattern is converted into a tone length pattern of only harmonic sounds (motif dispersed chord pattern). In other words, the length of the non-harmonic sound included in the motif is absorbed into the harmonic sound. FIG. 16 shows a pulse scale inherent in the flow shown in FIG. 15, and this pulse scale is used to obtain a motif dispersed chord pattern.

The conversion from the original note length pattern of a motif to a dispersed chord pattern basically follows the following principle. In other words, compare the weight on the pulse scale of the current note (non-harmonic note) with the weight on the pulse scale of the next note, and if the starting position 21SUM of the first note is heavier, then the a-length MRI of the current note is compared to that weight. If the starting position 21SUMH of the current note is heavier, its note length MRi is absorbed by the next chord RHN-1. The motif dispersed chord (tone length) pattern that is the execution result is set in the array (RHi].

FIG. 17 is an example of a detailed flow of steps 13-3 and 13-4. Through the above process of extracting and classifying non-harmonic sounds, the pitch sequence of the motif is converted into (HDi), and each HDi (i indicates the first note) has a pitch sequence of (HDi).

If the sound is a harmonic sound, the pitch data of mo is included.
If it is a non-harmonic sound, Zhou Youdi (a) indicates the type of non-harmonic sound.
(separate value) is included. In the flow for extracting the number of inharmonious tones and harmonic tones in Fig. 17, after initializing the parameter PAlj (
- column), the value of HDi is checked, and depending on whether the condition is met, each nonharmonic ef) force '>75' (PAS, 2,
PA2.2. PA4.4, PA3.3) are running. The counter PA1.3 stores the number of harmonic tones.

Figure 18 is an example of the detailed flow of 13-5 and 13-6, where smoothness can be determined by analyzing the joint wall (LLi) of the dispersed phase a (pattern obtained by the process in Figure 14). Parameter FA1 which is homophone progression with parameter FA1.2
．． I'm looking for 6.

FIG. 19 is an example of the detailed flow of step 13-8. The above rhythm evaluation (Figure 12) ensures that the number of various mini-rhythm patterns that can be included in a motif is (HRi), and in extracting this characteristic rhythm parameter, which mini-pattern is the dominant pattern. 0 determining whether there is
For convenience, in the flow of Fig. 19, only the number HR+ of 3 to l tone length patterns and the number HR3 of l#l tone length patterns are examined, and when there are many 3 to l tone length patterns, PAh
, 3″′ to +, otherwise PAi, “0” to +
is included.

Figure 20 is an example of the detailed flow of 13-7, in which the minimum rf length is found in the motif note length sequence (MRi), and P
It is set to A3.3.

As can be seen from the above detailed explanation, the motif feature parameter extraction means FIO (FIG. 2) is.

The motif data sent from outside the automatic composer is evaluated and analyzed, and the parameters collectively shown in FIG. 2 are extracted. The extraction result is used for PC generation and melody generation. For example, the pitch-type parameter (LLi) of the motif's harmonic tone is used as the initial pitch-type parameter of the melody.

The tone length pattern (RHi) of the harmonic tone of the motif can be used as initial information of the tone length sequence of the melody. Other PA
is used, for example, as a DC component in PC calculations.

The processing executed by the parameter calculation means F21 in FIG. 2 is shown in FIG. 21. FIG. 25 is an example of a flow when a melody is generated continuously.

25-9 is 21-4.21-5.21-6 in Figure 21
processing is executed. 21-1.21-2 in Figure 21.
21-3 is a part that reads out the input parameters FA, PR, and SR to be used all at once. Regarding the decoding of musical style identification data performed at 21-4 in FIG. 21, see FIG.
Please refer to FIGS. 23 and 24. In this example, the peak (SBP+~
5BP3) and hold data (S
BH) is being decoded.

At 25-10 in FIG. 25, a melody for one measure is specifically generated. The other parts in Figure 25 are for the transfer of motif data and melody data.

This includes incrementing the number of measures to be generated.

The specific melody generation is as shown in Figure 8.
First, the first step is to generate a melody consisting only of harmonic sounds (dispersed chords).
The addition of each non-harmonic note, and finally the selective injection of rests and characteristic mini-patterns.

The overall flow of the generation of dispersive chords is shown in FIG.

Reading of chord constituent sounds in this flow 26-1.
Change of scale weight 26-2. Calculation of number of suitable turns 26-4. The details of the inversion 26-5 of the chord constituent notes are shown in the 28th section.
v4; Figure 30A, Figure 30B, Figure 31: Figure 32,
This is shown in FIG. 26-5 is completed, the weight data of the scale to be used (data with weights assigned to each note on the scale), the chord in progress and its inversion,
In other words, the pitch of each harmonic tone used to generate dispersed chords is determined. The scale weight data is used to check the validity of melody note candidates raised during the melody generation process, and if they are valid, they are officially adopted as melody notes.

From 26-6 to 26-42, the simplified flow is shown in FIG. 35, and the pitch sequence of the one-dispersion chord is determined. The details of the determination from the note before 26-15 shown in this figure are shown in Figure 34, which is the process of making the harmonic note with the pitch closest to the last note of the Moe measure the first melody note of the current measure. .

When proceeding from 26-11 through 26-12 to 26-29, it is within the range where type A is maintained. During the holding range,
A mold (26-31) that is the same as or opposite to that of the previous measure is created.0 For example, at the beginning of the second measure shown in (A) Dispersed chord in Figure 8, the mold of the first measure is in the ascending line. (LL+=1.
LL2 = 2° LL3 = 3, LL4 = 4), the placement-related parameters from the PC calculation unit for the second measure are:
The sustain range is from the first to the fourth note (PCl,
a=0. PCl,3 = 4), and PCl,13 instructs to reverse the mold, so the downward direction (LL+
=4. LL2=3. LL3=2, LL4=1).

For a measure such as a motif measure in which a mold is not maintained, sound pattern formation is performed by randomization starting from 26-16. Parameters C are also included in 26-16 to 26-28, and these parameters are used to limit randomization.

Validity tests for pitch were conducted at 26-35 and 2.6-36.

26-44 is the determination of the tone length pattern of the dispersed chord. When the number of chord tones in the bar to be generated is equal to the number of chords I'f in the motif, the initial pattern (for example, the dispersed chord pattern of the motif) becomes the nri pattern of the dispersed chords.

When the number of harmonic tones is greater or less than the number of harmonic tones included in the initial pattern, the pulse scale is used to perform min;1 or combination. For details of 26-44, please refer to FIGS. 36 to 41.

The basic logic of division is the minimum number of divisions,
When dividing by one, the pulse scale is ToT+TtT3TsTsTiTITsTv〒+@T+
+T+2T+zTuT+5 (here, To NT's is the pulse point, To is the beginning of the measure, Ts is tjS
Using the pulse scale (intrinsed in the rule) indicated by the timing of the third beat (the number below is the corresponding weight), cross the heaviest pulse point in the current note length pattern. This means that the note being played is divided using the pulse point as a dividing line. The basic logic of combining is that the combination is performed with the minimum number of combinations until the target number of notes is reached, and that in one combination, the note that starts with the lightest weight in the current note length sequence is connected to the previous note. This is a method of combining.

Addition of overtones (Figure 42)? Jt, PCfi[fl,
l” Weight of overtones enhanced PO2, 2, addition position PC
2, 3, scale weight PO 2, 1, ascending line, descending line PC 2, 4
, the tone length limit PCz, 1, etc., and selectively add overtones according to internal strangulation addition logic. Regarding the tone length, the - part is removed from the tone length of the harmonic tone that follows the overtone to determine the tone length of the overtone.

The addition of transitional sounds (No. 431i4) is basically a transitional sound that transitions between harmony sounds, so it is a part that looks at the conditions related to these.

The part that decodes the plan of parameter C from the PC calculation unit (
Mainly the first column in Figure 43), the part that searches and tests whether there is a sound that can be used as a transition note between chord tones (mainly the second column), and the a-high conversion of the found transition note (pitch). It consists of a section (mainly the third column) that performs g-length conversion (incorporation into a tone pattern) and g-length conversion (incorporation into a note length pattern).

Addition of embroidery sound (Fig. 44) is only performed when the preceding and succeeding chord sounds are at the same pitch (see first column), and there are various other restrictions due to parameter C. For example, PCl
, 4 has the meaning of prohibiting the addition of embroidery sound when it is "θ"', when it is "l" it can be added once, when it is "2" it cannot be added continuously, and '3' In the case of , it is intended that it can be added without limit.This plan is deciphered in 44-9 to 44-14.44-15.

In the m3 column ■, the sweat height of an effective embroidery sound is determined by searching and incorporated into the sound length pattern (MED) (44-24
~44-30), a part of the note length is received from the adjacent chord and incorporated into the pitch pattern (MER). Note that 44-is to 44-20 in the second column are array shifts in preparation for incorporating the data of the stridor into the tone length pattern and pitch pattern.

Similarly, in the case of addition of ``a'' (Fig. 45), there is a part that decodes each parameter C, and a part that excludes or inspects illegal sound additions (for example, 45-2.45-3 and 45-4 to 45-18).
), 45-17, it has been determined that there is no problem with adding appropriate sound.

45-17 to 45-23 are provided for the differential processing shown at the lower right of F. - In a row of incense stones, we find the pitch that can be the appropriate pitch between the previous note and the following note and incorporate it into the pitch pattern. Note that in the case of adding appropriate notes, the number of notes does not increase; that is, adding appropriate notes is a pitch change of the last note of the bar.

Rests (Figure 46) are often added at the end, such as at the last measure of a passage.

Characteristic rhythm generation (FIG. 47) involves selectively incorporating characteristic mini-patterns into note length patterns according to the value of parameter C. In this example, characteristic rhythm generation is the final process of melody generation, and characteristic mini-patterns are created within the range allowed by the PC while avoiding strange changes to the note length patterns that have been created up to that point. , here, a pattern with a blind length ratio of 3 to 1 is incorporated into the tone length pattern. 47-9 to 47-12 are checking to see if there is any one of two consecutive notes that can be converted into a 3-to-1 pattern.

When the test passes, conversion is executed in the third column n.

The modified learning shown in FIG. 48 and the parameter change by learning shown in FIG. 49 are related to the learning ability of using the user's preferred parameter C in each measure selected by the user. In corrective learning (Figure 48), after the song is completed, the user indirectly specifies the parameters of the measure that the user does not like through the monitor 14, and the learning function learns the type and value of the specified parameter of the specified measure. memory 12
Put it in. As shown in FIG. 13, the parameter change due to learning is performed during parameter calculation, and the parameter C selected by the user is used with priority over the parameter C internally generated by the music composer.

Summary and Application to Melody Analyzer> From the detailed explanation above, the various features of the automatic music composer in this embodiment are clear. In particular, it is worth noting that the chord (chord) information intended by the user for the motif is used to extract the harmonic feature parameters from the motif line. This is true in the example.

This makes it easy to divide the function of motif non-harmonic sound extraction and classification into two parts. In other words, in general, in the harmonic evaluation of a motif or melody line, distinguishing between non-harmonic and harmonic sounds is not necessarily separate from identifying which sound is which type of non-harmonic sound. Made! Rather, if we think of a wide variety of melody lines, we can say that the two are closely related. Here, in this invention, non-harmonic sounds are directly and rationally separated from motifs or melody lines by using chord information. From another aspect, this invention is based on the idea that the harmony or chord appropriate to a certain melody line is generally not uniquely determined in relation to the performer's musical space. Allowing the user to freely designate chords makes it easier to approach the musical expression (song) that the user intends and expects in the area of application to automatic music composers. In other words, it contributes to expanding the selection range for users.

Embodiment L is an application of the present invention to the field of automatic composing machines, but it is clear that the invention can also be applied to melody analyzers that automatically perform harmonic evaluation of melody lines. Through the use of the melody analyzer, students are expected to be able to effectively learn to categorize melody lines, and also to acquire the skill of adding chords to melodies in a relatively short period of time.

[Effect of the Invention 1 As described in detail, the melody analyzer according to the present invention inputs not only melody data but also harmonic or chord data latent in the melody as input data to the non-harmonic sound extraction means. is passed, and the non-harmonic sound extraction means uses this code as a clue to identify which notes in the melody are harmonic sounds and which sounds are non-harmonic sounds, and the identification result is sent to the non-harmonic sound classification means. Based on this sent information, the non-harmonic sound classification means analyzes the sound sequence of the melody and estimates which sound is which type of non-harmonic sound. . Therefore, the learner is 1
This melody analyzer can be used as an assistant, and through its use, you can deepen your knowledge and ability to categorize and understand melody lines, and also improve your ability to envision chords or harmonies that harmonize well with each melody line. do.

Furthermore, the automatic composing machine according to the present invention utilizes the principle of the above-mentioned melody analyzer for extracting non-harmonic sounds of motifs, and this is a major feature. Then, the melody generation section of the automatic music composition machine generates a melody in which harmonic sounds and non-harmonic sounds are mixed, depending on the extraction result by the non-harmonic sound extraction section. Therefore, the resulting melody line changes depending on which chord the user specifies. Exchange? t, the automatic composer can provide the user with several preferred melody line candidates. In particular, for users who can freely create chords, we provide an environment where they can compose songs efficiently and efficiently. In terms of configuration, using chords as input information has the advantage of simplifying the non-harmonic sound extraction section and classification section.

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram of an automatic musical composition machine according to an embodiment of the present invention, FIG. 2 is a functional block diagram of the automatic musical composition machine, and FIG. 3 is a diagram showing a list of main variables used in the automatic musical composition machine. Figure 4 is a diagram showing a list of parameters C used in the automatic music composer.
FIG. 5 is a diagram showing an example of pitch data, FIG. 6 is a diagram showing an example of input data for explaining the operation, FIG. 7 is a diagram showing an example of the value of parameter C, and FIG. FIG. 8 is a diagram showing the melody generation results for each process for the data shown in FIG. Fig. 9 is a flowchart for extracting non-harmonic sounds, Fig. 1O is a flowchart for classifying non-harmonic sounds, Fig. 11A is a flowchart for classifying non-harmonic sounds using another logic, and 11th BR is a non-harmonic sound using another logic. Flowchart of classification, Figure 12 is a flowchart of rhythm evaluation, Figure 13 is a general flowchart of parameter extraction, Figure 14 is a flowchart for extracting parameters of chord type, Figure 15 is length of chord tone. A flowchart for extracting a pattern, Fig. 16 is a diagram for explaining the pulse scale used in the flow of Fig. 15, Fig. 1
Figure 7 is a flowchart for extracting the number of each non-harmonic tone and the number of harmonic tones; Figure 18 is a flowchart for extracting smoothness and homophone progression parameters; Figure 19 is a flowchart for extracting characteristic rhythm parameters; Figure 20 is a flowchart for extracting the minimum note length, Figure 21 is a schematic flowchart for calculating parameter C, Figure 22 is a diagram showing the data format of musical style identification data memory, and Figure 23 is a diagram for reading and decoding musician data. Flowchart, 2nd
FIG. 4 is a diagram showing an example of decoding, FIG. 25 is a flow chart of melody generation, FIG. 26 is a flow chart of dispersed chord generation, and FIG. 27 is a chord constituent note memory. A diagram showing examples of chord progression memory and root note memory data,
FIG. 28 is a flowchart for reading out chord constituent tones.
Fig. 29 is a diagram showing an example of scale weight data, Fig. 30A is a flowchart of changing the scale weight (1), Fig. 301 is a flowchart of reading scale weight data, and Fig. 3
Figure 1 is a flowchart of scale weight change (2), No. 32
The figure is a flowchart for calculating the optimum number of inversions, Fig. 33 is a flowchart for inversions of chord constituent notes, and Fig. 34 is a MED
Flowchart for determining the + (first note of measure) from the previous note, Figure 35 is a simplified flowchart of the important part of generating the pitch of a dispersed chord, Figure 36 is a flowchart for determining the note length sequence of a dispersed chord, Figure 37 38I54 is a flowchart of optimal tone length division processing; FIG. 39 is a detailed flowchart of the check in FIG. 38; FIG. 40 is a detailed flowchart of shift in FIG. 38; The figures are a detailed flowchart of the execution in Fig. 38, Fig. 42 is a flowchart for adding a stridor, Fig. 43 is a flowchart for adding a passing sound, and Fig. 44 is a flowchart for adding a passing sound.
Figure 45 is a flowchart for adding embroidery sounds, Figure 45 is a flowchart for adding appropriate sounds, Figure 46 is a flowchart for adding rests (presses), Figure 47 is a flowchart for generating characteristic rhythms, Figure 48 is a flowchart for corrective learning, Fourth
FIG. 9 is a flowchart of changing parameters by learning. l...Input device, 2...Chord composition note memory, 3...Chord progression memory, 9...
...CPU, 13...melody data memory,
Flo... Motif feature parameter extraction means,
F20... Melody control information generation means (control parameter generation means), F30... Melody generation execution means, Fil-IA... Non-harmonic sound extraction means, Fll-IB.・・・・・・Method for classifying harmonic sounds. Ku melody - IvI 4 N 1 Suga list now fishing 1 ridge 11 arta 5BN= 2 5B+ =
IQSB2-51 =-(ABsu2 Iku Saishi) Figure 13/Y U×-Tao Yuko012 3 4 5 6 7 B
9 10 11 12 13 14
15 Fig. 16 (-1 momme 5 L column sum 1; Yamamoto j; / 轡Bissu-ru) <13-3.4 details) Namura way J side n51 go 1 Mahmeh γ (1B-5,
6##) tv-7 meter 3'fl'4 A1 hosonaikigo1艮wachoA2\L't'ikennu1; Kawarama! 0 Tatsu 1 No. 1 B --- (9-12 + hiFF) No. A- (IIX fee to 13) Fig. 22 Gravel style!
) 37119 neck @ 7525 luster 25no 7525
Father 252,575 ~ 11 Korei 9
■ 25 75 25 Chi 25 (fund) (12~2
3rd year old] t5); tjlE'1. N7525502575
75 25 7525 50 2575 (2 Jinmanbanshichi) ξ26-2 119ノ (26-2r>-vIri(26-44/)-1FII) (38-8"!#j surrounding) Medium %1vL squirrel 'Mu pressure ν to 1

Claims (3)

[Claims] (1) A melody input means for inputting a melody, a code input means for inputting a melody code, and a non-harmonic sound extraction means for identifying or extracting a non-harmonic sound included in the melody based on the input melody code. and non-harmonic sound classification means for classifying each non-harmonic sound from the non-harmonic sound information identified by the non-harmonic sound extraction means and the input melody sound sequence. Melody analyzer. (2) a motif input means for inputting a motif; and a motif feature parameter extraction means for extracting feature parameters characterizing the motif, including harmonic feature parameters regarding non-harmonic sounds and harmonic sounds included in the input motif. , chord progression storage means for storing at least chord progression information of the melody to be generated; and chord progression storage means for storing at least chord progression information of the melody to be generated; melody generation means for generating a melody including non-harmonic sounds; and a code input means for inputting at least a code of a motif; 1. An automatic music composition machine characterized by comprising a non-harmonic sound extraction means for identifying or extracting non-harmonic sounds included in a motif based on the motif. (3) In the automatic music composition machine according to claim 2, the motif feature parameter extraction means further includes non-harmonic sound information identified by the non-harmonic sound extraction means;
An automatic music composition machine characterized by comprising a non-harmonic sound classification means for classifying each non-harmonic sound based on the sound sequence of a motif.