JP2541451B2 - Rhythm generator and rhythm analyzer - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a music apparatus, and more particularly to a rhythm generating apparatus that automatically composes a rhythm and a rhythm analyzing apparatus that analyzes a given rhythm musically.

[0002]

2. Description of the Related Art In the field of electronic musical instruments, those which automatically play a rhythm have been known. This type of device stores rhythm data for each rhythm style inside,
During operation (at the time of performance), the rhythm data corresponding to the selected rhythm style is read out and reproduced as a rhythm sound string. It also allows the user to enter rhythm data,
There is also known a device that stores input rhythm data and reproduces it when playing. However, neither device clearly lacks the ability to compose the rhythm itself, and the rhythm to play automatically is either stored in the device as data or stored in memory via user input. Only.

[0003]

Therefore, it is desirable to provide a rhythm generating device that can automatically compose a rhythm. Such a rhythm generation device can be used as a music composition support tool or a function of an automatic composer. Further, it is desired to provide a rhythm analysis device capable of performing musical analysis of rhythm.

[0004]

According to the present invention
For example, the original rhythm data that expresses the original rhythm in the sequence of note lengths
Original rhythm giving means for giving data and rhythm control data
Rhythm control data generating means for generating
A note that differs from the note length sequence of the above original note sequence based on your data
A rhythm that is a modification of the original rhythm by generating a long sequence
And a rhythm transformation means for generating
The data generating means uses the rhythm control data as music data.
Weighted for each pulse point representing each music time on the axis
Single pulse scale (hence multiple pulse scales
Of multiple pulse points with different weights of
A pulse scale generating means,
Is the original note sequence based on the pulse scale weights
Split one original note in two notes, or above
Combines two original notes in the original note string into one note
A note having a note dividing / combining means
A device is provided.

According to this structure, the rhythm transforming means transforms the original rhythm based on the rhythm control data from the rhythm control data generating means to generate a rhythm. Therefore,
This rhythm generator automatically composes the rhythm (compose)
can do.

In particular, the rhythm control data generating means uses pulse scales (FIG. 16; Ti in FIG. 53, FIG. 67) in which weights are applied to the respective pulse points in order to combine and divide the notes. It has a pulse scale generating means for generating as a parameter). Further, the rhythm transformation means has note division / combination means (F32-1, F32-2 in FIG. 2; FIG. 53, FIG. 55) for selectively combining or dividing notes based on the weight of the pulse scale. . In one configuration example, the note division / combination means is
When divided, the minimum or maximum on the above pulse scale
The original note being sounded at the pulse point of
Split at the small or maximum pulse points and combine notes
The minimum or maximum weight on the above pulse scale.
The original note whose pulse point is the sound generation start time
Combine with the original note immediately before the original note of.

Further, according to the present invention, the above-mentioned rhythm generation
A device that can be used in place of or in combination with the device.
The original rhythm is a note length sequence of the original note sequence.
Original rhythm adding means for adding the expressed original rhythm data
And the original rhythm data is processed into the original note
By generating a tone length sequence different from the tone length sequence of
Rhythm generator that automatically generates a rhythm different from the original rhythm
And the rhythm generation means includes
Data is processed, and from the original rhythm data,
A note length sequence pattern that appears frequently in note sequences (hereinafter
Characteristic rhythm pattern that extracts a rhythm pattern)
The above-mentioned characteristic rhythm pattern is applied to the generated rhythm
And a means for incorporating a characteristic rhythm pattern,
There is provided a rhythm generation device characterized by: This configuration
According to the feature rhythm pattern extraction means (F12 in FIG.
-1, characteristic of the original rhythm extracted by Fig. 11)
Rhythm generation based on various note length patterns (HRi in Fig. 4)
Characteristic rhythm pattern incorporating means (F32-3 in FIG. 2, FIG.
Since it was done according to 7), the characteristics of the original rhythm were reflected.
Can generate rhythm.

Further , according to the present invention, the rhythm is a musical note sequence.
Rhythm assignment that adds rhythm data expressed by a sequence of note lengths
Means and from the rhythm data to the note sequence with high frequency
Use the note length sequence pattern that appears to identify the characteristic rhythm pattern of the above rhythm.
Characteristic rhythm pattern extraction means for extracting as a turn,
There is provided a rhythm analysis device characterized by having
It

The technology of the present rhythm analysis device can be used as an analysis unit of a rhythm generation device. That is, the rhythm can be generated by incorporating the extracted characteristic rhythm pattern into the rhythm.

Further , according to the present invention, the rhythm is changed to a musical note sequence.
Rhythm assignment that adds rhythm data expressed by a sequence of note lengths
Means and each pulse point representing each music time on the music time axis
Pulse scale weighted by
Scale is from multiple pulse points with multiple different weights
And the weight pattern formed by the weights of these multiple pulse points.
The turn is called the pulse scale weighting pattern below)
Multiple types of pulses that are associated with each rhythm style
Scale generating means and the above notes that express the above rhythm
Sound generation pattern of string
Pattern formed by the sound generation start timing of several notes)
And a plurality of weight patterns of each pulse scale described above.
Pattern matching using the pulse point weight data
Rhythm evaluation means
An analytical device is provided. With the rhythm analysis device with this configuration
If each pulse scale is
Rhythm style such as B
Each pulse scale and assigned using Kale's weight data.
Pattern matching with the sound generation pattern of the rhythm
The rhythm that is the music background of the assigned rhythm.
You can analyze the style.

A rhythm analysis device using a pulse scale can also be developed into a rhythm generation device. For example, the rhythm assigning means is configured as a rhythm generating means (61-1 in FIG. 61) that randomly generates rhythms, and the rhythm generating means until the evaluation value evaluated by the rhythm evaluating means falls within a desired range. A desired rhythm can be obtained by repeating the trial of generation of the rhythm.

That is, this rhythm generating device produces a rhythm.
Randomly generate rhythm data represented by a note length sequence
Rhythm generating means and each music time on the music time axis
The pulse scale (weighted for each pulse point
Therefore, the pulse scale has multiple weights with different weights.
It consists of pulse points, and the weights of these multiple pulse points are formed.
The weight pattern of the pulse scale is
Pulse scale generating means for generating
Sound generation pattern of note length sequence of the above note sequence that expresses rhythm
And the weighting pattern of the pulse scale
Pattern matching using weight data of pulse points
The rhythm evaluation means and the pattern map of the rhythm evaluation means described above.
If the result of the ching is not within the specified range, the above rhythm is repeated.
Generate new rhythm data in the generation means,
For rhythm data, the pattern
And the rhythm evaluation method described above.
The pattern matching result of the stage is within the above specified range
When, the rhythm data at that time is stored as a generated rhythm
And a means for performing the rhythm generation device.
Composed of.

In the following embodiments, the rhythm generation device and the rhythm analysis device are described as one function of the automatic composer for automatically generating a melody.

[0014]

Embodiments of the present invention will be described below with reference to the drawings. Of the following headings, in particular, the part relating to the present invention is described in “rhythm evaluation”, “characteristic rhythm parameter”, “generation of distributed chord tone length sequence”, “characteristic rhythm generation”, and “rhythm relation”. ing. <Overall Configuration> FIG. 1 shows the overall circuit configuration of the automatic composer according to the present embodiment. In the figure, 1 is an input device, 2 is a chord component note memory, 3 is a chord progression memory, 4 is a root note data memory, 5 is a scale weight data memory, 6 is a motif memory, 7 is a parameter B memory, and 8 is a phrase. Identification data memory, 9 CPU, 10 work memory, 11 parameter C memory, 12 learning data memory, 13 melody data memory, 14 monitor, 15 CRT, 16 staff printer, 17 musical tone forming circuit, Reference numeral 18 is a sound system, and 19 is an external storage device.

The motif memory 4 is for storing information on a motif (input melody) input from the input device 1. The motif information is composed of a string of data of pitch and pitch (pitch). At the time of automatic composition, the CPU 9 extracts parameters (motif feature parameters) that characterize the motif from the motif information.
The chord progression memory 3 stores chord progression information represented by a string of chord names. The chord progression information may be input by the user sequentially specifying chords from the input device 1, or in response to a rough designation (for example, designation of a music format), the CPU 9 automatically generates chord progressions. You may do it. Automatic generation of chord progressions is possible, for example, by concatenating basic chord patterns (chord patterns that are frequently used) or concatenating chords that are allowed, and as the concatenation logic, for example, a Markov chain model can be used. . However, it is not important to the invention whether the chord progression is directly specified by the user or automatically generated by the machine.

The chord constituent note memory 2 stores the constituent notes of various chords (pitch data of chord members). In the case of this example, the contents (chord name) of each address of the chord progression memory 3 are used. A storage area of specific chord constituent sound data on the chord constituent sound memory 2 is designated. The CPU 9 advances the address of the chord progression memory 3 for each chord change timing (for example, for each bar) during the automatic tune, and calculates the address on the chord constituent note memory 2 from the chord name, which is the content of the chord progression memory 3. Each pitch data that composes the chord is read.

The root note data of the chord is stored in the root note data memory 4, and the weight data indicating the degree of effectiveness of each pitch forming the note (note scale) is stored in the note weight data memory 5. Weighted note scale data is stored (memory for storing a set of note scale data). At the time of automatic composition, a scale is selected by an appropriate method, and weight data of this scale is read out. The root note data memory 4 is used for root note shifting of the read scale weight data.

On the other hand, the parameter B memory stores data (parameter B) for controlling consistency and diversity in the flow of the melody. Further, the musical expression identification data memory 8 is used in order to give the music a higher hierarchy. At the time of automatic composition, the CPU 9 depends on the parameter B, the music identification data (data after decoding), the above-mentioned motif feature parameter, and the music progression section variable (for example, the bar number) C (melody plan information).
Create The parameter C has a property of controlling or characterizing the generated melody. The generated parameter C is stored in the parameter C memory.

The work memory 7 stores intermediate data (for example, melody data being processed) generated by the CPU 9 in the process of automatically performing music. The melody data memory 13 stores melody data forming a completed song. The completed song can be output to the monitor 14 as needed. For example, it can be auditioned through the tone forming circuit 17 and the sound system 18. Further, a copy of the score can be obtained from the staff notation printer 16.

Through the monitor 14, the user may desire to partially correct the song. In the present embodiment, in such a case, the user can request the correction via the CRT 15 and the input device 1, and the correction is executed in an interactive form. The corrected data is stored as knowledge in the learning data memory 12. At the time of later automatic music, the CPU 9 uses this knowledge to generate a melody. The external storage device 19 is used as a backup copy of the completed song, learned knowledge, other copies, or as a resource for an alternative automatic song program.

<Automatic music composition function> Next, the overall function of the automatic music composition apparatus according to this embodiment will be described with reference to FIG. As shown in the figure, the main function is to control the melody from a motif feature parameter extraction means F10 for evaluating and extracting its features from the motif data and information (generally indicated by PA) given by this means F10. Melody control information generating means F20 for generating information (comprehensively shown by PC) and melody generation executing means F30 for specifically generating a melody (single melody) according to the information PC given by this means F20. There is. The melody control information generating means F20 has the ability to associate and plan a melody based on the motif feature parameter PA, and the melody generation executing means F30 decodes this plan (represented by a PC) and creates a melody according to the plan. It has rules to generate. Broadly speaking, means F20 and F3
A melody generating means is constituted by 0. On the other hand, in terms of inference, the motif feature parameter extraction means F10 and the melody generation execution means F30 are basically in an inverse relationship, and the motif feature parameter extraction means F10 extracts the essence from a specific motif (input melody). On the other hand, the melody generation executing means F30 has the inference ability to derive the characteristic parameter PA, which is the characteristic parameter PA. The planning space of the melody control information generating means F20 is vast, and the degree to which PA is reflected on the PC can be freely and variously controlled. In the illustrated example, the melody control information generating means F20
Only PA from the means F10 can be input to
It may be configured to receive other input data (for example, a user's request input from the input device 1, global instruction information regarding a song (information that acts to narrow down the range associated with the means F20)). . For example, although not shown, all or one of the parameters B to be used from the parameter B memory of FIG. 1 and the musical identification data to be used from the musical identification data memory 8 may be selected by the user. You can entrust the department.

Generally, a song changes as it progresses. However, given a limited amount of time, certain features are fixed. This automatic composer also takes this point into consideration, and has the concept of a unit of "section" that is regarded as fixed. For example, each function in the motif feature parameter extraction means F10 extracts PA in units of a certain section. Similarly, the melody control information generation means F20 passes the value of the PC assigned to each section to the melody generation execution means F30, and the melody generation execution means F30 generates the melody of the section using the value of the PC. Of course, this does not mean that all PCs change in the same section (the same applies to PA). Depending on the type of PC, the intervals in which the value can be fixed are generally not equal. Therefore, in principle, it is also possible to provide a unit of a different section for each type of PC and generate a value in a different section for each type. However, this complicates the process. Therefore, in the embodiment, a section of a type much smaller than the type of PC, preferably a section of a unit that can be common to all PCs (section of the greatest common divisor) is used. The same applies to the motif feature parameter extraction means F10.

In the detailed description below, in particular, each function F10, F20, F30 has a common section as the simplest example. As this section, "1
"Measures" is selected. Of course, the present invention is not limited to this, and each function may have two or more different sections. For example, the section for the pitch sequence and the section for the pitch length need not always be the same. Further, while the means F20 assigns a PC value to each measure, the chord progression can also be made to proceed in a section in which the measure is not a unit.

Details of the main functions F10, F20 and F30 of FIG. 2 will be described. The motif feature parameter extraction means F10 comprises a motif pitch pattern extraction means F11 and a motif tone length (rhythm) pattern extraction means F12. The motif pitch pattern extraction means is for extracting characteristics relating to the pitch sequence of the motif, and is a non-harmonic sound extraction means F11-1 for extracting non-harmonic sounds included in the motif, and a sound obtained by removing the non-harmonic sounds from the motif. It includes a distributed chord pattern extraction means F11-2 for extracting a high pattern, that is, a distributed chord pattern (LLi described later). F11-1A
Is a means for classifying and extracting the types of non-harmonic sounds, which is one mode of the means F11-1. On the other hand, the motif tone length (rhythm) pattern extraction means F12 is for evaluating and extracting the features relating to the tone length sequence of the motif, and the feature mini pattern extraction means F12-1 for extracting the characteristic mini patterns included in the motif. , A distributed chord (rhythm) pattern extraction means F12-2 that forms a disperse chord pattern by absorbing the sound length of the non-harmonic sound included in the motif into the sound length of the harmony sound, that is, a sound length sequence of only the harmony sound. .

The pitch patterns and rhythm patterns of the distributed chords are extracted by the means F11-2 and F12-2 based on the idea of the embodiment that the chord sound is the basis of the melody (function. Element F31 in F30-
1 and F31-2 are based on the same idea).

In the block of the melody control information generating means F20, a progress-dependent parameter generating means F2 for generating a parameter depending on the section number indicated by the section counter F23.
2 is shown. Some of these parameters include
A parameter that regularly fluctuates as the music progresses is also included, and this kind of parameter is generated by the regular fluctuation parameter generating means F22-1. The random number generating means F24 acts on the parameter generated by the means F22, and a random number or fluctuation controlled by the parameter can be introduced into the parameter. Above F
The elements 22 and F23 are shown to show that the PC has a parameter having the above-mentioned property, and the elements of non-arithmetic type (for example, parameter C
It is shown in order to clarify that the parameter C (melody control information) can be generated also in the database (1). Actually, in the present embodiment, the parameter C is generated in the arithmetic type, and the part that executes this arithmetic is F21.
Is a calculation means. The calculation means F21 is the means F10.
Of the feature parameter PA, the information of the parameter B indicated by I1, the bar number indicated by the bar counter F23-1 and the musical expression identification data from the musical expression identification data generating means F25 as input, and these inputs as variables. The parameter C is calculated. The musical expression identification data generating means F25 has a phrase counter F25-1, which is used to select information on a specific musical phrase from the musical expression identification data I2. The information about the passage includes the type of the passage (repeated type, expanded type). The musical expression identification data generating means F25 reads the bar number indicated by the bar counter F23-1, inspects the positional relationship between the bar number and the musical interval, and decodes (decodes) the related musical expression identification data based on the inspection result. To do. The decrypted music expression identification data is the calculation means F2.
Passed to 1. The role of the musical expression identification data generating means F25 is to give a higher hierarchy to the music.

As can be seen from the above description, various PCs are output from the melody control information generating means F20, some PCs are constant over a relatively long period, and some PCs are in a unit section (here, one bar). It fluctuates in a cycle of a comparable degree, and one PC is affected by a musical expression and changes to another value at a specific section or point. The same PA depending on the value of the parameter B, the value of the music identification data, the type of the function used by the calculating means F21, etc.
Even so, various PCs are generated.

The melody generation executing means F30 has, as its main elements, a pitch sequence generating means F31 and a pitch length generating means F32.
And have. The pitch sequence generating means F32 is the PC itself from the chord in progress from the chord progression data I3 and the melody control information generating means F20, or the random number generating means F3.
A distributed chord generation means F31-1 for generating a pitch sequence of distributed chords using a PC in which fluctuations are introduced by 1-4;
Before or after the generated distributed chords, the non-harmonic sound adding means F31-2 is provided for adding the non-harmonic sounds according to the plan of the PC and according to the internal addition rules. Each non-harmonic sound classification adding means F31-2A is an example of the configuration of the means F31-2. F31-3 is a pitch control means used and has a function of controlling the use of the 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 note scale generation means F31-.
3A is used to generate data in which each pitch of the scale is weighted, and the effective pitch inspecting means F31-3B is caused to inspect the effectiveness of the candidate pitch based on the weight. The melody note that has passed the inspection is sent back to the means F31-1, F31-2 and used as a formal melody note here.

The tone length sequence generating means F32 is an optimum combining means F3.
2-1, an optimum dividing means F32-2, and a characteristic pattern incorporating means F32-3, and generates a tone length sequence according to the plan of the PC. Optimal combining means F32-1 and optimal dividing means F
In the present embodiment, 32-2 is used to form a tone length sequence of distributed chords. The optimum combination means F32-1 is based on an initial tone length sequence (for example, the dispersed chord pattern extracted by the dispersed chord (rhythm) pattern extraction means F12-2 of F10 or a pattern obtained by modifying this), which is a target dispersed chord. The note lengths are combined with the minimum number of combinations until the number of notes (given by PC) of Similarly, the optimum dividing means F
32-2 divides the tone length of the initial tone length sequence by the minimum number of divisions until reaching the distributed chord number PC. Furthermore, both means F3
2-1 and F32-2 are divided using a pulse scale,
Perform the join. The pulse scale may be given from the means F20 as a kind of PC (melody plan information), but in the flow showing the operation of the embodiment described later,
The pulse scale is inherent in the division and connection rules of both means. On the other hand, the feature pattern embedding means F32-3 operates according to the associated PC and incorporates the feature mini-pattern in the tone length sequence of the melody. In this example, the characteristic mini-pattern is injected in the final process of melody generation.

The control means F33 is for controlling the activation of each element of the melody generation executing means F30 and the transfer of data between the elements. The melody data which is the execution result of the melody generation executing means F30 is the melody data memory 1
3 (FIG. 1).

<General Flow> FIG. 3 shows the entire operation of the automatic composer according to the present embodiment in a partial flow manner and a partial block manner. This figure shows the position of each process,
It shows the numbers of the drawings to be referenced and will not be described further here.

<Preliminary Items> The detailed operation of the embodiment will be described below, but before that, the preliminary items will be described.
FIG. 4 shows a list of main variables used in the automatic composer. The description itself is clear and will not be described here. FIG. 5 is a list of parameters C. Although the melody control information may include parameters other than those shown in the drawings, in the following drawings, only parameters using PC characters are shown here. Since the description itself is clear, the description is omitted here. FIG. 6 shows an example of pitch data assigned to each pitch of the scale. In the following description, data (pitch data) is assigned to pitches as shown in b of the figure unless otherwise specified. That is, a continuous integer value, which increases by 1 each time a semitone is raised, is assigned to each pitch of the scale. Rests are expressed by a special value. Other than this, some pitch data can be assigned, an example of which is shown in FIG.

The minimum unit of note length (pitch) is a sixteenth note. That is, the note length data of a 16th note is 1, and the note length data of a note that is twice as long is 2. In addition, the unit of the “extraction section” and the “progression section” is one bar. In connection with this, chord progression information is also one chord per bar. Furthermore, the length of the motif (input melody) is one bar unless otherwise specified. It is also assumed that the bar lengths are the same regardless of the bar number.

For convenience of explanation, it is assumed that each chord has four pitches. Here, there are four independent chords as chords, for example, a type such as C major sevens (do, mi, so, si), and two of them have an octave difference.
There are chord (triad) types. Therefore,
In this example, the three chords (do, mi, and so) are

[Outside 1] And In relation to this, there are four chord areas in the chord constituent note memory 2, and each address contains a value corresponding to the pitch of each constituent note. For example,

In the case of [External 1], the data is (1, 5, 8, 13). The above premise is merely for convenience of explanation,
It is merely an example premise.

FIG. 7 shows a data example (No. 1) for explaining the operation. The musical score shown in FIG. 7 (1) is a motif given as an example in the following description of the operation. Since the rest of FIG. 7 is a self-explanatory description, its explanation is omitted here. FIG. 8 is a data example (No. 2) for explaining the operation, which relates to the value of the PC. FIG. 9 is a musical score showing the generated melody for each process. It is a melody created in the case of the input data of FIGS. 7 and 8. With the above, the explanation of the preliminary matters is completed, and the detailed explanation will be started from now.

<Motif evaluation> Motif information is shown in FIG.
Is input from the input device of No. 2 and stored in the motif memory 6. The first task that the automatic composer performs on this motif information is evaluation. FIG. 10 shows a flow of motif evaluation executed by the CPU 9. According to this flow, the CPU 9 evaluates the motif by using one bar as a unit of an evaluation or extraction section. The main part of the flow of FIG. 10 is 10-11.
Rhythm evaluation and 10-12 non-harmonic sound extraction. In the figure, i is the number of the motif data, that is, a variable indicating the number of the note in the motif measure of interest, and AFT is the variable of the number of notes after the i-th note within the same measure ( Register) and BEF are variables of the preceding note number (10-2, 10-3). 10-4-1
0-10 is a pitch difference between adjacent notes in the pitch sequence MD _{i-1 to} MD _{i + 4} of a total of 6 notes before and after the i-th note MDi, that is, a pitch is calculated. is there.
For example, a ₁ is M for MD (the note of interest)
It is the pitch of D _{i + 1} (the pitch of the next note). Note that there is no pitch concept in rests, so it is treated specially (1
0-5, 10-8).

As will be described later, in the non-harmonic sound extraction of 10-12, the positional relationship A with respect to the bar line of the note of interest is A.
FT, BEF and pitch progression before and after the note a ₀
From ~a ₄ Prefecture, it is extracted by classifying the nonharmonic tones contained in the motif. Therefore, the above 10-2 to 10-1
The processing up to 0 is a preparation for the non-harmonic sound extraction processing 10-12. NO is the total number of notes (including rests) included in the motif measure, and the flow shown in the figure is exited at the time when the evaluation of all the notes is completed (when i> NO is satisfied). The description of the flowchart of FIG. 10 is 10-11.
With the exception of and 10-12, they are self-explanatory and need no further explanation.

★ Rhythm Evaluation ★ FIG. 11 shows an example of a detailed flow of the rhythm evaluation 10-11 in FIG. The purpose of the rhythm evaluation here is to find out what kind of mini pattern is included in the motif and to what extent. In FIG. 11, as a mini pattern (sound length sequence) to be investigated, a pattern having a sound length ratio of 3: 1, for example,

[Outside 2] A pattern with a tone length ratio of 1: 3, for example,

[Outside 3] A pattern with a tone length ratio of 1: 1, for example,

[Outside 4] A pattern with a tone length ratio of 2: 1: 1, for example,

[Outside 5] A pattern with a tone length ratio of 1: 2: 1, for example,

[Outside 6] A pattern with a tone length ratio of 1: 1, 2:

[Outside 7] Is illustrated (notes include rests).

The number of patterns with a tone length ratio of 3: 1 included in the motif is counted in HR1, the number of mini patterns of 1: 3 is HR2, the number of 1: 1 patterns is HR3,
The number of 2: 1: 1 patterns is HR4, the number of 1: 2: 1 patterns is HR5, and the number of 1: 1: 1 patterns is H.
Stored in R6. Note lengths MRj, MR of adjacent notes
The calculation of the ratio of _{j + 1} is performed in 11-3. Since the inspection is performed on the notes in the measure, the pattern check of the note length ratio of the three notes may be performed two points before the last note in the measure (at that time, 11-10, j = NO- 1
Is established). By the way, the motif shown in the first bar of Fig. 9

[Outside 8] In the case of, the counter HR3 of the one-to-one pattern is 2

[Outside 9] The counter HR4 of the 2-to-1 pattern is 1

[Outside 10] The counter HR6 of the one-to-one-two pattern is 1

[Outside 11] And other pattern counters HR1, HR2,
HR5 indicates that the number is zero.

As will be described later, the mini-pattern (dominant mini-pattern) that characterizes the motif by checking the magnitude of the count value of each mini-pattern counter HR1 to HR6 (the number of each tone length pattern included in the motif). You can know. Furthermore, by incorporating the feature mini-pattern into the tone length sequence of the melody that produces it, the rhythmic feature of the motif can be reflected in the melody.

* Non-harmonic sound extraction * FIG. 12 shows an example of a detailed flow of non-harmonic extraction shown by 10-12 in FIG. This flow 12-1 as shown
12 to 58 have a tree structure, and the automatic composer has a positional relationship AFT and BEF with respect to a bar line of a note,
Progress a ₀ of the pitch to the note column of around the note to form ~
By examining a ₆ , it is possible to derive which note is which type of non-harmonic sound by the forward inference method. The derivation result that a particular note is a particular non-harmonic sound is the array {H
It is stored by writing a unique value to a register HDi of a specific number on Di} (where the array {HDi} is
This is an area prepared on the work memory 10 for motif evaluation, and the pitch data {MDi} of the motif is initialized as initial data in this area). For example, HDi = -40 means that the i-th note in the motif of the bar of interest is a non-harmonic sound named transitive sound. In the figure, as the non-harmonic sound, the pre-taken sound,
Sounds, embroidery, transitions, and other non-harmonic sounds are shown. The name of each non-harmonic sound is for illustration purposes only. In other words, there are some conclusions (eg, HD _{i + 1 of} 12-41 at the end of the flow (network) of the tree structure).
-40, that is, the conclusion that the (i + 1) th note is a transitive note), but there is more than one root from the top of the tree to that conclusion. These routes are the conditions and definitions that make the conclusion true.
That is, the flow of FIG. 12 classifies and defines each non-harmonic sound. This suggests that a different definition is possible for each non-harmonic sound. Those skilled in the art will readily be able to implement non-harmonic extraction according to other definitions in accordance with the teachings of the present invention. The description of the flow in FIG. 12 is clear, and the description will be used instead of the description of the specification.

<Motif Parameter Extraction> If the above-mentioned motif evaluation is called a “first analysis” of a motif, the motif parameter extraction described below is a “second analysis of a motif”.
Analysis ”. Both functions are unchanged in that they analyze the features of the motif. FIG. 13 illustrates an overall flow of motif parameter extraction. According to this flow, in 13-1, a chord type parameter, that is, a pitch flow or a vertical component pattern {LLi} for the distributed chords included in the motif is extracted, and in 13-2, the pattern is extracted as the motif. The flow of the tone length of the included distributed chord or the pattern {RHi} of the horizontal component is extracted (formed). In 13-3, the number of non-harmonic sounds is calculated, and in 13-4, the number of harmonic sounds is obtained.
In 13-6, the parameter of the smoothness of the pitch sequence of the motif is extracted, and in 13-7, the parameter of the number of continuous sounds at the same pitch (homogeneous progression) is extracted. 13-7 is where the minimum sound length included in the motif is extracted.
In the last 13-8, characteristic rhythm parameters of the motif are extracted. Note that the order of these processes does not have to be as illustrated.

To add a little explanation, the automatic actuator of this example handles harmony and non-harmonic tones separately.
13-1 to 13-6 are related to this. For example, in 13-2, the non-harmonic sound is removed from the tone length sequence of the motif itself. That is, the non-harmonic notes are absorbed by the harmonic notes to form a tone length pattern of only the harmonic notes. Further, the smoothness of 13-5 and 13-6 and the parameter of the homophonic progression are the parameters for the “dispersed harmony”, and the “non-harmonic” is excluded from the extraction target. Such an extraction approach is based on the idea of harmony.

* Harmonic tone type parameter extraction * FIG. 14 shows details of the process 13-1 of FIG. Is a flow for extracting. The purpose of the processing here is to find out what pitch each of the harmonic sounds distributed along the flow of the motif is in the collection of the harmonic sounds of the entire motif. The secondary purpose is to find out which note is the rest. In the figure, a variable indicated by ONPU represents a range (system range) used in the automatic composer, and is determined so as to include the range of the input motif.

In the illustrated flows 14-1 to 14-16,
From the lowest pitch of the system pitch, it is checked whether or not the sound (harmonic sound) is included in the motif in order, and if there is, the variable M of the harmonic pitch number is incremented,
The value is LLi. When HDi is “rest”, that is, when the i-th motif data is a rest, LLi is set to zero, and it is stored that the i-th motif data is a rest. Further, when HDi is negative, that is, the non-harmonic notes are skipped.

An example of the processing result of this flow is given in the upper right part of the flow of FIG. In this example, the first sound (indicated by HD ₁ ) of the motif is a harmony sound, which is the highest harmony sound among the illustrated motifs. The second sound HD ₂ is a non-harmonic sound. The third sound HD ₃ is a harmony sound (harmonic sound that appears second in the motif), but is a harmony sound with a lower pitch than the first harmony sound. The fourth note HD _{4 of the} motif is also a harmony sound (a harmony sound that appears in the third motif), but at a lower pitch. Motif 5
The fifth HD ₅ is a rest. The sixth (in this case, the last) note HD ₆ of the motif is a harmony and has the lowest pitch, like the 4th note in the motif. As can be seen from this figure, the flow of the harmony sound of the motif has a descending pattern as a whole. The result of the flow of FIG. 4 is as follows for this example motif. First, H
Zero indicates rests as its LLi allocated for D _5. HD ₄ is assigned 1 as its LLi, which is the lowest note in the motif. The same applies to HD ₆ , L to HD ₃
2 is assigned as Li, which indicates that it is the second highest harmonic sound in the motif. HD ₂ is a non-harmonic sound and LLi is not assigned. HD ₁ is the highest note among the harmonic sounds contained in the motif, and is 3rd harmonic pitch from the bottom, so 3 is assigned to that LLi. That is, LL _{1 of} HD ₁ (first harmonic sound)
Is 3, HD ₃ (LL ₂ of the second harmony sound is 2, H
The LL ₃ of the D4 (the third appearing harmony) is 1, the rest comes after the third harmony, the LL ₄ is 0, the LL ₅ of the next harmony HD ₆ is 1, and the flow of the harmony of the motif , Or the type (including rests here) has been specified. That is, {LL} = (LL ₁ , LL ₂ , LL ₃ , L
L ₄ , LL ₅ ) = ( ₃ , ₂ , 1, 0, 1) is obtained.

Although not shown in the form of PA, this type of chord (dispersed chord) type is also a parameter that characterizes the motif. If this pattern is used without any modification in the melody generation which will be described later in detail, a highly repeatable and highly uniform melody will be generated. The motif of Fig. 7 (Fig. 9)

[Outside 12] , LL ₁ = 1 of the _first harmony sound de,
LL ₂ = 2 of the second harmony sound, LL ₃ = 3 of the third harmony sound, LL ₃ = 3 of the third harmony sound, and the fourth harmony sound

[Outside 13] LL ₄ = 4. This is the ascending pattern.

[Extracting Harmonic Tone Length Pattern] FIG. 15 shows an example of a detailed flow of the processing 13-2 in FIG. The purpose of the processing here is to form a tone length pattern (tone length sequence of only harmony tones) in which nonharmonic tones are removed from a motif including nonharmonic tones. In other words, the process of absorbing the sound length of each non-harmonic sound into the sound length of the adjacent harmony sound is performed. Then, which non-harmonic sound is absorbed by which harmonic sound is illustrated in FIG.
It is determined by the pulse scale of the meter system. Now
If a non-harmonic sound is sandwiched between a certain sound and the weight on the pulse scale at the start position of the non-harmonic sound, the note next to the non-harmonic sound (for simplicity, it is regarded as a harmony sound) If the weight of the starting position of the next note is lighter, the length of the non-harmonic sound is absorbed by the length of the next note (for simplicity, imagine a harmonic sound) Is
If the starting position of the next note is heavier, the non-harmonic sound is absorbed by the previous note (you can imagine a harmony sound for simplicity). The above is the logic of sound length absorption of nonharmonic sounds by the pulse scale (positive logic). Harmonic sound that absorbs the sound length of non-harmonic sound has a wider sound length than its original length.

The pulse scale (5, 1,
2, 1 ...) are logical scales, and are scales incorporated in the logic of the flow of FIG. In FIG. 15, SUM is a register SUM that indicates the distance or length from the start position of the next note to the bar line (bar line in front).
MB is a register that indicates the distance and length from the start position of the current note to the bar line in front of it, MRi is the length of the i-th note of the motif, and RH _N is the length of the N-th "harmonic" of the motif Therefore, after the processing of FIG. 15 is completed, it has a length obtained by adding the lengths of zero or a plurality of non-harmonic sounds located before or after. NO is the total number of musical notes in the motif. S
UM MOD 2 ^j is the remainder of the value of SUM divided by 2 ^j, as well SUMB MOD 2 ^j 2 the value of SUMB is ^j
It is the remainder divided by.

Reference numeral 15-4 is a check as to whether or not the i-th note is a harmony tone. In the case of a harmony tone, N is incremented and added to the N-th harmony tone length RH _N (15-5).
15-6 is a check whether the non-harmonic sound is the first note,
15-8 is a check whether the non-harmonic sound is the last note. When the non-harmonic sound is the first note, the start position of the sound coincides with the leading pulse point on the pulse scale (FIG. 16). Its weight is 5, which is the maximum. Therefore,
According to the logic of the pulse scale, this note length is absorbed by the next note (15-7). Similarly, if the non-harmonic is the last note, the start of the next note has the greatest weight (assuming no note crosses the bar line), so this last note is absorbed by the previous note. (15-9). 15-
10 to 15-15 add the length of the non-harmonic sound which is neither the beginning nor the end of the measure to the note length RH _N of the preceding note (harmonic sound) or the note length RH of the next note (harmonic sound). _{N + 1}
Is being processed according to the logic of the pulse scale.

By the way, in the case of the motif shown in FIG.

[Outside 14] And Fah is a non-harmonic sound, and the others are harmony sounds. The starting position of the next sound of Ph is Seo is the eighth from the beginning of the bar, with one bar having a length of 16 (see the top of FIG. 16). Its weight is 4. On the other hand, the start position of the far is 6, and its weight is 2. Seo weight (4)
> Since the weight of fa is (2), the non-harmonic fa is absorbed by the preceding harmonic mi (in the flow, when j = 2, 15).
-11 becomes true, NF = 0, 15-16 at 15-12
Then RH ₂ = RH ₂ + MR ₃ ). After all,

[Outside 15] Becomes That is, the tone length pattern {RHi} of the harmony tone
Is {RHi} = (RH ₁ , RH ₂ , RH ₃ , RH ₄ ) =
(4, 4, 4, 4). From the above description and the clear description of the flow of FIG. 15, the operation of extracting the tone length pattern of the harmony sound is clear, and further detailed description will be omitted.

Here, the pulse scale of FIG. 16 is merely an example, and it is possible to form a tone length sequence of a harmony tone by another pulse scale. This is clear. For example, when (weight of starting position of non-harmonic sound)> (weight of starting position of next harmonic sound), the non-harmonic sound is absorbed by the "previous" sound, and (weight of starting position of non-harmonic sound) ) <(Weight of the starting position of the next chord sound), the non-harmonic sound is absorbed by the "next" sound (harmonic sound). If the reverse logic is used, the result for the motif is

[Outside 16] ({RHi} = (4,2,6,4)).

★ Number of non-harmonic tones and harmonic tones ★ A detailed flow of the processes 13-3 and 13-4 in FIG. 13 is shown in FIG.
7, for example. 17-1 to 17-7 are parameters PAi
It is the initialization of j. The number of pre-taken sounds included in the motif is P
A ₅ and ₂ , PA and PA ₂ and ₂ , respectively.
To _4,4, the number of passing notes is stored in the PA _3,3. The result of SUM shows the total number of non-harmonic sounds contained in the bar of the motif. The contents of PAs _{1 and 3} indicate the total number of harmonic sounds contained in the bar of the motif.

[Parameters for smoothness and homophonic progression] FIG. 1 is a detailed flow chart of the processes 13-5 and 13-6 in FIG.
8 for example. As described above, {LLk} is a distributed harmony tone pattern appearing in the motif measure. In the illustrated flow, the PA _{1, 2,} the degree of maximum jump between adjacent harmonic sounds (smooth parameter) is extracted, is set, PA _{1, 6}
Is extracted and set as the number of sounds (harmonic sounds) proceeding with the same sound as a parameter of the same sound progression. Note that rest LLk, which has no meaning of pitch, has a value of zero. Related to this, if either LLk or LLk + 1 is a rest, 1
The result of the check in 8-4 and 18-6 is NO (for example, after the processing in 18-3, LLk = 0 or LLk _{+ 1} = 0 is discriminated, and when they are satisfied, L = -1 is set. Can be achieved with).

[Characteristic Rhythm Parameter] FIG. 19 is an example of the flow of the process 13-8 in FIG.
As described with reference to FIG. 11, the automatic composer of the present embodiment seeks what kind of mini rhythm pattern is included in the motif and how much. The remaining question is which mini-pattern will be the pattern that characterizes the motif. Here, the dominant minipattern included in the motif is defined as the characteristic minipattern of the motif. An example of evaluation based on such an approach is the flow shown in FIG. In this flow, in the motif, a mini pattern with a tone length ratio of 3: 1

[Outside 17] And a mini pattern with a tone length ratio of 1: 1

[Outside 18] Is investigating which of them is dominant in numbers. If the number HR ₃ of mini patterns with a tone length ratio of 1: 1 is greater,
PA _6,1 indicates zero, and PA _6,1 is set to a value of 3 if the number HR ₁ of mini-patterns having a tone length ratio of 3: ₁ is more dominant.

By the way, in the case of the motif of FIG. 9, HR1
= 0, HR3 = 2, and the one-to-one mini pattern is dominant. Therefore, the value of PA _6,1 will be zero (meaning that the one-to-one _minipattern is the _minipattern that characterizes the motif). The flow in FIG. 9 is merely an example, and those skilled in the art can easily create a flow for extracting other characteristic rhythm parameters. For example, the number of mini patterns with other tone length ratios (HR ₂ , HR ₄ , HR ₅ , HR ₆
Etc.) may be considered. As will be described later, by incorporating the characteristic mini-pattern indicated by the value of PA _6,1 into the tone length sequence of the melody, the rhythmic characteristic of the motif can be reflected in the generated melody.

[Minimum Sound Length Extraction] FIG. 20 is an example of a detailed flow of processing 13-7 in FIG. In this flow example, the minimum tone duration that are found among the tone duration MRi motif is set to renewably min, the minimum tone duration finally obtained is stored in the PA _3,3. The minimum note length information PA _3,3 can be associated with the minimum (shortest) note length of the generated melody. For example, if the minimum tone length PA _3,3 of the motif is made equal to the minimum tone length PC _3,3 of the melody that produces the shortest melody note, the shorter melody note in the motif will not be created (of course, if not made equal). It does not mean that it does not mean that it can be controlled). Anyway, PA _{3, 3} is one of the motif length parameters (rhythm parameters).

[0058]Melody generation  In the above description, the automatic composer of this embodiment is given.
How to evaluate the motifs and how to evaluate their characteristics
It became clear whether to extract. This automatic bending machine
Later work will be based on the results of these feature evaluations and feature extractions.
It is to generate a melody based on. For this automatic composer
If so, the melody generation function is roughly divided and generated.
The part that plans the outline of the melody (the total melody
And the plan created by this melody planning department.
, The part that specifically generates the melody along (the melody
The generation execution part) and. The Melody Planning Department
It corresponds to the melody control information generating means F20 in FIG.
Yes (however, the function to automatically generate chord progression information as well)
If it is set, that function is also included in the melody planning department.
Will be. Controls the melody that also generates chord progression information
It is the same information that is controlled). Melody planning
The main work of Kanabe is extracted by the motif feature melameter extraction unit
Based on the result, the melody to be generated is "associative" or "total".
It is to draw. Melody planning department is a melody student
The planning information is stored in each part of the execution unit (F30 in FIG. 2).
That is, it gives melody control information. Melody generation execution unit
Then, according to the melody control information sent,
Execute the generation. Two main functions of the melody generator
Is a function to generate the pitch sequence of the melody and to generate the length sequence of the melody.
It is a function.

<Melody Planning> Melody planning information (melody control information) is referred to as parameter C (P
(It may be written as C). The melody created by the melody generation executing section is greatly influenced by the parameter C. In other words, the parameter C is a parameter that characterizes the generated melody. The overall task of this automatic composer is to compose a melody (here, the meaning of the song) based on the motif (input melody). As I often experience, some songs have a motif that appears many times in the same form or in a very similar form, while other songs have a motif that is transformed, expanded, developed, and developed. There is also a song that exists only temporarily, and a different motif appears and then moves on to another motif. The songs that belong to the first class have a high degree of "consistency", while the songs that belong to the second class remain "consistent" but rich in "diversity".
It can be said that the songs belonging to the third class have less "consistency" but more "freedom" and "diversity".

According to the present invention, it is possible to freely control from composition that is faithful to the melody (motif) input by the user to composition that emphasizes freedom and diversity rather than faithfulness to the motif. Parameter C (P
The characteristics of C) and various ways of generating the parameter C are described in detail in the patent application of the present applicant (filing date: April 8, 1987, name: "automatic bending machine"). Detailed description will not be given here. Parameter C
The first property of is that it depends on the progress section of the song. The second property is that it depends on the motif feature parameter. The first is related to the consistency and variety of songs. The second property is related to the degree of reflection of the motif in the created song.

In this embodiment, a section in units of "measures" is used as the music progression section, and the parameter C is assigned to each measure. Further, the parameter C is created by calculation, and as a component parameter for the calculation, a parameter B (sometimes written as PB) that controls the consistency and diversity of the melody flow and a motif feature parameter (written as PA There is also). Furthermore, in order to impart a higher level of hierarchy to the music, the music expression identification data (music expression identification parameter) is also used as a component parameter in the calculation of the parameter C. The parameter B is stored in the parameter B memory 7 shown in FIG. 1, and the music identification data is stored in the music identification data memory 8 shown in FIG. If desired, the user can request the automatic composer to use a specific parameter B with the input device 1, directly or indirectly, and likewise by specifying the musical expression of the music through the input device 1. , It is possible to request the automatic composer to use specific music identification data. However, the man-machine interface related to this (for example, conversion between the concept space for the user and the internal representation space of the automatic composer, for example, the parameter B, or the way of the user access to the parameter B) is related to the invention. Since it is not an eye, further explanation is omitted.

<Calculation of Parameter C> FIG. 21 exemplifies a flow of calculation of the parameter C. In 21-1, the motif feature parameter PA is read out and 21-2
Then, the consistency diversity control parameter PB is read out and the music identification data is read out from the memory 8 at 21-3. These three steps do not have to be performed for each bar, but can be performed collectively. In other words, it is a step that is processing unrelated to the bar number of the melody to be generated. 21-4, 21-5, 21-6 are executed for each measure (PC in step 25-9 in FIG. 25 described later).
The calculation corresponds to these 21-4 to 21-6). 21
-4 is a decoding of the musical identification data, which is a parameter PC assigned to the bar number of the melody generated in 21-5.
Is being calculated. As shown in the figure, the parameter PC is a variable or component parameter that includes a bar number, PA (motif feature parameter), PB (consistency diversity control parameter), and SB (comfort identification parameter, higher hierarchy control parameter). Formally expressed as a function. However, this does not mean that any of the individual parameters PC must actually depend on any of the bar numbers, PA, PB, SB. As illustrated in FIG. 5, the parameter C is a set of many parameters, and some of them may have a DC characteristic (a property that does not depend on the bar). Further, it does not mean that the value of PC must be determined "uniquely" if the measure number, PA, PB, and SB are given. Randomization can be introduced to the uniquely obtained intermediate PC value by the random number generation means (F24 in FIG. 2). Preferably, the randomization is a "controlled" randomization, by introducing a random number controlled by an intermediate PC value (eg by introducing fluctuations based on the intermediate PC value), the final parameter You can ask for a PC. Examples of controlled randomization are disclosed in the above mentioned patent applications.

21-6 is a parameter changing process by learning. Here, the parameter C generated in 21-5
The process of giving priority to the parameter C set by the user by correction is executed. For example, if the value of a certain parameter C at a certain measure is set to a value desired by the user through the learning function of the automatic composer, the value selected by the user is adopted as that parameter at that measure ( Details will be described later).

[Control of musical expression] In general, with regard to the developing portion of a music piece, the entire developing portion (rust) has a characteristic that it is different from that before the chorus, and a characteristic at a break point into the chorus. For example, the bar just before the chorus is usually accompanied by a feeling of end, which has an anticipatory psychological effect of entering the chorus. In addition, it is customary for the entire chorus to have different appearances due to changes in the range, number of notes, and rhythm. In order to provide such a high level hierarchy, the musical expression identification data is generated in the present embodiment to control the musical expression (21-3, 21-4, 21-5 in FIG. 21 described above). Is related to this).

FIG. 22 shows the music identification data memory 8 (FIG. 1).
2 shows an example of data stored in. Here, the data type is made up of A1 in the upper digit and A2 in the lower digit, where A1 indicates the bar number, that is, the bar number at which a certain passage starts. Part A2 defines the type of passage starting from the bar number indicated by A1. In particular, A
When the value of 2 is zero, it represents an iterative type, and when it is non-zero, it represents an expansion part, and its value represents the type of expansion.
As an example, SB1 = 10, SB2 = 91, SB3 = 1
The data of 30 describes that the music style of the song is A → B → A, the first verse A is from the first measure to the eighth measure, the type is repetitive, and the second verse B is The 3rd syllable A, which is a developmental syllable starting from the 9th bar and ending at the 12th bar, is a repetitive syllable starting from the 13th bar.

FIG. 23 shows an example of the detailed flow of 21-3 in FIG. 21 (however, 23-1 additionally shown for readability is the same as 21-2 in FIG. 21). . From the description of the flow of FIG. 23 and the above-described music style identification data (original music style identification data) of FIG. 22, the manner of the music style control of this column can be easily understood (see also FIG. 24). That is, in the musical expression control of this example, in order to characterize the melody over the entire expansion section (chorus),
The hold type (SBH) is prepared, and the pulse type (SBPi) is prepared in order to characterize the break (break between verses). Explaining according to the flow of FIG. 23, when the number of the measure for generating the melody is I, the CPU 9 (FIG. 1) enters 23-2 and resets the pulse type musical expression identification parameter register SBPj. In this example, FIG.
As can be seen from 4, the pulse type parameter is SB
P _1, SBP _2, a total of three are prepared of SBP _3, S
BP ₃ is a pulse parameter which is placed _two measures before the beginning of the syllable of the expansion part (which starts from the ninth measure in FIG. 24), SBP ₂ is a pulse parameter which is added one measure before the expansion part, and SBP ₁ is the expansion parameter of the expansion part. It is a pulse parameter attached to the start measure. In addition, only one SBH is prepared for the hold type, and it is attached between the expansion portions (between the ninth bar and the twelfth bar).

The processing from 23-3 to 23-12 in FIG. 23 is processing for realizing the above-mentioned function. 23-3
In 23-4, the phrase number counter j is initialized, and in 23-4, the upper digit of the musical expression identification data format, that is, the starting measure number of the phrase indicated by j is placed in the register a ₁ , and the lower digit, that is, the type of the phrase is entered. It is stored in register a ₂ . When the measure number I for melody generation is two measures before the start measure number a _{1 of the} passage, the variable a ₃ is set as the third pulse type identification parameter SBP ₃ (23-9,
23-10), when the melody measure number I points one measure before the start measure of the measure, the variable a _{2 of the} measure type
Is set as the second pulse type musical identification parameter SBP ₂ (23-7, 23-8), and when the melody bar number and the starting bar number of the passage match, a _{2 is set} to the first pulse type parameter SBP 2. Set to ₁ and hold type parameter SBH (23-5, 23-
6). The SBN of 23-12 is the total number of passages that compose the song. That is, in this flow, the measure number of the melody to be generated is used as the key data, and a search is made of all the phrases forming the song that match this measure number, and the corresponding processing is performed. Since the SBPi is reset every time the melody bar number is updated, it becomes a pulse type.
On the other hand, the SBH is only entered when the type of the phrase having the starting measure number that matches the melody measure number matches. Therefore, when the passage is an expansion type, S
BH has a non-zero value, and is zero in the repetitive type and is not reflected in the parameter C. The decrypted music expression identification data (music expression identification parameter) is calculated by the parameter C (see FIG. 2).
21-5) of 1 as a component parameter. For example, it is used as a DC component.

* Generation of melody * FIG. 25 is an overall flow when continuously generating melody. The main part of this flow is 25-9 PC calculation (see FIG. 2).
21-4 to 21-6 of No. 1) and melody generation of 25-10 (specific generation of melody, details will be described later). The rest is processing for transfer between memories, etc., and is the function of the melody controller in FIG. That is, 25-
In the processing from 1 to 25-5, the motif data in the motif memory 6 (FIG. 1) is stored in the melody data memory 13
Is the process of transferring to. No. 25-5 is the number of notes included in the motif.

MN ₀ shown in 25-6 and 25-16 is the total number of notes of the already created melody when continuously creating melody. Since the melody is generated in units of measures, the number of measures of the motif is calculated (for example, it can be calculated from the tone length data of the motif data) (25-
7). Here, it is shown that the motif may have two or more measures. The handling of motifs of two or more measures will be described later, but for the time being, it will be one measure as originally assumed. 1 is added to the calculated number of motif measures (25-8), and when a melody of one measure occurs (25-
9, 25-10) and write the data to the melody data memory 13 (25-1 to 25-15). 25-15
No. is, of course, the number of notes of the melody for one bar generated in the processing of 25-10. CN ₀ shown in 25-17 is the total number of chords used in chord progression, and is chord / measure in this example, so that the generation of the melody is completed when the measure number reaches the total number of chords. Now, the details of the concrete generation of the melody (25-10 in FIG. 25) will be described later.

<Concrete Generation of Melody> The function of concrete generation of a melody is roughly divided into a function of generating a melody pitch sequence (a melody pitch sequence generating function) and a melody pitch sequence. It consists of a function (melody tone length sequence generation function) that Both functions operate in accordance with the melody plan information (melody control information) sent from the above-mentioned melody plan planning unit for each bar, and each melody control information (collectively referred to as parameter C) is internally stored. The melody note is created in a form incorporated in the production rule of (for example, a form incorporated as a constraint condition). The melody pitch sequence generation function is a non-harmonic sound addition unit that adds a non-harmonic sound between the pitches of the distributed chords generated by the distributed chord unit and the pitches of the distributed chords that are generated by the distributed chord unit. including. Furthermore, the melody pitch sequence generation function is
The pitch control section determines the scale (note scale) used by the distributed chord section and the non-harmonic sound addition section and checks the validity of the pitch of the melody note candidate. Only valid candidates are accepted as melody notes.

On the other hand, the melody tone length sequence generation function generates a tone length sequence of distributed chords, a tone length sequence adjustment unit associated with the addition of a pitch of a non-harmonic tone by the non-harmonic tone addition unit, and a melody tone length. It consists of parts that add rests and incorporate characteristic mini-patterns as finishing lines. In the flow charts illustrated below, the above-described functions are activated and operated in a timely manner. In terms of the process, first, a melody containing only distributed chords (a sequence of pitches and a pitch sequence) is created, and then a non-harmonic sound is added to include a melody containing a distributed chord and a non-harmonic sound (a sequence of note lengths). The pitch sequence) is created, and finally the final sequence of the musical length sequence of the melody is performed.

Generation of Distributed Chord FIG. 26 is an example of a flow including generation of dispersed chord. Two
The processing from 6-1 to 26-5 is basically 26-
This is a preparation for creating a pitch sequence of distributed chords of 6 to 26-42. Reference numeral 26-1 is a read-out of the constituent notes of the chord assigned to the melody measure to be generated from now on.
Is the inversion of the constituent sounds of a chord. The PCs _{1 and 12} of 26-3 are for checking whether the number of turns is the optimum number of turns or the normal number of turns (the number of turns indicated by the PCs _{1 and 7} given by the parameter C calculation unit) _{. When 12} > 0, the optimum number of turns is calculated at 26-4, and PCs _{1 and 7} are changed to the calculated value.
Will be updated. The turn of the chord component sound influences the range of the generated melody bar, and in the present example, the larger the number of turns, the higher the range of use. The function of optimal turning is to create a melody in this measure that optimally connects with the melody flow of the previous measure. In particular, it is effective when connecting a sound pattern similar to the sound pattern of the previous bar.

In the scale weight change of 26-2, the weighted note scale data {SCLi} is read out, and the weight of each read out pitch is a predetermined condition (whether or not it is a chord component note, root shift is It is changed depending on the conditions such as whether it is required. The changed note scale data {SCLi} is referred to when the candidate of the melody note of the distributed chord is obtained, and is the official melody if it is equal to or more than the threshold value (PC _{1, 1} ) sent from the parameter calculation unit. It has been adopted as a note. The changed note scale data {SCLi} is also referred to in the process of adding non-harmonic tones, which will be described later, and is adopted as a formal melody note if the pitch of the candidate for addition has a weight equal to or greater than the threshold value of each non-harmonic tone. To be done. The last process 26 in FIG. 26
Reference numeral -44 is a process for determining a tone length, that is, a process for forming a tone length sequence of distributed chords. This is the end of the description of the outline of FIG. Hereinafter, a more detailed description will be given for each individual.

★ Reading of chord constituent notes ★ To generate distributed chords, chords in progress (chords)
You first need to know what is. The chord progression sound memory 2 and the chord progression memory 3 in FIG. 1 give the chord progression information. FIG. 27 shows an example of the chord progression stored in the chord progression memory 3 and the constituent note data of each chord stored in the chord constituent note memory 2. Further, FIG. 27 also shows the contents of the chord root data memory 4. In the example in the figure, each chord has four constituent tones (independent four-voice type or two-voice octave type).
As a result, the chord component sound storage area of the chord component sound memory 2 has four consecutive addresses, and the pitch data of the chord component sounds are entered in ascending order from a young address. For example, the contents of the addresses 1 to 4 of the chord component sound memory 2 are 1 (= do) and 5 (=
Mi), 8 (= So),

[Outside 19] And these data describe the constituent sounds of C _maj .

On the other hand, the chord progression memory 3 stores data indicating a chord name or chord type at each successive address. This sequence of data describes the chord progression. From the data (representing the chord type) stored in each address of the chord progression memory 3, the start address of the area storing the corresponding chord in the chord component note memory 2 can be calculated. Similarly, the root note storage address of the corresponding chord in the root note data memory 4 can be obtained from the data stored in each address of the chord progression memory 3. In the illustrated example, the start address of the chord component note memory 2 for storing the chord is obtained by calculating (data-1) * 4 + 1 using the value of the data representing the chord type. To read chord constituent sounds, chord progression memory 3
This can be executed by reading out a more advanced chord and using the value of the chord as a pointer to read the data of the chord constituting sound from the chord constituting sound memory 2.

Since one code / measure is assumed in this example, at the time of the I-th measure, the I-th code of the chord progression memory (chord progression file) 3 is read (the I-th address is By specifying). At the end of the chord progression file, EOF (End of File) indicating completion of chord progression is entered. However, regarding the reading of the chord component sound, in this example, all the chords of the chord progression file are read, and the constituent voices of all the read chords are taken out from the chord component sound file (of course, in 26-3 and later of FIG. 26, Turn around for the chord sounds that are in progress). That is,
The flow shown in FIG. 28 is the batch reading of chord constituent sounds (details of 26-1 in FIG. 26).

In the flow of reading the chord component sound shown in FIG. 28, the processes from 28-1 to 28-4 are performed while the chord progression memory 3 (see FIG. 27) is reading out the chord number data (chord name) sequentially. is there. 28-2 represents setting the contents of the address of the chord progression memory 3 indicated by the value of i, that is, the i-th chord name in the register CN _i . 28-3 EOF
Is the end code stored at the address next to the last code name, and the reading of the code name is completed when the end code is read. 28 to 28 in FIG. 28
In the processing up to -12, the pitch data of the chord constituent tone of each chord name is read by referring to the chord constituent tone memory 2 from each read chord name. Two
J = (CN _i −1) × 4 + 1 in 8-7 is the calculation of the read start address of each code, and is 28-8 to 28-1.
At 0, four pitch data are read from the start address and are set in the register KD _ij .

In the following description, KD _i1 , KD _i2 , K
Instead of D _i3 , KD _i4 , simply KD ₁ , KD ₂ , KD ₃ , K
Let's call it D ₄ . KD ₁ is the lowest note in the harmony, K
D ₂ is the next lowest harmony, KD ₃ is the next lowest harmony, KD ₄
Is used as the highest harmonic register. At this stage, in KD ₁ , KD ₂ , KD ₃ and KD ₄ , the pitches of the chord constituent notes (in the i-th measure) are in the basic form, that is, they are stored in the chord constituent note memory 2 as shown. That is why. For example, the chord at the third bar of the musical score of FIG. 9 is G ₇ , and KD ₁ to KD ₄ corresponding thereto are KD ₁ =
3, KD ₂ = 6, KD ₃ = 8, KD ₄ = 12 (Le, Fa,
So, Shi).

[Modification of Scale Weights] As will be described later, when a melody note candidate is obtained in the process of the distributed chord section or the non-harmonic sound addition section, the pitch of the candidate is regarded as an effective pitch. It is checked whether it is possible pitch. The information that is the basis of such a check is the scale (note scale). In this embodiment, a set of weighted note scales is stored in the scale weight data memory 5 of FIG. An example thereof is shown in FIG. Here, four types of note scale data are illustrated. It is a minor scale without jonas, a major scale without jonas, an Okinawan scale and major scales (Europe). For example, in the major scale (European) note scale, the pitches of do, le, mi, fa, so, la, and si are assigned a value of 75, and other pitches are 2
A value (weight) of 5 is attached. The weight assigned to each pitch represents a measure of the likelihood that the pitch will be used. In terms of memory, the scale data of this major scale (Europe) is 36 to 4 of the scale weight data memory 5.
It is stored at address 7, and when the data of this major scale (Europe) is read in the data reading 30A-1 in the flow of FIG. 30 (A), the array {SCL
In the first register of i}, that is, SCL1, the data at address 36, that is, the weight of 75 is set, and S
CL2 is set to a sharp weight of 25 at address 37. Below, SCL3 = 75, SCL4 = 25, SCL
5 = 75, SCL6 = 75, SCL7 = 25, SCL8
= 75, SCL9 = 25, SCL10 = 75, SCL1
1 = 25 and SCL12 = 75.

Data reading 30A-1 in FIG.
The details of the above are illustrated in FIG. User input device 1
The type of scale is input by (30B-1). Based on the input information, the CPU 9 calculates the start address of the designated scale data on the scale weight data memory 5 (30B-2), and arranges the respective data of 12 addresses following this start address. Write to SCLi}. PC 1 and ₁₄ shown at 30A-2 in FIG. 30 (A) are parameters indicating whether or not the root _sound is shifted, which is given from the parameter C calculation unit, and when PC 1 and ₁₄ > 0 are satisfied, 3
The routine proceeds to the root shift routine from 0A-3 onward, but if it does not hold, nothing is done and this flow is exited and the flow proceeds to the flow of FIG. 31 (the flow of FIG. 30 (A) and FIG. 31 combined is shown in FIG. 26). 26-2 in detail).

The flow from 30A-3 onward in FIG. 30A is for the root shift. In 30A-3, the root note data for the current chord name (CNi) is read from the root note data memory 4. The read root data R is 1
When, the root note of the scale (fundamental note or tonic) and the root note of the chord match, so root note shift is unnecessary.
On the other hand, when the root note of the chord and the root note of the scale do not match, R indicates a value larger than 1, that is, a value obtained by adding 1 to the pitch difference for the mismatch. For example, the root note of the chord is G and the root note of the scale is C.
R in the scale data shown in FIG. 29 is such that the root note (fundamental note) of any scale is C.
= 7. In this case, the root shift processing from 30A-5 to 30A-13 is executed. In the above example (G and C), the weight of the root note in SCL1 is moved to SCL8 indicating the position of G, and similarly, SCL2 → SCL
9, ... SCL5 → SCL12, SCL6 → SCL1,
SCL7 → SCL2, ... SCL12 → SCL8. The root shift function here can be applied to, for example, a jazz mode (mode).

In the case of the operation example of FIG. 9, all measures are in the long scale (Europe), PCs 1 and ₁₄ are zero, and root shift is not performed. For convenience of explanation, it is assumed that there is no root shift in the following explanation. That is, the scale weights are not changed by the flow of FIG.
At the time of entering the flow of FIG. 31, it is assumed that the value of the predetermined scale data in the scale weight data memory 5 remains in {SCLi}. FIG. 31 is a flow of performing weight change having a meaning different from that of root note shift. Here, referring again to FIG. 29, the highest value (weight) of each scale shown is 75.
And the lowest value is 25. Some scales include 50 as an intermediate value. The sound with the highest value of 75 will be referred to as a “scale note”. For example, the major scale (Europe) is
Sounds on the diatonic scale of Do, Re, Mi, Fa, So, La, and Si have 75 points, and others have a low value of 25 points. In general, the western major scales are
The scale data shown in (4) of FIG. 29 simulates this fact.

The meaning of the flow of FIG. 31 will be clear from the above description. That is, in the scale weight changing process of FIG. 31, 100 points are weighted for pitches that are “scale notes” and are also chord constituent notes, and leave only 75 points for simple “scale notes”. It is to keep. Three
The calculation of ((KD _k -1) MOD 12) +1 of 1-7 is a process for obtaining the k-th chord constituent note from the bottom is the note in the scale (selected note scale). , If the weight (SCL _a1 ) of the note of that number is 70 or more (that is, it indicates a scale note), the weight is set to 10
It is increased to 0 (31-8, 31-9). 31
-1 to 31-5 are initialization processing for a new chord, and the scale weight register SCL _k , which had a value of 100 due to the previous chord, is returned to the scale weight of 75. In short, the flow of FIG. 30 (A) is the pitch shift of the pitch that is the reference of the scale (cyclic shift of the array {SCLi}), while the flow of FIG. It is an upgrade of the weight of scaled notes.

★ Optimum turn ★ As an example, F that can be best linked with Domisodo's melody
Consider the distributed chords of.

[Outside 20] In the case of the connection, it is unnatural how to connect Do and Fa before and after the bar line (depending on the song). The most natural connection would be | Domisodo | Dofarad |. In this embodiment, an optimal turning function is provided for optimal connection between such tone patterns. That is, as shown in FIG. 26, the optimum number of rotations is calculated when PC ₁ , ₁₂ > 0 holds (26-3, 26-4). A detailed flow of the calculation of the optimum number of turns is illustrated in FIG. In the figure, PCs _{1 and 10} are parameters (additions to the optimum number of rotations passed from the calculation unit) (32-18, 32-
19). PCs _{1 and 11} are the upper limits of the optimum number of revolutions sent from the parameter C arithmetic unit (32-22, 3).
2-23). LL ₁ is the position information of the harmony sound appearing at the beginning of the previous bar (information indicating the number of the harmony sound from the bottom), LL
₂ is position information of the harmony sound appearing second in the previous bar (see 32-12 and 14). PCs _{1 and 13} are parameters as to whether or not the tone pattern is turned upside down. KDi is the i-th pitch from the bottom of the chord constituent tones in this measure,
The pitch before turning as read from the chord component sound memory 2 is shown.

The first column in FIG. 32 (32-1 to 32-11)
So, I'm looking at which is the closest of the last note of the previous bar (the last note that is not a rest) and the note that composes the chord of this bar (the note before the turn). However, when there are two equivalents, the upper sound is given priority. As a result, the number of the constituent sound from the bottom of the chord constituent sound closest to the last sound of the previous bar is set in the register a ₂ . FIG.
In the second column 32-12 to 32-17 of No. 2, the position information of the first chord of the previous measure (rests are not included here) and the information PC regarding whether or not there is reversal of the note type of the current measure.
Position information a ₁ of the first _chord of this measure from ₁ and ₁₃
Then, the number a ₆ to be turned is found from this information a ₁ (position information after turning the code) and optimum position information a ₂ before turning the code. 33-18 to 32
-24 exceeds the upper limit (PC _{1, 11} ) of the optimum number of revolutions when the optimum number of revolutions (PC _{1, 10} ) is added, when the obtained optimum number of revolutions (a ₆ ) is negative. If so, the optimum number of turns is being corrected. The finally obtained optimum number of revolutions a ₆ is put into PCs ₁ and ₇ . That is, the values of the inversion parameters PC _{1 and 7} directly generated by the parameter calculator are updated here.

[Rotation of chord constituent tones] FIG. 33 illustrates a detailed flow of the chord constituent tones shown in 26-5 of FIG. The function of turning chord constituent sounds is to change and adjust the tone range of the generated melody over time (in this example, in units of measures), and thereby the excitement of the song can be controlled.

In the flow of FIG. 33, for example,

[Outside 21] Then in one turn

[Outside 22] In two turns

[Outside 23] In case of reference, in one turn

[Outside 24] In two turns

[Outside 25] I am trying to become.

[Outside 26] The inversion logic is different between a chord whose both ends have an octave relation and a chord which does not have an octave relation such as Reference (G ₇ ). That is, the pitch sequence of chords before turning is KD ₁ (old), KD ₂ (old),
When expressed by KD ₃ (old) and KD ₄ (old) (low order), for codes that do not have an octave relationship, the sequence after turning is KD ₁ (new), KD ₂ (new), KD ₃ (new). , KD ₄ (new)
KD ₁ (new) = KD ₂ (old) …… (fa) KD ₂ (new) = KD ₃ (old) …… (so) KD ₃ (new) = KD ₄ (old) …… (shi) From (M1) KD ₄ (new) = KD ₁ (old)

[Outside 27] For octave-related codes, KD ₁ (new) = KD ₂ (old) …… (mi) KD ₂ (new) = KD ₃ (old) …… (so) KD ₃ (new) = KD ₄ (old) …… (do) Same as above, but from (M2) KD ₄ (new) = KD ₂ (old)

[Outside 28] Need to be

Whether it is an octave relation or not is seen in 33-3, the distinction between (M1) and (M2) is made in 33-4 and 35-5, shift from 33-6 to 33-10, 33 -1
1 is the finishing of the distinction between (M1) and (M2). PC of Figure 33
_{1 and 7} are parameters that indicate how many times to perform turning,
It takes a value directly given from the parameter C calculation unit or a value updated for optimum turning.

[Generation of pitch string of distributed chords] [0125] Preparations for generation of pitch string of distributed chords are completed from 26-1 to 26-5 in FIG. For example, the range of the distributed chords to be generated is determined, and the pitches of the four chord constituent notes are processed from the bottom of this range.
_{6-5 KD 1, KD 2, KD} 3, are set to KD ₄ through this range, the parameter C computation unit PC _{1, 7} or given directly from runs for optimal coupling of the sound type It is controlled by the PCs _{1 and 7} indirectly given through the optimum number of rotations calculation 26-4. Further, process 26-
Through "2", the normal chord notes are used as the chord constituent sounds this time.
It is given a higher weight.

After the above preparation, 26-6 to 26-4
At 2, the pitch sequence of dispersed chords is specifically generated. For easy understanding, a simplified flow of the processing here is shown in FIG. Several features are included in this process. First, in the PC generated by the parameter C calculation unit, the information (PC
_{1 and 3} ), and from where (P
Range information of the range from (C _1,4 ) to where (PC _1,3 ) should be maintained is also given. On the other hand, the pitch string generator of the distributed chords maintains the tone pattern in that range. Further, when there is an instruction to reverse the tone type (when PC _{1, 13} > 0), the tone type {LLi} is inverted (for the tone type of the previous bar). For example, if the previous bar is the upper row pattern, {LLi} having the lower row pattern is generated.

When the maintenance of the tone pattern is not planned from the beginning, the parameter C arithmetic unit can give the instruction information (PC _{1, 5} > 0) that the sound at the beginning of the measure is determined from the last sound at the preceding measure. (You don't have to do that). On the other hand, the pitch string generator of the distributed chord smoothly connects the beginning sound of the bar to the preceding sound (this means the smooth connection of the sounds by the optimal turn of the chord (the optimum connection of the tone pattern)). Can act as an alternative). Of course, when smooth connection between measures is not required, the parameter C calculation unit attaches a value indicating that to the related parameter C and passes it to the pitch string generation unit of the distributed chord.

Further, the parameter C arithmetic unit issues parameters (PC _{1, 9} ) for controlling the jumping progress of the distributed chords. On the other hand, in the pitch string generation unit for distributed chords,
_{When 1 or 9} is zero, it is interpreted that jumping is prohibited, when 1 is interpreted as one jump is possible, when 2 is interpreted as continuous jumping is impossible, and when 3 or more, jumping is restricted Interpret as none. The pitch string generation unit inspects the presence or absence of jumping and uses the flag for confirmation in order to keep the instruction from the PCs _{1 and 9} . The parameter C calculator may allow the generation of distributed chords by random numbers (when the tone pattern is not maintained). However, some restrictions are also given so that completely random distributed chords are not created (for example, jump upper limit PC ₁ and PC ₂ and homophone progression PC ₁ and PC ₆ ). On the other hand, the pitch string generation unit for the distributed chord generates a pitch string satisfying the predetermined constraint, that is, a pitch string controlled by the PC.

Further, when the pitch string generator for the distributed chords obtains the distributed chord candidates, it refers to the note scale data {SCLi} to check whether the sound is a valid sound, Take out the weight. The parameter C computing unit issues a parameter (PC _{1, 1} ) for the threshold value of the effective pitch, and when the extracted weight is equal to or greater than this threshold value, that sound is a valid sound, that is, a distributed chord. Adopted as the melody note of. The work of the pitch string generation unit of the distributed chord is completed by generating the pitch string as many as the target distributed chord number (the value of PC given by the PC calculation unit).

As can be seen from the above description, the parameter C arithmetic unit assigns planned values to various parameters C and delivers them to the pitch string generator of the dispersed chords in order to generate the dispersed chords. On the pitch string generation unit side, various parameters C passed from the parameter C calculation unit are “decoded” and distributed chords are generated according to the decoding result. The distributed chord pitch sequence generation unit has a rule for generation of a pitch sequence along the planned parameter C (similar relationship is also established between the parameter C calculation unit and other melody generation execution units). Exist).

In 26-6 to 26-42 of FIG. 26, i is a note number, fl and flb are flags for controlling jumping progress, and NF is SLC _a1 <PC at 26-36.
It is a flag to show that it was _{1, 1} (invalid pitch). PC _{1, 9} = 2 means continuous prohibition of jumping (26
-8). fl = flb indicates that the previous jump was not a jump (see 26-38 and 26-39). 26-16
26 to 28 are randomization LLI determination processing. The above flag NF also has the meaning of avoiding an infinite loop operation including this randomizing process (36-3).
7, 26-19, 26-23, 26-27). Further, the pitch valid / invalid check (26-36), that is, the limitation by the weight control parameters PC _{1, 1} is prioritized over the instruction content indicated by various kinds of jump control parameters PC ₁ , ₉ . 26-38). This is 4
Considering that when one chord constituent sound {KDi} is limited by every other scale weight (PC ₁ , lower than ₁ ), it conflicts with the limitation of jumping progress (PC _{1, 9} ). Is. The details of the process of determining from the sound before 26-15 are as shown in FIG. That is, of the chord component sounds {KDk}, the effective pitch is (36-
6 is true), the sound with the smallest difference from the previous sound MED _i-1 (see 36-3, 34-5, 34-7, 34-8)
Is the chord sound from the bottom of the chord constituent sound
It is set to Li (actually LL ₁ ).

The above description and 26-6 to 26- in FIG.
From the explicit description of the flow itself of 42, the operation for the concrete generation of the note length sequence of the distributed chord is clear. Therefore,
Further description will be omitted.

[Generation of Sound Length Sequence of Distributed Chords] In this embodiment, the process of generating the tone length sequence of the melody is basically executed independently of the process of generating the pitch sequence of the melody. More generalized, given the number of notes and the total length of those notes (one bar in this example), some kind of note length generation rule
A note length sequence having the number of notes can be generated. In the case of this example, the number of distributed chords in each measure is planned and issued in the parameter C calculation unit (initial value of PC in FIG. 5). Further, as described above, at the time of forming the tone length pattern of the dispersed chords of the motif (FIG. 15), the non-harmonic sounds included in the motifs are removed, and the tone lengths are absorbed by the harmony sounds. A sequence of sound lengths is formed and extracted. This tone length sequence will be referred to as a motif distributed chord pattern.

The number of distributed chords planned by the parameter C calculator depends on the progression of the piece of music. Therefore, the number of dispersed chords in a certain melody bar is equal to the number of dispersed chords in the motif bar, or is larger or smaller than the number of dispersed chords in the motif. If the number of distributed chords in a melody bar is equal to the number of distributed chords in a motif bar, what sequence of dispersed chords is preferable as a distributed chord pattern for that melody bar? In this embodiment, the motif distributed chord pattern is used as the distributed chord pattern of the melody when the numbers match. This is to strongly reflect the rhythmic characteristics of the motif in the melody. Of course, if desired, a pattern obtained by modifying or transforming the distributed chord pattern of the motif may be used as the distributed chord of the melody (an example of a means for realizing this is described in another section). In this embodiment, consideration is given to the tone length sequence that faithfully maintains the rhythmic characteristics of the motif.

Next, there is a problem in that the number of dispersed chords in the melody bar and the number of dispersed chords in the motif bar do not match. On the other hand, in this embodiment, the distributed chord pattern of the melody bar is created by the minimum number of times of sound length division or sound length combination. It is an approach that emphasizes consistency rather than freedom. That is, a pattern with the smallest degree of change from the initial pattern (a pattern that is closest to the motif distributed chord pattern if the motif distributed chord pattern is the initial pattern) is obtained by the minimum number of division and combination operations. Furthermore, in the present embodiment, note length division and combination are executed by the same pulse scale as the pulse scale (see FIG. 16) used when obtaining the motif distributed chord pattern.
As a result, the characteristics of the pulse scale will be generated, but the pattern of the rhythm related to the pulse scale will be maintained, rather than the length pattern changing irregularly. Changes. In short, well-controlled pitch pattern conversion is performed.

The generation of the tone length sequence of the distributed chord is indicated by 26- in FIG.
44 (although it may be done at other times). Details of this processing are illustrated in FIGS. 36 to 41. In FIG. 36, i is the note number within the bar, RHi is the length of the i-th harmonic sound of the motif, and MERi is the i of the melody being generated.
Th tone length (meaning the tone length of the i th chord here), PC is the number of distributed chords assigned to the current measure (the number of distributed chords planned in the parameter calculation unit), PA
Numbers _{1 and 3} indicate the number of dispersed chords included in the motif bar, and S indicates the difference between the numbers of both dispersed chords. At 36-1 to 36-4, the motif distributed chord pattern {RHi} is used as the initial pattern of the melody distributed chord pattern {MERi}. S =
If 0, that is, if the numbers match, this motif distributed chord pattern becomes the distributed chord pattern of the melody, and the work is completed.

When S> 0, that is, when the number of distributed chords of the melody is smaller than the number of distributed chords of the motif, optimal combination processing (FIG. 37) is executed, and when S <0, that is, the distributed chord of motifs. When the number is smaller, the optimum division process (FIG. 38) is executed. In the optimal combination of notes, (a) the note length is combined a minimum number of times, and the number of notes is the planned value PC.
When it reaches, the process is completed. (B) A note starting with a pulse point having a low value on the pulse scale (FIG. 16) is preferentially combined with the preceding note. Is adopted.

For example, referring to FIG. 16, of the positions 0 to 15, the next note starts at positions 1, 3, 5, 7, 9, 11, 11, 13 and 15 (positions having a weight of 1). Is checked first, and if there is such a next note,
Combine that note with the previous note (unless there is one). If the target number PC is not reached, then it is checked whether or not there is a note starting at the position where the weight is 2, and if there is a note, the note is combined with the previous note (the note currently being focused on). If the target number PC is not reached, the presence or absence of a note starting at the position where the weight is 3 is checked, and the note is combined with the preceding note in the note length. Since the same applies to the following, description will be omitted. In short, the logic is to preferentially combine a note starting at a pulse point with a light weight with a previous note. In FIG. 37, SUM represents the start position of the next note (the length from the bar line to the note). VT is an array {MER
j}, that is, the current number of distributed chords (more precisely, the current number of distributed chords minus 1, the number of distributed chords to be achieved next). The reason i> 4 is satisfied in 37-9 is that
This is the case for whole notes. 37-11 is the note length ME of the next note
This is a process of combining R _{j + 1} and the note length MERj of the current note. 37-12 to 37-16 are shifts of the array {MERj}. The meaning of 37-5 is, for example, when j = 1, it is a judgment of whether or not the next note starts with a weight of 1 on the pulse scale of FIG. To the execution routine of. When j = 2, it is checked whether the next note starts at the position of weight 2 (pulse point). When j = 3, it is checked whether the next note starts at the position of weight 3, j. When = 4, it is a check whether the next note starts at the position of weight 4. When the number of distributed chords reaches the target number PC, S = 0
Is established and is completed (37-17).

On the other hand, in the optimal division of notes, (a) the note length is divided by the minimum number of times and the number of notes is the planned value PC.
When it reaches, the process is completed. (B) A note crossing a heavy pulse point is preferentially divided with the pulse point as a boundary. Is adopted.

According to the flow of FIGS. 38 and 39 to 41, first, the weight 3 on the pulse scale (FIG. 16) is set.
A note crossing the pulse points having the above is searched, and each time it is found, two notes are divided and generated using the pulse point as a dividing line. When there are no notes that cross the above pulse points, or when there are no notes that cross the above pulse points as a result of division, but the planned number PC is still not reached, a pulse point with a weight of 2 is selected. Search for the notes that cross, and perform the division in the same way.

In the check routine of FIG. 39, a note crossing the pulse point in question is searched. If found, the note number is written down in the flag fl, and the start position information of the note is written in the flag ff (see 39-8 to 39-10). 39-1 f
Since l is reset to zero, if there is no note that crosses the pulse point in question as a result of the search, fl = 0 holds in 38-4 of the main flow (FIG. 38),
It shifts to the check of the note that crosses the pulse point having a lower weight (note that j = 2 in the main flow shown in the figure).
Although only up to (at the pulse point of weight 2) is investigated, strictly speaking, it is necessary to perform until the achievement of the target number of dispersed chords PC is guaranteed. Consider the case where the parameter C calculation unit plans an extremely large number of distributed chords.
The modification of the flow for this is very easy. ).

As a result of the check (FIG. 39), if a note straddling the pulse point in question is found, the note length array {MERj} is shifted to lengthen one note (FIG. 4).
0). Then, the division is performed as shown in FIG. That is, the fl-th note to be divided is divided at the pulse point where the note crosses as a boundary. From the above description and the clear description of the flow itself in FIG. 36 to FIG. 41, the operation of the flow regarding the sound length determination, that is, the formation of the sound length sequence of the distributed chords is clear, and therefore further description will be omitted. With the explanation up to this point, the pitch string and the pitch string of the distributed chord are completed. FIG. 9A is an example of generation.

Addition of non-harmonic sound After non-harmonic sound is generated, non-harmonic sound is added. Regarding addition of non-harmonic sounds, there are problems such as "whether or not to add", "where" to add, and "what pitch" and "what length" of non-harmonic sound. As described above, in this embodiment, the parameter C calculation unit plans the melody. Therefore, the parameter regarding the non-harmonic sound is also calculated by
It is created in consideration of the feature extraction result of the motif and the progress of the song. The created non-harmonic sound related parameters are passed to the non-harmonic sound addition execution unit. The non-harmonic sound addition execution unit
It is equipped with logic or rules for decoding the sent parameters and adding non-harmonic sounds according to the plan. This rule is a rule for solving the above-mentioned problem regarding addition of non-harmonic sounds. The rule for adding non-harmonic sounds in this embodiment changes various addition results depending on the value (data) of the parameter C. In this sense
It is a data driven rule. However, whatever the value of the parameter C, the addition of "illegal" non-harmonic sounds is never done. In other words, the data (parameter) does not have the power to destroy the rule.

In the illustrated example, the addition of non-harmonic sounds is performed for each type. Addition of sound (Fig. 42) and addition of progress sound (Fig. 43)
And embroidery sound addition (FIG. 44) and missing sound addition (FIG. 45). Here again, the names of "sounding", "progressive sound", "embroidering sound", and "missing sound" are for convenience. For example, the sound addition is performed when a predetermined condition is satisfied under the constraint of the parameter C. In other words, it can be said that each non-harmonic sound lower part performs inference to add a non-harmonic sound. That is, the given situation is a pattern containing only chords (a melody containing only distributed chords), and the addition execution unit
The "hidden" non-harmonic sound is found using the pattern of only the harmony sound as a clue and according to the plan from the parameter C calculation unit. Such inference is basically the reverse of the inference when extracting and removing nonharmonic sounds from a melody containing nonharmonic sounds (the former is A → A + B).
The latter is the type of A + B → A).

The rules for adding each non-harmonic sound to be described below are merely examples, and those skilled in the art can easily invent other rules according to the teaching of the present invention.
For example, the number of notes is increased by confirming the addition of the non-harmonic sound (however, in this example, the missing sound is treated as an exception). On the other hand, in the following non-harmonic sound addition flow, the sound length of the note (harmonic sound) before or after the non-harmonic sound to be added is cut, and the cut-off portion is set as the non-harmonic sound length. As a result, a note length sequence with an increased number of notes is formed. Alternatively, the pulse scale division described above is possible. Before reading the flow instructions for adding each non-harmonic note, it is helpful to look at FIG.

★ Addition of spells ★ FIG. 42 shows a flowchart of addition of spells. In the case of this flow, the following conditions must be satisfied in order to add the sound. (i) The parameter PC _2,2 of the weight of the sound from the parameter C calculation unit permits addition of the sound (PC _{2, 2} > 0)
42-1). (ii) The note next to the onset is not a rest (42-5). (iii) The next note is longer than the shortest allowable note length indicated by the parameters PC _{2 and 7} (42-6). 42-2 to 42-4 are determining the note numbers to which the notes are added, and the information thereof is set in the register a ₁ . 42-7 to 42-11 are array shifts of the pitch sequence {MEDi} and the tone length sequence {MERi} of the melody (securing a place for the sound data). 42-12 to 4
2-14 is deciding whether to add the sound on the upper side (a ₂ = 1) or on the lower side (a ₂ = -1) according to the instruction information PCs _{2 and 4} . 42-15 to 42-21 are pitch determination processing of the low sound. Here, the valid note found at the end, starting from the note 4 degrees (K = 5) above or below the pitch of the next note (the note following the note), and gradually narrowing the pitch by half steps. A high pitch (SCLa ₄ > PC _{2, 1} is satisfied) is set in MEDa ₁ as the pitch of the low sound. 42-2
From 2 to 42-16, the length of the part of the sound from the next note is determined. At 42-27, reduce the note length of the next note by the amount of the onset, and set it to MERa ₁ , ₁ .
The tone length of the 42-28 sound is set in MERa _1, and the counter PC for the equal number of sounds is incremented at the last 42-29. FIG. 9B shows an operation example of adding a sound.

[Addition of Progressive Sound] FIG. 43 is a flowchart of addition of progress sound. In this example, after the jumping progress, the device is devised so that the progress is likely to occur sequentially (43-2 to 43-16, 43-27,
43-30). The value of r ₁ indicates a scale to which the progress sound is easily added. 43-2 is a random calculation of r ₁ based on PC ₃ and ₂ . When the front is jumping (a ₁ ≧ PC
_{When 3 and 4} are satisfied), a high value (here, 5) is forcedly inserted into r ₁ .

In the flow shown in the figure, the necessary conditions for the addition of the elapsed sound are as follows. (i) There is a certain pitch difference before and after the elapsed sound (43-7, which is wider than the whole sound here). (ii) Neither rest before nor after (43-7, 43
-8). (iii) Both the preceding and following sounds have a certain length or longer (43-32, 43-3). Other conditions are also required, but this condition is 4
3-15 to 43-24 depending on the result of the search or check of the candidates for the elapsed tone. 43-10 before that
Reference numeral 43-14 is a preparation for the search. For example, if the rear note is higher than the previous note (if it is traveling upward), the note that is one semitone above the previous note is the first candidate for the passage note, and if it is descending, A sound that is a semitone lower than the preceding sound is set as the first candidate for the passing sound. In the search, the candidate is shifted by a semitone each time the loop is rotated once (43-24). 43-10 to 43-14
In the search of, the number of candidates having effective pitch is counted in SUM. However, if a pitch that is a harmony sound is included between the preceding sound and the following sound, a special value is set in SUM (to prohibit addition of elapsed sounds). The effective pitch test (43-15, 43-1) is included in the array {b _SUM }.
Each pitch data which passed 6) is written. The search is repeated until there is a semitone difference from the back tone.

Here, an example of a condition in which the addition of the elapsed sound is prohibited as a result of the search will be illustrated (condition of failure). (B) In the search, no valid pitch was found on the chromatic scale between the front note and the back note. That is, there is no effective pitch on the chromatic scale from the front to the back (see 43-25). (B) Harmonic tones are included on the chromatic scale from front to back (43-21, 43-22, 43-2
5). (C) Too many (here, 3 or more) valid pitches are included on the chromatic scale from the front note to the back note (43-16, 43-29 indicates skip this time)

[Outside 29] Exit). Positive conditions, i.e., the remaining conditions for the addition of elapsed sounds, are described in 43-25 to 43-31. That is, the (V) chord should not be included on the chromatic scale from the front to the back. (VI) There must be one or two effective pitches. Then, (a) one is larger than a certain value (here, r ₁ > 1) of the measure of ease of addition, while (b) two is a certain value of the measure of ease of addition ( Here, r ₁ > 2). When one, a ₂ = 1 and it is shown that there is one non-harmonic sound (progressive sound) to be inserted between the preceding sound and the following sound,
When there are two, a ₂ = 2, and it is determined that there are two elapsed sounds inserted between the front sound and the back sound.

The third column, 43-34 to 43-39, is for determining the note length. 43-40 to 4 in the third row
Up to 3-43, an array {MED
i} and {MERi}. When a ₂ = 1 (additional number 1), in 43-44, the already-determined pitch data and duration data are stored in the registers MED _{i + 1} and MER _{i + 1} for the notes of the elapsed tone. Insert and rewrite the note length register MERi of the previous note. That is, the sound length of the elapsed sound at one time is obtained by receiving a part of the length of the preceding sound. After 43-44, 43-45 fails and 43
Go to -47 (note number counter processing, etc.). a ₂ = 2
In the case of (addition number 2), the sound length and pitch of the first elapsed sound are set at 43-44, and then the sound length and pitch of the second elapsed sound are set at 43-46. The length of the second elapsed note is received from the back note.

* Embroidering sound addition * FIG. 44 shows a flowchart for adding an embroidering sound. Since the flow (FIG. 43) has been carefully described in the preceding addition of the elapsed sound, the meanings of the symbols in the flow may be unnecessary except for exceptional ones. Here, a brief description will be given. In order for the embroidery to be added, the pitches of the front and back notes must be the same (44-2 to 44-).
4). The length of the embroidery sound is to be received from the preceding sound (44-6 to 44-8, 44-31, 4).
4-32), the preceding sound must have a certain length or longer (44-5). The parameter C calculator sends out parameters PC _{4 and 4} for the addition of embroidery sounds. When PC _{4 and 4} are zero, it means that embroidery sounds cannot be added. When 1, PCs _{4 and 4} can be added only once. Ban. If it is 3 or more, it can be added freely. In order to satisfy the request from the parameter C calculation unit,
The processes shown in -9 to 44-14 are prepared. Further, the parameter C calculation unit issues an instruction to put the embroidery sound up or down according to the values of PC _{4 and 2} . Measures against this are taken from 44-21 to 43-23. 44-16 to 44-20 are arrays {MEDi} and {MERi} shifts for embroidering sound addition. Also, 44-24 to 44-30 are pitch embroidery sound pitch determination processing (processing similar to 42-15 to 42-21 described above).

* Addition of Missing Sound * FIG. 45 shows a flowchart of adding the missing sound. In the flow of this example, the rule that the number of notes does not change is adopted in the case of a missing note, and attention is paid to the pitch progression of three consecutive notes. Also, the missing sound is supposed to be the last sound in the bar. Therefore, based on the pitch difference between the penultimate sound of the previous measure and the beginning sound of the next measure, the pitch of the matching missing sound is calculated, and the last of the measures already generated at that pitch is calculated. The sound is being changed (pitch replacement). When the prohibition of the addition of noise is instructed by the parameter C calculation unit (PC _{3, 2} ≤ 0
In case of), follow this instruction (45-1). Further, since the missing note addition is a process for three consecutive notes (notes that are not rests), addition is not performed when the preceding and following notes are rests (45-2, 45-3). Due to the difference in the pitch progression of the three sounds, the missing sounds are divided into three types. It is shown below the flow chart. (i) embroidery type, the end of the previous measure and second sound MED _-1 and MED ₁ beginning of the next measure is generated when the homophone. (ii) Progressive types are MED _-1 and MED
_This occurs when there is an effective pitch (a sound that has passed the pitch effective inspection 45-11, 45-12) on the chromatic scale between ₁ and ₁ , especially when there is only one in the illustrated flow. (iii) The low tone type occurs when there is no effective pitch at all when MED _-1 to MED ₁ are advanced in semitone steps.

The number of effective pitches between the two tones MED _-1 and MED ₁ is 45-4 to 45-15, (especially 4
5-9 to 45-15). The number of effective pitches found in fl is counted. When the number of effective pitches is 2 or more, it is not added (45-16). All of the above three molds have an upper mold and a lower mold. For example, in the case of the embroidery type, the anomalous sound is divided into two, depending on whether it is higher (upper) or lower (lower) than the homonyms MED _-1 and MED ₁ before and after. The processes related to this are 45-17 to 45-19 (P
Determined by the values of C _{5 and 3.} ) MED _{-1 for} transitional type
From MED ₁ to MED ₁ is divided into two depending on whether the pitch progresses upward or downward. The same is true in the case of the low noise type. However, in the case of the low-pitched sound type, there is no effective sound (usable sound) on the chromatic scale from MED _-1 to MED _1, so it is escaped. That is, MED _-1 to MED ₁
Progress to

[Outside 30] If so, a sound higher than both sounds is selected, and MED _-1 to ME
Progress toward D ₁

[Outside 31] In that case, a sound lower than both sounds is selected (4
From 5-4-45-7 and 45-23,

When [outside 30], a ₇ = 1,

When [outer 31], a ₇ = −1. Further 45-24
After that, the effective pitch is searched in semitone steps from the pitch shown by (MED ₁ + 5 × a ₇ ). ).

These 45-24 to 45-30 are shown in FIG.
The same processing as 44-24 to 44-30 of No. 4 is performed. Here, it is effective while starting from the pitch 4 degrees above or 4 degrees below the beginning pitch MED ₁ of the next bar (determined by a ₇ whether it is above or below) and moving closer to the beginning pitch MED ₁ by a semitone. Search for sound. The last valid sound found, that is, the valid sound closest to the pitch of the beginning of the next measure is determined as the lost sound. Although the effective sound is searched by the decrement method (pitch of a certain sound, here, a method of approaching the first sound), the effective sound may be searched by the increment method. In this case, the first valid sound found is determined as the sound to be added. I need to get out).

Adding Rests FIG. 46 is a flowchart of adding rests (breaths). In this example, 16-minute rests are added when possible (46-2 to 46-4). PC _9-1 of 46-1 is a parameter of whether or not rests can be added, which is given by the parameter C calculation unit. Typically, the parameter C calculation unit sets PC9, ₁ to the permissible value at the last measure of the passage.

Generation of Characteristic Rhythm As described above, in the automatic composer of the present embodiment, the rhythm evaluation section extracts what kind of mini rhythm pattern and to what extent the motif includes, and the characteristic rhythm parameter. In the extraction section of, which mini-pattern is the mini-pattern that characterizes the motif is evaluated. Then, based on this evaluation result, the parameter C calculation unit systematically generates a mini-pattern scale (melody rhythm control information) that characterizes each measure of the melody. According to the melody rhythm control information given by the parameter C calculation section, a section for injecting a mini pattern into the tone length sequence of the melody is a characteristic rhythm generation section (feature pattern incorporation section). The function required for the characteristic rhythm generator is the parameter C.
It is the ability to decipher the plan indicated by the melody rhythm control information sent from the arithmetic unit and change the tone length sequence of the melody in accordance with the plan. FIG. 47 shows an example of the flow executed by the characteristic rhythm generation unit. In this flow, one parameter PC _6, 1 is used as the melody rhythm control information.
I'm just using This is for convenience of description, and a plurality of parameters for controlling the melody rhythm may be passed from the parameter C calculation unit to the characteristic rhythm generation unit, and the characteristic rhythm generation unit decodes those parameters, The functions may be combined so as to generate a rhythm according to the decoding result.

Parameters PC _{6, 1} (47 in FIG. 47)
The meaning of (see -1) is as follows. First, when PC _6,1 is negative or zero, mini-pattern injection is prohibited. When PC _{6, 1} is 1, injection is allowed only once.
When 2, continuous injection is prohibited. When it is 3, it is allowed to inject the mini pattern freely. Further, the parameter PC _6,1 has the following meaning in relation to the rhythm-related processing described above (for example, FIG. 19, PA ₆ , ₁ ). That is, it is related to the mini pattern with a tone length ratio of 3: 1. That is, in the systematic sense, the parameter PC
The sizes of ₆ and 1 indicate the degree of appearance of a mini pattern having a note length ratio of 3: 1 in a melody bar. Therefore, the flow of FIG. 47 depends on the value of PC _6,1 and is 3
It has an algorithm to generate mini patterns with a tone length ratio of 1: 1.

Furthermore, the characteristic rhythm generation process shown in FIG. 47 is the final process of melody generation. Before entering this process, the distributed chord generation process and the non-harmonic sound addition process are completed. ing. In other words, in the process of generating distributed chords, a sequence of tone lengths of distributed chords (melody distributed chord pattern) is created based on a motif distributed chord pattern in which non-harmonic sounds are eliminated from motifs, and hidden in the process of adding non-harmonic sounds. The non-harmonic sound that has been recorded is inferred and determined, part of the preceding or following harmony sound is trimmed, and the length of the trimmed portion is used as the non-harmonic tone length to make a minimum change. There is. Then, a conversion process from a motif containing a non-harmonic sound to a motif chord pattern in which the non-harmonic sound is omitted, and a melody containing a non-harmonic sound from the melody distributed chord pattern in which the non-harmonic sound is omitted are performed. The conversion process up to formation is formally in a reverse inference relationship. Therefore, at the stage of entering the process of generating the characteristic rhythm of this example, the pitch sequence of the melody is already completed. In addition, there are many cases where the sequence of melody lengths is almost complete. Therefore, this last process should be done carefully. In other words, if you forcibly create a 3-to-1 note length pattern by shaving or increasing the note lengths of some notes, this powerful note length pattern It is highly possible that a ridiculous length pattern is generated as a by-product due to the formation of the. That is,
Originally, a rhythm that should not exist is born. In consideration of such a point, in the characteristic rhythm generation of this example, the mini-pattern according to the plan of the parameter C calculation unit is injected while the rhythm of the already-completed tone length sequence is not impaired as much as possible. ing. In a word, it has the rule of the minimum transformation of the given sound length sequence.

More specifically, in the characteristic rhythm generation of FIG. 47, tone length conversion, that is, a tone length ratio of 3: 1 is achieved only for tone length pairs that satisfy the limited conditions. Is being converted. This condition is as follows. (i) The duration of two consecutive notes is equal (47-
9) (ii) The note length is 8 minutes or longer (47-10) (iii) The note length is an integral multiple of 8 minutes (47-1)
1) (iv) Two consecutive notes (to be exact, the preceding note) starts at a position that is a multiple of 4 (including 0) (47-12). Satisfied the above conditions, and PC values and related conditions _6,1 when (planning conditions PC _6,1) is satisfied, the above-mentioned two consecutive notes tone length ratio of 3: 1 of the sound It is converted to a long ratio (47-13 to 47-19). The condition (iv) is such that syncopation does not occur (other than that, it has no special meaning).

The processing from 47-5 to 47-8 is performed by the PC.
_{6 and 1 This} is a process to keep the constraint indicated by each value. Flag f
l is when PC _{6, 1} is 1 (when injection is possible only once) and 2
It is used for (when continuous injection is prohibited).
At the time of injection, the flag fl changes from 0 to 1 (47-1
4). When PC _{6, 1} = 1, after injection of 1 time and it becomes fl, it shows completion after going from 47-5 to 47-6 and 47-7.

[Outside 32] Exit the flow at. In case of PC _{6 and 2} , f by injection
When l = 1, 47-8 from 47-7 on the next pass
Through,

By jumping to the next sound processing as indicated by [Ex. 29], the injection in the next pass of fl = 1 is prohibited. The illustrated flow is merely an example, and those skilled in the art can incorporate a mini pattern characterizing a melody into a note length sequence according to other rules. By the way, in the operation example of FIG. 9, the operation of generating the characteristic rhythm ends with NOP. In the case of the melody generation shown in FIG. 25, the above-described processes of generating distributed chords, adding non-harmonic sounds, adding rests, and generating characteristic rhythms are executed in bar units. As illustrated in FIG. 9, the motif is 1 bar and the melody to be automatically generated is 7
In the case of a bar, the music is completed when these processes are executed 7 times.

<Modification Learning> In the present embodiment, the user can request the modification of the completed music piece from the automatic composer through the monitor (not shown in FIG. 2). The correction learning will be described below. In this embodiment, the correction is made for each bar. FIG. 48 is a flow of correction learning. First, in the same manner as described above, the characteristic parameter PA is extracted from the motif (48-
1). Here, the parameter C or the like may be generated and stored for all measures of the generated melody (the finished melody is the same as before), but this is not preferable in terms of storage capacity. The user is asked to input which bar should be modified (48-2). On the other hand, the automatic composer activates the internal parameter C calculation function to calculate all the parameter C (PC _x ) of the bar for which correction has been requested (48-3).

Next, there is a parameter conversion from objective data or internal expression data (parameter C) to subjective data or conceptual data for users (4).
8-4). The purpose of this parameter conversion is to provide information that is easy for the user to understand and judge. For example, since the turning parameters PC _{1 and 7} mainly control the swell, when PC _{1 and 7} = 2, the degree of swell is 90% in this measure, It is easier for the user to make a subjective judgment, and it becomes easier for the user to make a correction. The result (message) of parameter conversion is CRT12 (Fig. 1)
Etc. (48-5). On the other hand, the user inputs the parameter type EDB and the correction value EDC '(48-6, 7). For example, if the user wants to change the excitement from 90% to 50%, the user inputs that the type of the correction parameter is excitement with 48-6, and sets 50% (EDC ') of the correction value to 48. It is input with -7.

On the other hand, the automatic composer inversely converts the subjective parameter value (EDC ') into the objective parameter value (48-8). In the case of the above example, the inverse conversion is executed for the request to increase the excitement to 50%, and the turnover parameter PC
_Set the value of _{1 and 7} (EDC) to 1. Then, as shown in 48-8 to 48-9, the correction contents are written in the learning data memory 9. Here, P is a pointer of the learning memory 12. The pointer P is incremented and the correction data is written one after another. When the bar to be modified, the type of the parameter to be modified, and the value are no longer present, the user inputs the completion of modification, and the modification learning process ends (48-12). The above-mentioned correction learning works so as to reflect the user's preference in the melody that is generated. This will be further clarified from the operation of parameter change by learning illustrated in FIG. 49.

That is, at the parameter change by learning, the parameter C that has been learned by the correction learning is given priority over the parameter C that is normally generated by the parameter C calculation unit. 49 is a normal calculation of the parameter C at 49-3. 49-4 to 4
In the portion 9-13, the learning memory 12 is searched and the parameter C used for melody generation is replaced with the learned one. That is, a modified measure that matches the current measure (value of i) is found (i = * Pa), and a parameter type that matches the focused parameter type is found (j = * (Pa + 1)). And the correction data EDC
Becomes the value of that parameter PCj in that measure (PCj = EDC). In short, once learning memory 1
The learning data stored in No. 2 is read out every time the user requests a composition, and the learned data is directly reflected in the parameter C. The parameter C controls the generation of the melody. Therefore, the composed melody reflects the taste of each user, and the learning result appears here.

<Characteristics of Embodiment> From the above detailed description, the characteristics of the automatic bending machine according to this embodiment are clear. Some of them are listed below. (B) Since the melody is generated based on the motif feature parameters obtained by evaluating the motif and chord progression information, the essence and concept of the song's motif are reflected throughout the song itself and controlled along the chord progression. And a melody that can be changed in various ways is created. (B) A means for generating a parameter C (melody characteristic parameter, melody control information) for controlling the melody to be generated is provided in the melody generating means, and the parameter C for each measure is set in units of the progress section (measure). Have decided. The melody of each measure is controlled by the value of the parameter C assigned to each measure. Therefore, it is possible to express unity and variety in the flow of the melody. (C) The means for generating the parameter C uses the compressed data (parameter B) and the motif feature parameter,
The parameter C is generated by the calculation. Therefore, various values of the parameter C can be obtained with a relatively small storage capacity. (D) Further, the generating means (melody control information generating means) for the parameter C includes a musical expression identification data generating means, whereby a higher-order hierarchy can be imparted to the music. (E) Since the non-harmonic sound and the harmonic sound are treated separately, the flow of the music becomes very musical. (F) Motif is designed to be input by the user, and the song can be created in a way that reflects that motif, so the user can get a sense of participation in the composition and a sense of satisfaction at the same time. . (G) Regarding the reflection of the characteristics of the motif in the melody, it is controlled according to the characteristic elements of the motif, and the space for composing is very wide. (H) In terms of rhythm, characteristic mini-patterns included in the motif are extracted, and characteristic mini-patterns are systematically incorporated into the melody based on the result. Therefore, the rhythmic feature of the motif can be reflected in the melody to any degree. (I) Further, regarding the formation of the sound length sequence of the distributed chords, means for dividing the sound length by the pulse scale a minimum number of times and combining the sound lengths is provided. The rhythmic features not present in the motif may appear in the melody due to the influence of the pulse scale, but the changes are consistent and achieve well-controlled length pattern conversion. (Nu) A pitch control means is used for the effectiveness of the melody note candidates. It is possible to obtain an effective note scale that far exceeds the types of scales provided in memory. For example, it can be achieved by controlling the threshold value. (L) Also, depending on the parameter settings, it becomes a distributed chord (arpeggio) creation device. (Wo) An artificial intelligence function is included, and evaluation and analysis of motifs (conversion from motif to parameter) and conversion from parameter to melody are performed by forward inference and backward inference. (W) Furthermore, no specialized musical knowledge is required, and if you can think of even a few motifs, the automatic composer will compose accordingly. (F) A correction learning function is built in, and this function learns the user's preference. The automatic composer prioritizes the learning data acquired by this learning function and performs subsequent composition. Therefore, the composed song will reflect the user's preference and the user will be more and more interested.

<Modification, Development, Application> The above embodiment is only one aspect of the present invention. The modification, development, and application will be described below.

[0131] Ward During First, in the above embodiment are any measures same length the length of the bar, but may also be a variable length bar. This can be realized, for example, by providing a bar counter and using the bar length assigned to the count value (bar number). Further, in the above-mentioned embodiment, it is assumed that the bar of the motif is given as the beginning of the music, but any bar may be used as the input bar of the motif. Modification for this is easy. Further, in the above embodiment, the length of the input motif was assumed to be one bar, but it may be a plurality of bars. For this purpose, for example, in the case of two measures, a first motif feature parameter (for example, a phonetic parameter LLi) is extracted from the motif of the first measure,
A second motif characteristic parameter is extracted from the motif of the second measure, the first parameter C is calculated from one of the two types of parameters B and the first motif characteristic parameter, and the second parameter of the other parameter B is calculated. The second parameter C is calculated from the motif feature parameter of. The first parameter C is used, for example, to control the melody generation of odd-numbered measures, and the second parameter C is used to control the melody generation of even-numbered measures (measures where modulo 2 is zero). Use (Set the bar number counter value to 2
If so, it is possible to immediately determine which parameter C should be generated). However, depending on the music, for example, the format is A, B, A, where A is 8 measures and B is 7
In the case where the measure is a measure, the last A is a measure of 8 measures, for example, the value of each measure (8, 7, 8) = (the number of measures of the first measure, the number of measures of the second measure, the third) The number of measures in the measure) is compared with the value of the measure number counter, and when the measure number becomes 9 and it is the ninth measure from the beginning of the song, that is, when the start of the second measure is detected, the measure number You can reset the counter. In short, it suffices to generate the first parameter C in the odd-numbered bars and generate the second parameter C in the even-numbered bars in each phrase.

In the above embodiment, the number of chords per one bar is assumed to be one (one chord / bar), but the number of chords per one bar can be two or more.
For example, assume a bar where the first two beats are the first chord (eg C) and the second two beats are the second chord (eg F). In the first configuration example, the parameter C generation function (melody control information generation unit) generates the parameter C in units of bars. On the other hand, the distributed chord generating function generates a distributed chord according to the first chord in the section of the first chord, and generates a dispersed chord according to the second chord in the section of the second chord. However, the parameter C is used in the bar section. The remaining non-harmonic sound addition function adds non-harmonic sounds in units of measure sections. In the second configuration example, the non-harmonic sound adding function also adds the non-harmonic sound in units of two beats (chord unit). Other configuration examples are possible. If desired, the note length correction function may be changed to include a function of correcting the total note length of the melody in chord units to the chord length (or a length close thereto).

Furthermore, a function for selecting a musical style may be added. For example, the entire data of the parameter B is classified.
That is, the data structure of the parameter B memory is made into a classified structure. And by selection input from the input device,
The data to be read from the parameter B memory is determined, and the parameter C is generated by these selected parameters B. As described above, the parameter C depends on the parameter B, and if the value of the parameter B changes, the value of the parameter C also changes, and as a result, the melody of the generated song changes. Moreover, the data structure of the chord component sound memory 2 related to the chord progression can be a structure classified based on the wind style. From the selected set of chord component sounds, each chord component sound is read according to the chord name sequence indicating the chord progression on the chord progression memory. As a result, the dispersed chord generated by the dispersed chord generating function is different from the case where the dispersed chord is based on another set of chord constituent tones. As a result, the characteristics of the melody also change.

Scale Relation (Used Pitch Control) Next, a modification of the used pitch control function relating to the generation of the melody pitch sequence will be described. As described in the embodiment, the used pitch control function is composed of note scale generating means (note scale giving means) and effective pitch checking or determining means (see FIG. 2). According to the flowchart showing the operation of the embodiment, the note scale generation means reads the scale weight data (weighted note scale) from the scale weight data memory 5 (see FIG. 29), and the read note scale weight. It is composed of means for changing the root note of chords and chord constituent sounds. Further, the effective pitch inspecting means inspects the weight on the note scale assigned to the candidate for the melody note candidates that are intermediately created in the process of generating the distributed chord and the process of adding the non-harmonic tone, Speaking of which, the weight is the parameter C
Compared with a threshold value (for example, PC ₂ , ₁ ) planned by the operation unit (melody control information generation unit), and when the weight of the candidate is equal to or greater than the threshold value, the candidate is determined as a valid note. Has become.

In the example of FIG. 29, a certain scale has two kinds of weights and a certain scale has three kinds of weights.
As shown in the reading of scale weight data (scale data) in FIG. 30B, the user can determine which note scale to use from the input device 1. However, in this example, for convenience of explanation, it is assumed that the note scale read from the memory is independent of information such as chords. Figure 50,
The examples shown in FIGS. 70 and 51 are examples in which the note scale read from the memory depends on the type of chord. The concept of the available note scale is used here. There is more than one available note scale that can be used for a given chord. The user can select which available note scale to use for a particular chord, but this is done automatically here (see FIG. 51).

More specifically, in FIGS. 50 and 70, the memory 1 is a memory for storing chord progress information, and numerical values (code numbers) representing chord types are arranged at consecutive addresses (((b) )reference).
The chord progression is represented by this array of chord numbers {CNi}. The memory 2 is a table describing the number of available note scales for each chord (CNi). On the other hand, in the memory 3, the names of the available note scales (ANS) for the respective chords are expressed numerically. From the name information of the available note scale, the scale data (data in which the available note scale is shown by a weight pattern) stored in the memory 4 can be specified. That is, A of the memory 3
In the NS column, "0" means natural mode, "1" means pentatonic scale, "2" means blue note scale, "3" means whole tone scale, and "4" means alternate scale. From this ANS value, the start address of the scale data on the memory 4 can be calculated. On the other hand, between the memory 2 and the memory 3, a specific ANS of the memory 3 can be designated based on the content of the memory 2, and between the memory 1 and the memory 2, the content (code type) of the memory 1 can be specified. The number of available note scales for the chord type can be read from the memory 2. For example, when CNi has a value of "7" indicating V7, the value obtained by subtracting 1 from that value (6 in this case) is used to access the memory 2, so that there are three available note scales of V7. I understand. If it is desired to specifically know the three available note scales of V7, that is, in the form of weighted scale data, the following is performed. First (CNi-1) Here, the number of available note scales is accumulated up to 6 (using the memory 2).
This accumulated value (10 in this case) indicates the start address of the list of available note scale (ANS) names for the chord type V7 on the memory 3. It is already known that there are three V7 ANS. Therefore, this start address, the next address and the next address are individually calculated to calculate the start address of each available note scale to be read from the memory 4. The length of each note scale data on the memory 4 is 12 addresses. As a result, the data of the natural mode, the pentatonic scale, and the blue note scale, which are the available note scales of V7, are read from the memory 4.

Therefore, the memory illustrated in FIGS. 50 and 70 can be used as an available note scale database, and if desired, a list of available note scale data or names for a target chord can be displayed by a CRT or the like. It's easy. With such a monitor, it is easy for the user to select an available note scale that matches the progression of the song. Further, the available note scale database illustrated in FIGS. 50 and 70 is easily expanded.
For example, a code not included in the code list (see (b)) can be added to the list. For example, the new code type is given a value larger than CNi having the current maximum value. Given the name of each available note scale of the new chord type, expand memory 2 and write that chord type (chord number) and its number of available note scales to the next address. Write. When the available note scale data that has never existed is given, a new name (5 in the example of the figure) is given and the memory 4 is expanded. Therefore, the user can easily register a new chord, the number of available note scales for the chord, the name, and the weight pattern (scale data). In addition, it is easy to change (add, delete, correct) the list of available note scales for registered chords.

Now, in the flow shown in FIG. 51, selection of the available note scale for a given code (CNi) is executed using random numbers (51-
5, 51-6). In other respects, it is the same as the method for reading the name or scale data of the available note scale for the specific code described above. The right side of the flow in FIG. 51 is annotated, and no further explanation is necessary.

* Synthesis of Note Scales * A set of weighted note scales can be stored in the scale weight data memory (memory 5 in FIG. 1, memory 4 in FIG. 70). The problem here concerns the synthesis of note scales (scale weight data). Now, let's express the scale data {SCLi} of the chromatic scale as follows. Next, the scale data of the pentatonic scale {SCL
Let i} be as follows. European scale scale data {SCLi} is as follows. Adding the above three scale data, Is obtained. This (d) is completely in agreement with the data of the major scale without the distortion of (2) in FIG. Further, when the data of (a) is added to twice the data of (c), the data agrees with the data of (4) major scale Europe in FIG.

The above example is to show that data of a certain scale can be synthesized by combining scale data of two or more scales by linear combination or the like.
This has two meanings. First, the scale data representing a known scale can be obtained by synthesizing another scale data representing a plurality of known scales. The second point is that scale data representing a scale of a type that does not exist until now can be obtained by synthesizing scale data representing two or more known scales. Therefore, it has the possibility of being an automatic tune on an experimental scale.

In terms of storage capacity, the set of scale data stored in the scale weight data memory can be made small, which is advantageous. The desired scale data may be automatically generated by the linear combination when necessary. This kind of linear combination can be easily executed if the scale names to be linearly combined and the coefficient for each scale are determined. S
Let CLij be the weight data of the j-th sound on the i-th (named i) scale, and the scale data after synthesis is SC.
Expressed as L _{i, NEW} , SCL _{i, NEW} is SCL _{i, NEW} = Σ
It can be easily obtained by calculating a _j SCLij (a _j is a coefficient).

Threshold Threshold for valid pitch test (eg PC _{2, 1} ) is
In the case of the embodiment, it is created in the melody control information generating section. The effective pitch determination means compares the weight assigned to the candidate pitch (weight given by the note scale generation means) with a threshold value, and when a predetermined condition is satisfied, in the case of the embodiment Then, when the weight has a weight equal to or greater than the threshold, the candidate is adopted as a melody note. When the condition of effective pitch is that the weight of the candidate ≧ the threshold, the action of the threshold on the candidate is as follows. First, consider the case where the note scale given by the note scale generating means has only two types of weights. In this case, when the threshold is heavier than any weight, the effective pitch does not occur. When the threshold is in the middle of both weights, the heavier pitch is treated as valid and the other is invalid. When the threshold is lower than any of the weights, all pitches on the note scale are treated as valid (becoming chromatic). Even when there are three or more kinds of note scale weights, the action of the threshold value can be similarly understood.

In the case of the embodiment, the scale notes which are also the chord constituent notes are promoted and have a higher value than the normal scale notes. Therefore, if the threshold value is selected so that weight (scale note) <threshold value <weight (chord constituent note and scale note), the automatic composer of the embodiment becomes a distributed chord generator. In the case of the embodiment, the note scale generating means does not include means for introducing fluctuations or fluctuations. Such variation introducing means can be easily realized. As a result, a certain pitch is stochastically defined as an effective sound. For example, array of note scale data {SCLi}
For each element of, a variation (for example, addition) controlled by the size of each element is introduced by a random number. Alternatively, a relatively small value generated by a random number is added to an element whose difference from the threshold value is small to some extent. By using such controlled variation introducing means, the pitch that is always an effective sound, the pitch that is sometimes an effective sound, and the pitch that is never an effective sound are defined.

In a sense other than the primary combination of the note scales, the enrichment of the note scales that can occur is brought about. It has already been explained that note scale data can be made user programmable. If desired, the threshold value can be freely set by the input device (this is very easy to realize).

Rhythm Relationship <Review and Development of Embodiments> In the above embodiments, the following means have been provided for the melody rhythm.
The first is the extraction and injection of mini-patterns, and the second is the combination and division of sound lengths by the pulse scale. More specifically, in the evaluation of the motif, the dominant small note length sequence included in the motif including non-harmonic sounds is evaluated and extracted as a characteristic minirhythm pattern, and based on the extraction result,
In the final process of melody generation, the characteristic mini rhythm pattern is incorporated into the melody tone length sequence. On the other hand, the combination and division of the note lengths by the pulse scale is used as a means of forming a motif distributed chord sound pattern (sound length sequence of the distributed chord sounds of the motif) in which the non-harmonic sounds are omitted from the motifs. It is used as a means for forming (a sequence of tone lengths consisting of a melody consisting only of distributed harmony tones). The means by mini-pattern is direct,
In a superficial way, it shares some aspects with "words" in natural language. On the other hand, the pulse scale method is an indirect approach rather than the mini pattern method.

In the above-mentioned embodiment, an approach based on the concept of "distributed chords" as the basis of the melody is adopted for the analysis and synthesis of pitch sequences. With respect to (for melody) and for generation (for melody), the means of duration combination and division by the pulse scale works effectively. The rhythm or the tone length sequence is finely adjusted and controlled by both the means using the pulse scale and the means using the mini pattern.
However, this does not mean that both measures are absolutely necessary. That is, according to the present invention, only the means by the mini pattern or only the means by the pulse scale can function as the means for adjusting and controlling the rhythm. Also, it should not be understood that the pulse scale means can only be used for "distributed chords". It is possible to use the means by the pulse scale and the means by the mini pattern, independently of the "dispersed chord". The relationship between the means by the pulse scale and the distributed chord is that one effective means of forming the sound length sequence of the dispersed chord is the means by the pulse scale, and the formation of the sound length sequence of the dispersed chord by other means. I do not deny it. Conversely speaking, one means effective for forming the tone length sequence of the distributed chord, that is, the pulse scale means, can be used for forming other tone length sequences (such as the tone length sequence of a general melody). Specific examples thereof will be described later in detail.

<Concatenation and division of notes by table reference>
In the above embodiment, a pulse scale in which each pulse point is weighted is used to combine and divide the notes. However, the pulse scale is in the form inherent in the program (see FIGS. 15 and 16). In another pulse scale generating means, pulse scale data is stored in a table (memory). The advantage of realizing the pulse scale in the table is that by preparing data of various pulse scales and selectively sending them to the note combining / dividing means, even if the rule of the note combining / dividing means is the same, different division, combining results, That is, various tone length sequences can be changed. Conversely, complicated and complex rules inherent in various pulse scales are unnecessary.

The table reference type sound length combination and division means will be specifically described below. First, FIG. 52 illustrates a flow of table reference type pattern extraction used to obtain a distributed chord pattern of a motif. The points not specified are the same as those in the flow of FIG. 52-1, T _SUMB is the start position of the note of interest (SUM
It is the weight in B) and is the SUMB-th data in the pulse scale table {Ti}. T _SUM is the pulse scale at the start position (SUM) of the next note {T
i} is a weight on. Therefore, 52-1 to 52-
The meanings of the processes up to 4 are as follows. That is, (i) when the start position of the next note has a heavier weight,
The note of interest (one note before the next note above)
Is added to the note length RH _N of the preceding note (preceding chord note). (ii) When the starting position of the note of interest has a heavier weight, the note length MRi is set to the next note (harmonic sound after).
To the note length RH _{N + 1} (However, if the weights are the same,
For the time being, it is absorbed in the next note). Therefore, if the table has a pulse scale as shown in the upper right of the flow shown in the figure, that is, the same pulse scale as that in FIG. The result of the sound length pattern extraction is the same. The point that should be noted is Figure 1
The flow of _No. 5 is the ordering of weights (T ₀ > T ₈ > T ₄ = T ₁₂ > T ₂ ...... = T ₁₄ > T ₁ = ... T as illustrated in FIG.
₁₅ ) can be extracted only with the pulse scale that has been performed, the flow of FIG. 52 is that any pulse scale can be used, and the result changes variously depending on which pulse scale is used.

FIG. 53 is a modification of FIG. 37 and exemplifies the flow of optimal note combination by table reference. 53-1, each value of the pulse scale (T _{1 to} T ₁₆ )
Is rearranged in ascending order and substituted into SORTi in order to speed up the subsequent processing. In 53-2, the weight on the pulse scale at the start position SUM of the next note is S
It is compared with the value of ORTi. First SORTi is SOR
T1 and has the lightest weight on the pulse scale.
The next note (MER _{j + 1} ) starting at the point of the same weight as this lightest weight is preferentially combined with the previous note (MER _j ). Hereinafter, similarly, the sound lengths are combined in order from the lightest. 54 and 55 show an example of the flow of sound length combination by table reference. The flow of FIG. 54 shows details of the check 55-2 in FIG. FIG. 55 can be used as an alternative to FIG. 38 of the above-described embodiment. Shift 55- in FIG.
3 and execution 55-4 are as shown in FIGS. 40 and 41, respectively.

In FIG. 54, SUM represents the start position of the next note (end position of the current note) and SUM represents the start position of the current note. In 54-3, the value of the heaviest pulse point among the pulse points at which the current note crosses is obtained. By repeating the loop from 54-2 to 54-8 until K> PC is satisfied (until all the notes have been checked), the start position of the note that straddles the heaviest pulse point of all the notes. Is set to the flag ff, the note number is set to the flag fl,
The maximum weight of the crossing pulse points is the register MAX
Is written to. In the execution 55-4 of FIG. 55, the note found as a result of the check 55-2 is divided into two note lengths with the maximum weight pulse point as a boundary. When the number of notes reaches the target number, i> | S | is established, and the division work is completed.

56 and 57 show concrete operation examples of the division and connection of notes by the above-mentioned table reference type. Both of them are original length patterns

[Outside 33] Are using. 56 is a pulse scale that is divided and combined using the illustrated scale (to be referred to as a positive logic scale), and FIG. 57 is in contrast to the pulse scale shown in FIG. This is an example of division and combination using a (complementary) scale (which will be referred to as an inverse logic scale).

Positive logic, inverse logic, evaluation value ★ Here, the pulse scale is represented by {Ti} (i = 0 to 15). Therefore Ti is the weight at the i-th point on the pulse scale. Also, the rhythm pattern is {Ri}. Ri is zero or 1, and Ri =
1 indicates that the note start position is at the i-th point, and Ri = 0 indicates that the note does not start at the i-th point. Therefore, the number N of notes is

[Equation 1] Is represented by next,

[Equation 2] Consider the function (evaluation value) represented by.

Now, let us review the logic of note division and connection described above, using this function as a guide. The basic rule of the above note division means is that a note that crosses a pulse point having a "maximum" weight among the given notes is divided into two notes with the pulse point as a boundary. is there. Therefore, the evaluation value V before division
The difference between (BD) and the evaluation value V (AD) after division is related to the above-mentioned “maximum” weight. That is, the above note division logic causes the function V to change (increase) to the maximum by dividing. On the other hand, the basic rule of the above note combination means is that the note starting from the pulse point with the "minimum" weight among the given notes (with that pulse point as the connecting line) It is a long bond. Therefore, the difference between the evaluation value V (BT) before combining and the evaluation value V (AT) after combining is related to the above-mentioned "minimum" weight. Let's call the above two basic rules "positive logic" or simply "positive logic".

"Positive logic" in this sense is a rule that is not destroyed by the pattern of pulse scale weights used. However, given a pulse scale and its complementary (contrast) pulse scale, rule execution produces the opposite result. The mutually complementary pulse scales are two pulse scales in which the point having the maximum weight on one pulse scale has the minimum weight on the other pulse scale. For example, the pulse scale shown in FIG. 56 and the pulse scale shown in FIG. 57 are complementary. When the pulse scale shown in FIG. 56 is divided according to the rule of "positive logic", the point of the "maximum" weight where the notes cross on the pulse scale becomes the dividing line of the note, and the pulse scale is divided into "positive logic". When combining with the rule of “”, when the rhythm pattern and the same pulse scale are overlapped, the point with the “minimum” weight among the notes in the rhythm pattern is noted, and this minimum value point is the connecting line of the notes. Becomes On the other hand, when a pulse scale (FIG. 57) complementary to the pulse scale of FIG. 56 is used, “FIG.
On the pulse scale of "6", the point of the "minimum" weight at which the notes cross becomes the note dividing line, and "Fig. 56"
The point of the "maximum" weight at which the note starts on the pulse scale of is the note connecting line. Even with the same rules, complementary pulse scales produce opposite results. Therefore, the pulse scale shown in FIG. 56 is called a “positive logic” scale, while the pulse scale shown in FIG. 57 is called a “reverse logic” scale.

The rule used by the tone length division / coupling means can be changed to "inverse logic" instead of "positive logic". Alternatively, it is easy to selectively divide and combine the sound lengths by "forward logic" and "reverse logic".
According to the "inverse logic" rule, at the time of division, the point of the "minimum" weight on the pulse scale where the notes cross each other becomes the note dividing line. Therefore, the fluctuation (increase) of the evaluation value V is minimized. In the case of connection, the point of the "maximum" weight on the pulse scale where the note starts is the note connection line. Therefore, the fluctuation (decrease) of the evaluation value V becomes maximum.

Therefore, possible combinations of logics regarding division and combination are as follows. (i) Partition and join are positive logic (ii) Partition is positive logic, join is inverse logic (iii) Partition is inverse logic, join is inverse logic (iv) Partition and join are inverse logic I will explain again from another angle later.

[Synthesis and Decomposition of Pulse Scale] A pulse scale has a property that it can be composed from several pulse scales and, conversely, a property that it can be decomposed into some simple pulse scales. In FIG. 58, some rhythm patterns and rhythms of liz are illustrated. Shown in (A) is an example of a samba rhythm. It is played on three (or four, as shown in brackets) instruments here. A simple addition of the number of sounds played by each musical instrument is shown below (A). That is, 2122 3212 1213 3111. The place with a high value is also the place where the generated sound pressure is high. This numerically expressed pattern can be used as a pulse scale. The pattern here is also a numerical expression of polyrhythm (composite rhythm) that has a samba-like character as a whole. FIG. 58B shows an example of 16 beats. The numerical expression of this composite rhythm is shown as 3010 4010 3111 4020 (again, it is a simple addition).

58 (C) to (G) are examples of monorhythms. It is a direct indication of the characteristics of pure rhythm. If you look closely, you can see that the 16-beat complex rhythm includes 4-beat-like, rock-like, 8-beat-like, and so on. In other words, numerically linearly combining the paste patterns shown in (C), (D), (E), (F), etc., the numerical expression of 16 beats (composite rhythm) shown in (B). It looks exactly like the pattern.
From a certain aspect, a pulse scale represented by a pattern of three or more weights represents a complex rhythm.
And, as just illustrated, the known complex rhythm is a more pure monorhythmic combination (a rhythm formed by the performance of several musical instruments). If this monorhythm is expressed by the weight pattern of 1 and 0 (pulse scale),
By combining several pulse scales, such as a linear combination, one obtains a pattern that characterizes a particular polyrhythm.

Further, a pulse scale characterizing a novel polyrhythm can be easily obtained. For example, the rhythm pattern of the second row of the samba shown in (A)

A pulse scale characterizing the complex rhythm of samba rock can be obtained by linearly combining the pulse scale 1010 1101 0101 1000 representing ## EQU33 ## and the pulse scale 0000 1000 0000 1000 representing rock nori.

In one implementation, the pulse scale Tij
Are stored in memory. The linearly combined pulse scale Ti (synthesis) is synthesized by calculating Ti (synthesis) = Σ ^a _j Tij (where ^a _j is a coefficient) using data of two or more pulse scales. Such an arrangement produces a very large number of pulse scales from a limited number of pulse scales. In order to save the storage capacity to the maximum, an independent pulse scale ti having only two weights 1 and 0 is used.
It is better to prepare a set of j.

* Dependence of pulse scale on song progression * For example, the characteristics of the rhythm of the melody usually change at fill-in, chorus (tail opening), and 4 Bar variations. For this purpose, the pulse scale can be changed as the song progresses. For example, the parameter C calculation unit systematically generates or selects the pulse scale. It is also possible to use means for introducing very small fluctuations or fluctuations into the pulse scale.
For example,

[Outside 34] Next to the measure

[Outside 35] The pulse scale can be slightly changed so as to obtain the rhythm (sound length sequence).

<Evaluation of motif and generation of melody by PS> It is assumed that a motif is given. As described above, the motif is represented by the pitch string data and the pitch string data in this embodiment. Here, the data {MRi} of the tone length sequence of the motif or the data {Ri} of the start position of each tone length can be evaluated by the pulse scale as described above. Although various evaluation functions can be used, here, for convenience of explanation,

Let's select [Equation 2] as the evaluation function. Where {T
Let i} represent a pulse scale with three or more weights. The i-th pulse point has a weight of Ti. N is the number of notes in the motif. Further, for convenience of explanation, the pulse scale {Ti} will be considered as a numerical expression of a specific composite rhythm.

Now, what does the person who inputs the motif, that is, the user, expect after the motif? This is a difficult problem. We will consider this issue later. The evaluation value V of the motif can be calculated by applying a certain pulse scale {Ti}. The evaluation value depends on the tone length sequence of the motif. In some cases it will be relatively large and in others it will be small. On the assumption that the pulse scale {Ti} is a numerical expression of a complex rhythm, the fact that the evaluation value V showed a large value means that the rhythm pattern of the applied motif has a large value in the position used in the pulse scale used. It is highly possible that there are many note start points Ri. In other words, it can be regarded as a rhythm pattern in which the character and impression of the combined elements (for example, rock-like, 8-beat like) are strongly expressed. On the contrary, the small value of the evaluation value V suggests that there are many notes that make the weakest personality among the combined elements stand out.

Here, some of the logics or rules that can be used by the note division and combination means for generating the melody rhythm pattern (here, the meaning of the note length sequence) will be presented. In the first approach, both splitting and joining are performed in "positive logic". That is, the value of the evaluation function V is
The point that increases the most with the minimum number of divisions is the note dividing line, and the point that gives the minimum decrease with the minimum number of couplings is the note connecting line. This approach acts to maintain a compounded character on the pulse scale. In the second approach, the points at which the evaluation function V shows the minimum increase by division and the maximum decrease by connection are the note division line and the note connection line, respectively (inverse logic approach). This approach tends to create new features that are independent of pulse scale features.

In the third approach, the points at which the evaluation function V has the smallest variation in both division and connection are set as division and connection lines. In the fourth approach, positive logic is used when the value of the evaluation function V is large, and negative logic is used when the value of the evaluation function V is small. The first approach is suitable for a user who intends to develop a rhythm from a motif to a melody in accordance with the character of the complex rhythm inherent in the pulse scale. Second
This approach is suitable for the user who is aware of the complex rhythm but intends to express a characteristic different from the complex rhythm in the melody rhythm. The third approach is suitable for users who are not aware of complex rhythms. This is because it is against the user's expectation that unintended characteristics of the composite rhythm are exposed due to division and combination.

The above approach is effective for each user. However, when the user asks "what"
Whether or not the motif was created intentionally cannot be known from the motif alone. In order to solve this, it is necessary to promote interaction (interaction) between the automatic composer and the user. (Although the form of the interaction is considered to be various (for example, if the user wants a lock pattern, the user can make a “rock-like” request via an input device).) There exists a pulse scale that best matches the rhythm, and it is considered meaningful to use this pulse scale as the basis for the creation of the melody rhythm.

Evaluation of degree of syncopation * Here, as an example of evaluation of the rhythm of a motif, an example of measuring the degree of syncopation will be given. In general, syncopation is the movement of accents in passages from the normal position. As a pulse scale (4-beat system) for measuring the degree of syncopation, the first beat (T ₀ ) has the maximum weight, the third beat (T ₈ ) has the next highest weight, and the second beat (T _4). ) And the fourth pulse (T ₁₂ ) is lighter. Furthermore, divide one beat into four equal parts,
The first pulse is the heaviest of them, the third pulse is the heaviest next, and the second and fourth pulses are the lightest (hierarchical structure of weighting pattern). An example of this type of pulse scale is shown below the flow in FIG. Here, T ₀ T ₁ T ₂ T ₃ T ₄ T ₅ T ₆ T ₇ T ₈ T ₉ T ₁₀ T ₁₁ T ₁₂ T ₁₃ T ₁₄ T ₁₅ 5 1 2 1 3 1 2 1 4 1 2 1 3 1 2 It is 1.

In the flow of FIG. 59, i is a note number,
SUM is the start position of the (i + 1) th note (the total length of each note from the beginning of the measure to the i-th note) when calculating 59-4, and S is The cumulative value of the weights for the note start position sequence {SUM},
That is, S = ΣT _SUM (SUM = 0 to the last SUM value). This is because when Ri and Ti described above are used,

(Equation 3) Is equivalent to Therefore, V calculated by 59-6
Is

(2) (N is the number of notes), which has the same format as the evaluation function described above. Here, V represents the measure of syncopation of the rhythm of the motif (note length sequence {MRi}).
FIG. 60 shows syncopation evaluation values calculated by the pulse scale and flow shown in FIG. 59 for various rhythm patterns.

Generation of Controlled Duration Pattern * As described above, for the rhythm of the given motif, the rhythm evaluation value can be calculated by using the pulse scale. From now on
This is a technique for generating a controlled note length pattern (which can be used as a melody note length pattern) based on the rhythm evaluation value of the motif thus obtained. The value of the evaluation function V changes depending on the tone length sequence to be evaluated. Therefore, generally, the evaluation function V also depends on the number of notes n of the note length string to be evaluated. Now, it is assumed that the value of the evaluation function V of the original (motif) tone length sequence having the number of notes N is V (N). Consider the expansion of the original note length sequence for an arbitrary number of notes n. In this case, the evaluation function V is V when N
It can be a curve that passes through the point (N). This evaluation curve can be obtained in various ways (for example, by calculating the evaluation value of the note length sequence obtained by the above-mentioned dividing and combining means for all the note numbers).

A random tone length sequence is generated by the random number generation means. It is checked whether or not this random tone length sequence is in good agreement with the above evaluation curve. If the result of the inspection is acceptable, the tone length sequence is adopted (as the tone length sequence of the melody or as the tone length sequence of the intermediate melody). This makes it possible to obtain a tone length sequence controlled by the evaluation function. FIG. 61 is an example of a flow for generating a controlled tone length pattern. At 61-2,

[Equation 4] (T _Ri is the weight of the Ri-th pulse point) is calculated, but this arithmetic expression is only an example. For example, a pulse scale {Ti} and a sequence of random sound generation points {R
It is also possible to use a coincidence function that evaluates the degree of coincidence with i} (a function that becomes 1 for the best match and 0 for the most disagreement). For example,

(Equation 5) Alternatively, the pulse scale {Ti} is decomposed into a plurality of sub-pulse scales {tij} having only weights of 1 and 0 (as M), and the M sub-pulse scales are used to

(Equation 6) Etc. can also be used.

According to the flow, 61-in FIG.
The details of 1 are shown in FIG. Since the operation is clear from the clear description of the flow itself in FIG. 62, detailed description will be omitted. In short, as shown in FIG. 61, N−1 different numerical values (representing the sound generation point Ri) are generated by random numbers from 1 to 15. 62-7 R
₀ = 1 suggests that a note always starts at the beginning of a bar (not necessarily required). FIG. 63 shows the details of 61-2 in FIG. No explanation is necessary.
FIG. 64 shows details of 61-4 in FIG. A clear description of the flow itself in FIG. 64 and an operation example (ME
Ri represents the tone length data of the i-th note), so that no explanation is necessary. PCx and PCy shown at 61-3 in FIG. 61 are, for example, parameters given by the melody control information generating means. The lower limit and the upper limit of the allowable evaluation value V are shown respectively. Those passing the test of 61-3 are converted into data of the tone length sequence of the melody in 61-4.

Optimal PS Estimation for Motif Rhythm Estimating or associating a pulse scale that best fits the rhythm of a motif is effective for later melody generation. As an example, FIG. 65 shows a samba-like pattern.

The characteristics of the evaluation value when the notes are divided and combined using three different pulse scales, that is, a positive logic scale, an inverse logic scale, and a samba scale, are shown. FIG. 66 is a normalized evaluation example (a characteristic based on negative logic is normalized so that both ends match other characteristics). As an evaluation function,

(Equation 7) Are using.

As can be seen from FIG. 66, the evaluation value changes to the most “smoothness” in the case of division and connection by the samba scale (calculation of smoothness is omitted here). Therefore,

The optimum pulse scale that best fits the original tone length sequence of [Ex. 33] can be evaluated as the samba scale. Next, a technique for estimating a pulse scale that gives an evaluation of a peak value (maximum value or minimum value) as an optimum pulse scale for a rhythm pattern of a target motif will be described.

As mentioned above, in the case where the pulse scale is assumed to be a numerical representation of a complex rhythm, the pulse scale may be represented by a linear combination of several purer pulse scales (which we shall call subpulse scales). it can. Here, it is assumed that the sub-pulse scale is a weight pattern having only two weights. Then, without loss of generality, the sub-pulse scale can be expressed as pattern data having only values of "1" and "0" (including the case of only "1"). The pattern of only 0 is not applicable.

Now, assuming that a plurality of pulse scales are prepared, a predetermined evaluation function (function including pulse scale data as a variable) is applied to a rhythm pattern (sound length sequence) of a certain motif. The value of the evaluation function changes depending on the type of pulse scale and the rhythm pattern to be evaluated. The evaluation value on a certain pulse scale shows a peak for a certain rhythm pattern, and the evaluation value on another pulse scale shows a peak for another rhythm pattern. When there are a wide variety of pulse scale types,
If the characteristics of the rhythm pattern have a similar relationship (similarity) to two (or more) pulse scales, there may be two (or more) pulse scales giving peaks. In this case, either pulse scale is estimated to be the optimum pulse scale within the set of applied pulse scales.

In the following description, the peak is used in the maximum meaning (for convenience of description). Two evaluation functions are illustrated as the evaluation function. The first evaluation function is expressed as follows.

(Equation 8) Here, {Ri} represents a rhythm pattern (sound length sequence) represented by weights of 1 and 0, and {tij} represents the j-th sub-pulse scale. On the other hand, as described above, the pulse scale (composite one) is represented by a plurality (N) of sub-pulse scales {tij}. That is, for each element Ti of the pulse scale {Ti},

[Equation 9] Is established.

The above V (mean) represents the average of the subject rhythm pattern {Ri} evaluated by each component scale {tij} of the pulse scale {Ti}. The possible value of this evaluation function V (mean) is from 0 to 1. The value in [] also takes a value from 0 to 1. It becomes 1 when tij = Ri holds for all i. That is, when the generation point matches the rhythm pattern {Ri} and the sub-pulse scale {tij}. In this sense, V (mean) is a matching degree function.

The second evaluation function is expressed as follows.

[Equation 10] This is the maximum evaluation value within []. That is, N
The sub-pulse scale that gives the maximum value among the sub-pulse scales dominates this evaluation function V (max). In the following description, the first evaluation function V (mean) may be simply referred to as “average”, and the second evaluation function V (max) may be simply referred to as “maximum”.

FIG. 67 exemplifies five patterns as the rhythm patterns to be evaluated (for each motif tone length). “Samba” is a pattern that has the characteristics of samba. "Enka" is a pattern that seems to be Enka. "Now" is a recent trend of syncopated pattern. “4” is a pattern of four quarter notes, and “2” is a pattern of two half notes. further,
In FIG. 67, four types of pulse scales are shown as a linear combination of sub-pulse scales. The first is a positive logic (normal) pulse scale, which is a composite of five sub-pulse scales. The second is the first inverse pulse scale (inverse logic pulse scale) and is composed of five sub-pulse scales. The third is the samba scale, which has four sub-pulse scales as components. Fourth
Is a 16-beat pulse scale, which is a composite of five sub-pulse scales.

FIG. 68 shows the evaluation result. The five types of motif tone length sequences illustrated in FIG. 67 are classified into the first evaluation function V (mean) (column indicated by “average” in the figure) for each pulse scale by the four types of pulse scales shown in FIG. 67. It is the result of evaluation by the second evaluation function V (max) (column indicated by "maximum" in the figure). The result suggests that, firstly, all the evaluation functions make similar evaluations. For example, the highest score for the "samba" motif is given both "average" and "maximum" to the samba pulse scale. The pulse scale giving the highest score for "enka" has changed, but both "average" and "maximum" are for the positive logic pulse scale. Second
The point is that the first evaluation function “average” exhibits higher selectivity. Conversely, it can be said that the second evaluation function has a rougher evaluation (wide association). However, this is also related to the complexity of each pulse scale. Many existing complex rhythms are formed by several rhythm patterns that are strongly related to each other.

The pulse scale giving the highest point is set as the optimum pulse scale, and the melody rhythm pattern (sound length sequence) is formed on the basis of the pulse scale.
This is one effective approach. However, this
It should not be interpreted in an "absolute" sense. That is,
The root is the pattern that the user wants (as long as the primary goal is to satisfy the user's expectations). Therefore, the user may freely select the pulse scale. As yet another approach, the composition may be performed on the basis of each of several pulse scales, and the evaluation may be left to the user. Here again, the interaction or interface in the composition between the user and the automatic composer poses a problem, but we will not enter any further.

* PS association in consideration of pitch sequence * Rhythm in melody cannot always be evaluated only by pitch sequence. That is, pitch strings and other elements also operate. In the present embodiment, for the sake of simplicity, the melody and the motif are represented by two components, that is, the data of the pitch length string and the data of the pitch string, and other elements are not considered. Here, we will describe an example of the association of pulse scales that also considers pitch sequences.

For example,

[Outside 36] In such a pattern, the position of "do" acts to increase the weight of that position on the pulse scale. More generally speaking, the regularity of pitch patterns regulates future patterns to a greater or lesser extent. This is something I often experience. Reflection on future patterns based on such regularity can be expressed by upgrading the weight data on the pulse scale. That is,
The upgraded weighted pulse points play a central role in the division and combination of note lengths, incorporating the regularity of the present pattern into future patterns to a greater or lesser extent. To put it another way, humans have the experience that when they hear a passage, they often can predict what passage will come next.

The example shown below is a technique for associating a pulse scale by measuring the autocorrelation of a pattern of a tone train or an element very similar thereto. In this example, the pulse scale is associated with the following. (i) The position data of each note is obtained from the note length string data, the weight at that position, for example, 1 is attached, and the other positions are unweighted (zero). (ii) DFT (Discrete Fourier Transform) of pitch sequence data (pitch time series) is taken, and its spectral component is extracted and 2 ^X
Overtone components are overlaid on the pulse scale as weights.

In FIG. 69, (a) is time series data of the sound to be evaluated. Within one bar at 8 minute intervals,
It is a pattern of domilefamisofara. The DFT result for this pattern is shown in FIG. In (b), 1 is the magnitude of the fundamental tone (a component having one measure as one period), 2 is the magnitude of the second overtone component (a component having half a measure as one period), and 4 is the The magnitude of the fourth harmonic component is shown. As shown, the magnitude ratio between the 1st, 2nd and 4th overtones is about 1: 0.5: 1. The position of C is the same as the position of the first overtone (fundamental tone) (the start position of each cycle) and the position of the second overtone and the fourth overtone. In position. In addition, the position of the rear Mi coincides with both the position of the second overtone and the position of the fourth overtone.

Therefore, according to one weight upgrade,
It looks like this: The position of the first do is 3.5 points in total, including 1 point of 1st harmonic, 0.5 point of 2nd harmonic, 1 point of 4th harmonic, to 1 point of rhythm data. Similarly, the first pulse point is 2 points, the latter pulse is 2.5 points, and the last pulse is 2 points. That is, The pulse scale {Ti} of is created.

Furthermore, the rhythm data of the distributed chord may be added. As can be seen from the above description, the pulse scale can also be regarded as a pattern having weights weighted based on the time-series regularity. That is,
The scale on which a note is likely to appear at the same position in each bar (meter) is represented by a weighted weight on the pulse scale.

<Others, Applications> As described above, various modified examples including the embodiment have been described in detail. Regarding the rhythm relationship, a means using a mini pattern (that is, a means for extracting a mini pattern that characterizes a motif and a means for incorporating a mini pattern that characterizes a melody into a tone length sequence of the melody based on the extraction result) and a means using a pulse scale. (Sound length division by pulse scale, combination means, motif rhythm evaluation means by pulse scale, controlled sound length sequence generation means based on the evaluation result, etc.) can be used jointly or independently. You can also

Further, regarding the pitch sequence generation of the melody, the above-mentioned embodiment adopts the approach in which the distributed chord is the basis for the generation of the melody, but the invention is not necessarily limited to this. For example, it is possible to generate a melody based on a scale, some of which are already known. For example,
It is possible to use a melody pitch sequence generation unit that specifies a scale to be used and determines the order of pitches selected from the scale by the 1 / F fluctuation method. Alternatively, a means for generating a pitch sequence is specified by specifying a scale to be used and executing a process in which a pitch greatly depends on the immediately preceding pitch, that is, a Markov process using a frequency table or the like. May be. Also, regarding the melody control information generating means,
You may make it generate | occur | produce control information (melody plan information) which considered the interaction between each element of a melody.

The means using the pulse scale is applicable to a rhythm-only generator. For example, it can be applied to a rhythm variation device that varies the motif rhythm.
Alternatively, it may be used for a rhythm analysis device that analyzes the rhythm characteristics of the melody of a song. Further, it can be applied to a device (arrangement device centering on rhythm) for converting and changing the rhythm characteristics of a completed song.

[0191]

According to the rhythm generation device of the first aspect,
For example, a pulse scale is generated as rhythm control data.
Change the rhythm given based on this pulse scale.
It forms and creates a new rhythm. Therefore just
Instead of the random number method, the pulse-scale controlled format is used.
It is possible to realize the automatic generation (compose) of
Wear. Note that claim 2 is the note division / concatenation of claim 1.
This is a technical limitation on the combination method. Claim
According to the rhythm generation device described in 3, the frequency is high in the original rhythm.
The note length sequence pattern that appears in
Extracted, and based on this extracted characteristic rhythm pattern,
Is generated, the characteristics of the original rhythm were reflected.
Can generate rhythm. Rhythm according to claim 4
According to the analysis device, the characteristic rhythm pattern of the applied rhythm
Can be extracted, can be surface analysis of the rhythm
It Such a rhythm analyzer is applied to a rhythm generator
can do. According to the rhythm analysis device of claim 5,
If so, analyze the background (rhythm style) of the assigned rhythm
Can be. According to the rhythm generation device of claim 6, the original
Automatically controlled rhythms without the need for rhythms
It can be compose.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram of an automatic composer according to an embodiment of the present invention.

FIG. 2 is a functional block diagram of the automatic composer.

FIG. 3 is an overall flow chart of the automatic composer.

FIG. 4 is a diagram showing a pitch data string.

FIG. 5 is a diagram showing a list of variables used in the automatic composer.

FIG. 6 is a diagram showing a list of parameters C used in the automatic composer.

FIG. 7 is a diagram showing an example of input data for explaining the operation.

FIG. 8 is a diagram showing an example of a value of a parameter C.

FIG. 9 is a diagram showing, for each process, a melody generation result for the data of FIGS. 7 and 8;

FIG. 10 is a flowchart of motif evaluation.

FIG. 11 is a flowchart of rhythm evaluation.

FIG. 12 is a flowchart of classification extraction of non-harmonic sounds.

FIG. 13 is a schematic flowchart of parameter extraction.

FIG. 14 is a flowchart for extracting parameters of a chord tone type.

FIG. 15 is a flowchart for extracting a tone length pattern of a harmony tone.

16 is a diagram for explaining a pulse scale used in the flow of FIG.

FIG. 17 is a flowchart for extracting the number of non-harmonic sounds and the number of harmonic sounds.

FIG. 18 is a flowchart for extracting parameters for smoothness and homophonic progression.

FIG. 19 is a flowchart for extracting a characteristic rhythm parameter.

FIG. 20 is a flowchart for extracting the minimum note length.

FIG. 21 is a schematic flowchart of parameter C calculation.

FIG. 22 is a diagram showing an example of musical expression identification data.

FIG. 23 is a flowchart of reading and decoding of musical expression data.

FIG. 24 is a diagram showing an example of a decoding result.

FIG. 25 is a flowchart of melody generation.

FIG. 26 is a flowchart of distributed chord generation.

FIG. 27 is a diagram showing an example of data in a chord constituent note memory, a chord progression memory, and a root note memory.

FIG. 28 is a flowchart of reading chord constituent tones.

FIG. 29 is a diagram showing an example of scale weight data.

FIG. 30 (A) is a flowchart for changing the scale weight (1), and FIG. 30 (B) is a flowchart for reading scale weight data.

FIG. 31 is a flowchart of changing a scale weight (2).

FIG. 32 is a flowchart for calculating the optimum number of turns.

FIG. 33 is a flowchart for turning chord constituent sounds.

FIG. 34 is a flowchart for determining MED ₁ (bar leading sound) from the previous sound.

FIG. 35 is a simplified flowchart of a pitch string generation part of a distributed chord.

FIG. 36 is a flowchart for determining a tone length sequence of distributed chords.

FIG. 37 is a flowchart of an optimal combination process of sound lengths.

FIG. 38 is a flowchart of an optimal sound length division process.

39 is a detailed flowchart of the check in FIG.

FIG. 40 is a detailed flowchart of the shift shown in FIG. 38.

41 is a detailed flowchart of execution in FIG. 38. FIG.

FIG. 42 is a flowchart of adding a sound.

FIG. 43 is a flowchart for adding a progress sound.

FIG. 44 is a flowchart for adding embroidery sounds.

FIG. 45 is a flowchart for adding a missing sound.

FIG. 46 is a flowchart for adding a rest (breath).

FIG. 47 is a flowchart of characteristic rhythm generation.

FIG. 48 is a flowchart of correction learning.

FIG. 49 is a flowchart of parameter change by learning.

FIG. 50 is a diagram showing a modified example, and is a diagram showing a data structure of a scale-related memory (No. 1).

FIG. 51 is a flowchart for reading weight data of a scale.

FIG. 52 is a flowchart for extracting a chord tone duration pattern of a type that refers to a pulse scale table.

FIG. 53 is a flowchart of optimal combination of table-referenced note lengths.

54 is a detailed flowchart of the check shown in FIG. 55.

FIG. 55 is a flowchart of optimal division of sound length by referring to a table.

FIG. 56 is a diagram showing an example of division and combination of notes by a positive logic pulse scale together with a positive logic (normal) pulse scale.

FIG. 57 is a diagram showing an example of division and combination of notes by the inverse logic pulse scale together with the inverse logic pulse scale.

FIG. 58 is a diagram illustrating various rhythms and tracks.

FIG. 59 is a flowchart for evaluating the degree of syncopation based on the positive logic pulse scale.

FIG. 60 is a diagram showing a specific example of a syncopation evaluation value.

FIG. 61 is a flowchart of generation of a controlled tone length pattern.

62 is a detailed flowchart of random number generation in FIG. 61. FIG.

63 is a detailed flowchart of the evaluation in FIG. 61.

64 is a diagram showing an example of data conversion together with details of the data conversion of FIG. 61.

FIG. 65 is a diagram showing characteristics of evaluation curves by division and combination by a plurality of pulse scales.

66 is a view similar to FIG. 65 but normalized.

FIG. 67 is a view showing some pulse scales and sub-pulse scales constituting each pulse scale, together with an example of a motif tone length sequence.

68 is a diagram showing evaluation values by each pulse scale for each motif tone length sequence shown in FIG. 67. FIG.

FIG. 69 is a diagram showing a time series of sound and its spectrum.

FIG. 70 is a diagram showing a data structure of a scale-related memory (No. 2).

[Explanation of symbols]

 F12 Motif sound length (rhythm) pattern extraction means F12-1 Characteristic mini pattern extraction means F32 Sound length sequence generation means F32-1 Optimal combination means F32-2 Optimum division means F32-3 Characteristic pattern incorporation means

Claims (6)

(57) [Claims] 1. An original rhythm represented by a note length sequence of an original note sequence.
Original rhythm applying means for applying rhythm data , rhythm control data generating means for generating rhythm control data, and tone length sequence of the original note sequence based on the rhythm control data
The original rhythm is changed by generating a note length sequence different from
A rhythm transformation means for generating a shaped rhythm , the rhythm control data generating means is configured to generate the rhythm control data.
Each pulse point representing each music time on the music time axis
Pulse scale weighted by
Scale is from multiple pulse points with multiple different weights
Has a pulse scale generating means for generating a composed), the rhythm deformation means, based on the weight of the pulse scale
Therefore, one original note included in the above original note string is converted into two notes.
Split into notes, or two original notes included in the above original note string
The has a note division and coupling means for coupling to a single note, rhythm generator and wherein the. 2. The rhythm generating device according to claim 1, wherein the note dividing / combining means is arranged to perform the above-mentioned division when the dividing is performed.
At the pulse point with the minimum or maximum weight on the pulse scale
The original note that is currently sounding is replaced by the minimum or maximum
At the same pulse point, and when combining, the minimum or
Indicates that the pulse point with the maximum weight is the sound generation start time
A rhythm generator characterized by combining an original note with an original note immediately before this original note . 3. An original in which an original rhythm is represented by a note length sequence of an original note sequence.
An original rhythm assigning means for assigning rhythm data, and data processing of the original rhythm data to produce a sound of the original note.
By generating a note length sequence that is different from the
Rhythm generation means to automatically generate a rhythm different from rhythm
When provided with said rhythm generation means, and data processing the raw rhythm data, the original Rizumude
Pattern that appears frequently in the above original note sequence.
A feature that extracts a sequence (hereinafter referred to as a characteristic rhythm pattern).
Characteristic rhythm pattern extraction means, and a characteristic that incorporates the characteristic rhythm pattern in the generated rhythm
And a rhythm pattern incorporating means . 4. A rhythm in which a rhythm is expressed by a note length sequence of note sequences.
And rhythm providing means for providing the data, the Rizumude
Note pattern that frequently appears in the above note sequence
And a characteristic rhythm pattern extracting means for extracting as a characteristic rhythm pattern of the rhythm. 5. A rhythm in which the rhythm is represented by a note length sequence of note sequences.
Rhythm giving means to give data and each on the music time axis
A weighted pulse scale for each pulse point representing the music time (hence the pulse scale consists of several different weights).
Of multiple pulse points with
The pattern of weights formed by the point weights is
The weight pattern of the
Pulse scale generating means for generating a plurality of types of notes, and a sound producing pattern of the note length sequence of the note sequence expressing the rhythm.
Turn (starts sound generation for multiple notes that make up the note sequence above
The pattern formed by the timing) and each pulse
Weight pattern of the above
Rhythm analyzing apparatus comprising: the rhythm evaluation means for pattern matching using data, a. 6. A rhythm in which the rhythm is expressed by a note length sequence of note sequences.
And rhythm raw forming means for generating data to random numbers, heavy with each pulse points representing each musical time on music time axis
Pulse scale (hence pulse scale)
Consists of multiple pulse points with different weights.
The pattern of weights formed by the weights of multiple pulse points
(Hereinafter referred to as the pulse scale weighting pattern)
The pulse scale generating means and the sound generation pattern of the note length sequence representing the rhythm described above.
The turn and the weight pattern of the pulse scale are
Pattern matching using weight data of several pulse points
And the pattern matching result of the rhythm evaluation means is
If it is not within the range, the rhythm generator is re-newed again.
Of the new rhythm data.
Let the rhythm evaluation means perform pattern matching
The pattern matching result of the repetition means and the rhythm evaluation means is the above
If it is within a certain range, rhythm data at that time is generated
And a means for storing as a rhythm.