JPH0869283A - Music session device and system - Google Patents

Abstract

PURPOSE: To realize a music session having ample individuality from a musical odd feeling by making possible musical communication by introduction of respective parameters and having an individuality profile given from an individuality data base. CONSTITUTION: This device includes a music hearing section 12, a music forming section 14 and the individuality data base 16. Music primitive is recognized by a music primitive recognition agent group 18 of the music hearing section 12 and respective means extract the tension parameter TTV, music interest parameter MIV and partner insistence parameter PIV according to the individuality set in the individuality data base 16 from the recognized music primitive. Solo/accompaniment patterns are formed according to these parameters and the formation is controlled in the music forming section 14.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a music session device, and more particularly to a music session device and system for synthesizing performances in a music session in real time.

[0002]

2. Description of the Related Art Music has traditionally been a time art with a real-time characteristic, but the widespread use of records and music tape has established the recognition that music can be recorded. Originally, computer music mainly used tapes, but with the recent advancement and development of electronics, the direction of liveness, that is, human intervention in real time has been re-recognized. . Interaction between human and machine in music field
The (Interaction) approach can be broadly divided into interactive composing systems, automatic accompaniment systems, and music session systems.

The interactive composing system is a system in which an algorithm is provided in a computer so that the composition can be made by the interaction between human and machine. An automatic accompaniment system is a system in which a computer has musical score data in advance and performs an accompaniment to follow a human performance expression or a soloist performance with a mistake.

On the other hand, the music session system to which the present invention is directed is intended to construct a system simulating a session in which a plurality of people play music and enjoy improvisation. The major difference between the music session system and the automatic accompaniment system is the degree of dependence on the score data, and in the former, how to realize real-time music recognition and performance generation is an issue.

For example, Nishizima M. and Kijima Y. "Le
arning on Sense of Rythm with aNeaural Network- Th
e NEAURO DRUMMER ", Proc. ICMPC, (October 1989),
And Nishizima M. and Watanabe K. "Interactive Mu
sic Composer based on Naural Networks ", Proc. ICMP
The neuro-musician disclosed in C, pp53-56 (1992) uses a neural network to realize the interaction of drums and the interaction of a keyboard (keyboard musical instrument). In both cases, the set of the performance part corresponding to the first half and the latter part corresponding to the response, respectively,
The neural network is trained as an input part and an output part, and the model is used for multiplication. It is argued that interpolation in multiple playing patterns is a kind of creation. However, the constraint of the harmony part is implicitly used for the melody used for the learning set, and the extensibility that each data example must be learned from the beginning when handling individuality etc. Problems with
In other words, it has a problem that the inside is treated as a black box.

As a system considering the internal model,
Rowe R .: "Machine listening and composing with Cyph
er ", CMJ, Vol.16 No.1, pp43-63 (Spring 1992) and Penny
cook B. et al .: "Toward a ComputerModel of a Jazz
A system called Cypher, which is disclosed in Improviser ", Proc. ICMC, pp228-231 (1993), is known.
This system consists of a listener section and a player section. In the listener section, a multi-agent model is used to recognize chords, phrases, etc. in addition to sound density, range, and strength. From the outside, the user defines the response method consisting of three types of sequence data performance, composition program calling, and input MIDI signal serial processing for the recognition result obtained in the listener section. Perform reactions according to the performance. Cypher is a system aiming at interactive composing, and is not touched on the transmission and control of messages in a session.

As a system focusing on the sensual reaction in a jazz session, Waki et al. "Cooperative Performance System Considering Performers'Feelings-JASPER", Proceedings of Summer Symposium on Music Information Science, pp43-46 (198)
9) There is. JASPER is a system in which a computer performs accompaniment of drums and bass to a piano solo, targeting 12-bar blues. From the range, strength, and number of notes of the piano performance, a tension parameter corresponding to the excitement of the performer is selected, and when the tension parameter remains low, responsiveness is realized by changing the accompaniment pattern. ing.

[0008]

[Problems to be Solved by the Invention]
Is an interesting system that takes into account sensitive information, but even if it is played hard, even if it makes no sense musically, a similar answer will appear. This is because all the control variables are reduced to a one-dimensional signal called a tension parameter, and no musical recognition is performed in the process. In other words, in JASPER, although subjective recognition is taken into consideration, there is a problem that musical recognition is insufficient.

Therefore, a primary object of the present invention is to provide a new music session device. Another object of the present invention is to provide a music session device capable of not only subjectivity but also musical recognition.

[0010]

According to the present invention, there are provided a music listening means for receiving a music signal and outputting a listening parameter, a music generating means for receiving a listening parameter and generating music, and an individual characteristic of the music listening means and the music generating means. The music session apparatus is provided with a personality profile assigning means for assigning a profile.

[0011]

A music signal, for example, a MIDI signal is given to the music listening means. The music listening means has, for example, a multi-agent music primitive recognition means and extracts music primitives such as chords, chord progressions, and next phrases. Based on the music primitives and the personality profile, for example, the listening sensitivity parameter extraction unit may listen to the listening sensitivity parameter, for example, EC (Emotional).
Character) and tension parameters (Total Tension V
lue: TTV) is extracted. The listening sensitivity parameters are given to the music generating means, and the music generating means generates music with a performance pattern that depends on the listening sensitivity parameter and the individuality profile.

[0012]

According to the present invention, by introducing the listening sensitivity parameter, more musical communication becomes possible, and since the performance pattern is generated according to the listening sensitivity parameter and the individuality profile, the individuality is rich. Moreover, it is possible to realize a music session that is musically comfortable.

Further, a plurality of music session devices are provided,
By giving different personality profiles to each, a variety of music sessions can be realized. In this case, each device may be in charge of a different musical instrument or part. The above-mentioned objects and other objects and features of the present invention will become more apparent from the following detailed description of the embodiments with reference to the drawings.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Before the detailed description of the embodiments, the features of the jazz session considered in the music session device of the present invention will be described. In jazz performance, there is no fixed governing relationship between the players. As the performance progresses, the role of the asserted performance, the admitted performance, the ignored performance, or the newly claimed performance can change freely. In addition, the degree of reaction to the assertion is not constant, and responds to the strong assertion with a stronger performance, or leaves the form of the ensemble, and with the agreement of all the players, selects one player who particularly strongly wants the assertion, Others may turn to the viewer, moving to his solo performance. In the actual session, the overall control is performed by the individuality of the participants, and modeling of this individuality is extremely important.

As a matter of course, in the session device, it is necessary to understand what kind of music the partner is trying to perform. For example, it is necessary to monitor whether or not the other party is trying to take a solo, is trying to transpose, or is performing a high tension solo. Other,
Whether or not the music can be traced musically is also an important monitoring target.

In jazz, a peculiar sense of tension may be expressed by intentionally leaving a sense of polysemy. For example, in interpreting a melody line, IIm7-V7-Im
II ♭ 7 is often used as a substitute for V7 when aj7 progresses.
In the performance, when the situation at a certain point of time makes it impossible to know whether to pass V7 or II ♭ 7, the chord player either stops the performance for a moment or the common note of V7 and II ♭ 7 (3rd, 7th
It is common practice to play only the degree) and wait for the bass player to emerge. In such a state, the tension note (9
Since the chord player does not give the instruction of (13, 13 degrees), the degree of freedom is increased in the melody line, and the melody player naturally advances the performance while recognizing this fact. This created sharing of tension between the players, and this is one of the characteristics and the real charm of jazz performance whose main purpose is improvisation.

The simplest method of constructing a music session apparatus is to determine a performance schedule according to the performance style of a human player who observes it, and control all parts of the session. However, with this method, the performance pattern output from the computer becomes definite and a jazz performance with a unique sense of tension cannot be realized without introducing a certain kind of stochastic model.

Further, when the probabilistic model is entirely introduced, it is difficult to control it from the viewpoint of communication. In a jazz session, the player's personality, physical condition,
Simple advance meetings (scenarios) and cues about songs,
In addition, the surrounding conditions including the audience are important factors that influence the performance. Therefore, in the following embodiment, the player participating in the performance is regarded as an agent including a music listening unit, a music generating unit, and a personality profile, and an interactive model thereof is constructed.

The music session device 10 of this embodiment shown in FIG. 1 includes a music listening unit 12, a music generating unit 14, and a personality database 16. The music listening unit 12 includes a music primitive recognition agent group 18. The music primitive recognition agent group 18 recognizes music primitives by using a multi-agent model (a method of realizing a complicated function by combining agents that realize a simple processing of multiple inputs and one output in a network). . Music primitive recognition agent group 1
Each agent, which will be described later in detail in Section 8, has musical knowledge stored in a database, and this musical knowledge uses the harmony theory of the 1960s as a basis. For example, Ushio Sakai: “Jazz Organ Improvisation 1, 2”, Littow Music (1970) and Sadao Watanabe: “Jazz Study”, ATS (197)
0) etc. In this harmony theory, 1 rather than general jazz harmony
Practical rules of thumb as a professional musician are given priority. This is because, due to the essential nature of jazz, its harmony has changed over time, and it has never been established as a comprehensive theory, and the set of empirical rules is the theory.

Generally, when a person listens to music, he or she recognizes at various levels. Usually, the harmonics, rhythm, and melody are recognized after understanding the musical instrument type and sound source included in the accumulated acoustic signals. In the music session system 10 of the embodiment, the MID
Upon receiving the output of the I musical instrument, that is, the MIDI signal, the music primitive recognition agent group 18 first extracts the volume, the number of notes, the range, etc. as the acoustic perception level. The number of notes is
For example, it is given by the number of note-on per unit time. The musical perception level includes basic musical elements (musical primitives) such as chords, melodies, tension notes and rhythms.
Furthermore, as the higher music perception level, there are music primitives such as chord progression, tonality, and closure. The music primitive recognition agent group 18 performs these analyzes in real time. The music primitives extracted by the music primitive recognition agent group 18 are used not only as a basis for the physical constraints of the performance in the music generation unit 14, but also as a pre-stage of the sensitivity information extraction described later. The recognition of typical music primitives will be described below.

The chord recognition agent 20 receives the MIDI signal from the chord part player and the bass agent 22.
Possible as a chord candidate by inputting the root note information from the above and comparing it with a chord form stored in a knowledge database (not shown) formed in advance in the musical primitive recognition agent group 18 Select all. In the database, all chord forms that can be played are registered, from complete ones of closed form and open form to incomplete ones of root note, third and fifth. The actual performance data reaches the agent in a few milliseconds or more, even if the notes of the chord are played simultaneously. It may also be played as a distributed chord. By the retrograde recognition, the agent 20
The candidates are always output while verifying these. Code recognition agent 20 typically cooperates with base agent 22. This is because the chord recognition agent 20 cannot determine the chord name by itself without the root note information. For example, even if E, G, and B are played, Cmaj7,
It can be Em, G6, etc., and cannot be determined as a single chord without root note information. The chord information thus obtained is sent to the chord progression agent 24, where the chord progression is determined. Also,
The chord recognition agent 20 also gives information to the melody tension agent included in the listening sensitivity parameter extraction means 32 described later.

The main function of the bass agent 22 is to receive a MIDI signal from the bass player and output a possible root note candidate. This output is sent to the code recognition agent 20, as described above, on which the final determination of the codename is made. The function of the base agent 22 changes according to the music genre. For example, in 2 beats of rock, 8 beats, and swing, the chord component sound is mainly detected with respect to the input, and in 4 beats of the swing, the input is compared with the scale note. The base agent 22 also performs beat tracking in cooperation with the beat tracking agent 28.

The chord progression agent 24 determines chord progressions including reharmonization and tension resolve based on the chord candidates output by the chord recognition agent 20 and the root note information of the bass agent 22. The chord progression agent 24 examines all combinations of adjoining candidates, and excludes those that are subject to the prohibition rules. Next, with respect to the remaining combinations, the continuity as chord progression is investigated while going back in the past, and as an example, the chord progression having the longest continuation is selected. The chord progression agent 24 processes by local retrograde recognition, but for chord progressions whose use is established stylistically, such as cliche and 4th ascending progression, by using model-driven hypothesis test processing together. Performance is improved. After determining the chord progression, the chord progression agent 24 determines the tonic, subdominant, and dominant (hereinafter,
Called "T", "SD", and "D". ) Function code and scale. In order to determine the chord progression, the following five possibilities of progression are verified. However, the higher the list below, the higher the priority.

(1) Joseki progression (2) 4th ascending progression (3) Increase 4th ascending progression (4) Chromatic descending progression (5) Chromatic ascending progression (1) As for joseki progression, there are cyclic chord progressions and other He has the knowledge of the routine. Circular chord progressions are made up of more than a dozen types, and are progressions that are connected around four ascending progressions of 4 degrees. The joseki progression is, in a sense, the personality information of the chord progression agent 24, and is exemplified below.

(A) Dm7 / G-G7-Cmaj7 (b) Cm-Cm / B-Cm / B ♭-Cm / A-Cm / A ♭-Cm / G (c) C-A7 / C #-Dm (d) C-E7 / B-Am-Am7 / G D7 / F In recent performances, the bass line is also considered as a melody, and it is often played in a melody rather than a line. The example above shows that. (a) is IIm7-
Considered V7 progress too clear and used as a proxy. In (b), which is called a cliche, a chromatic downward-progressing baseline produces a unique effect in contrast to a chord that suppresses movement. (c) and (d) are the use of chromatic progression as a workaround from the frequently used four-degree ascending progression. Chromatic progression has a fairly strong character in chord progression, so the interpretation is given the highest priority. After determining the function code in this way, the chord progression agent 24 evaluates the chord in units of four bars and the position of the dominant motion to determine the tonality.

The beat tracking agent 26 is
Based on the information from the bass agent 22 and the performance information of the drums, the tempo and the beat are analyzed, and the change in the rhythm pattern is recognized. There are relatively many studies on the recognition algorithm for beat tracking. For example, Desain P. and Honing H .: "The Quan
tization of Musical Time: A Connectionnist Apploar
ch ", MIT press computer Music Jounal, 13 (3), pp56-
56 (1989), Rosenthal D .: "Machine Rythm: Computer E
moution of Human Rythm Perception ", Ph.D. Thesis,
Media Lab.MIT (1992), Allen P. and Dannenberg R .:
"Tracking Musical Beat in Real Time", Proc. ICMC,
pp140-143 (1990), etc. However, in this embodiment, considering the actual playing situation in each music genre,
Paying particular attention to the bass part player, M from the drums player
The IDI signal is only used supplementarily.

This beat tracking agent 26
Has a database of the characteristic rhythm patterns of each genre and their variations. Music genres are currently classified into swing (2 beats, 4 beats), rock, bossa nova, 16 beats, Latin, and samba. For example, in bossa nova, bass beat is used to maintain the beat, and sycopation is often used for instruments such as cowbells, cymbals, and rim shots. In addition, in the swing, the beat is generally maintained with cymbals, and variations are added with the snare drum or bass drum, but in that performance, it is performed while consciously avoiding concentrating on the beat. In the example, this point is taken into consideration. Then, the beat tracking agent 26 recognizes the beat based on the characteristic points on the rhythm pattern of each genre and the positions of the characteristic points detected from the input. However, in actual performance, the bass line often holds more accurate beat information, and variations are mainly extracted from the drums signal.

The recognition of a phrase or the like is calculated based on the output of the agent indicating the changing point of the agent at a lower level or the sense of break formed culturally, for example, the output when a dominant motion is detected. The next phrase prediction required for performance generation is achieved by the next phrase prediction agent 28. An example of the algorithm is shown below. The closure recognition agent 30
Is to add a change value with a change in the musical perception level (for example, if there is a jump between a certain sound and a certain sound in the melody, the jumping degree is the change value) to give a closure. To do.

Step S1: When the closure recognition agent 30 outputs a relative value of its strength at the beginning of a closure-when there is a large output above a predetermined threshold, the time is recorded and the next closure recognition agent When there is a large output of 30, the melody sequence in the meantime is registered in the working memory. The score of the melody sequence is calculated from the sum of the two outputs of the closure recognizing agent 30 that sandwiches the score. The melody sequences registered in the working memory may overlap.

Step S2: Matching of a plurality of registered melody sequences and a sequence of tones sequentially input. At this time, the expected degree of closure at the ending time of the melody sequence is calculated according to the degree of matching and the score of the melody sequence calculated previously.
The degree of matching is calculated using the onset time (identity of the rhythm) and the degree of change of the pitch in the melody (and the initial pitch).

Step S3: If the expected closure end time matches the large output from the closure recognition agent 30, the score of the melody sequence is increased. This indicates that the memory was repeatedly amplified. It is well known that human beings have some emotion when they listen to music, which also affects the production of performance during the session. Although there are various discussions regarding the definition of the emotional information, in this embodiment, from the viewpoint of the validity as the user interface of the system, the Emotional Characte
r (EC), Musical Interest Value (MIV), and Partner '
s Consider three of the Insistence Value (PIV). These are used as parameters to control the performance.

What is EC? Listening sensitivity parameter extraction means 32
That is, it is given by the melody tension agent as a data set consisting of adjective names such as bright, light, dark, heavy, etc. and a value representing its strength. Among them, Tension Total T, which is equivalent to tension
The tension value (TTV) is positioned as an important parameter when controlling the session, and the EC corresponding to each music primitive is calculated.
The tension parameter TTV is calculated by multiplying these by the individuality database 16, that is, the network parameter described in the individuality profile, and adding up. By using the individuality profile 16, individuality such as a harmony-oriented type or a rhythm-oriented type is realized. The partner claim parameter PIV is a factor representing the claim level of the other party output from the partner claim parameter extracting means 34, which will be described later, and is calculated such that its value increases when the solo is taken. The musical interest parameter MIV is a factor that is output from the musical interest parameter extraction means 36, which will be described later, and indicates a musical interest, and has a role to stop playing if it is too low.

The personality database 16 is used for listening to music,
This is a database that describes the individuality of performance generation. Different people have different ways of listening to music. The personality database 16 is a network parameter group that describes which musical elements are likely to respond to listening (for example, rhythm-oriented type or melody-oriented type), self-assertion parameters that determine whether it is easy to enter solo or stick to accompaniment, It has a parameter group for converting a listening sensitivity parameter into a performance sensitivity parameter, and a pattern describing the tendency of accompaniment pattern / solo pattern generation.

Output 16a: Outputs a parameter given at a level of 0 to 1 such as a melody-oriented type or a rhythm-oriented type. Output 16b: Gives weights for the three music interest parameters MIV to levels 0-1. Output 16c: Outputs the self-assertion parameter given at a level of 0 to 1.

Output 16d: Outputs the parameters corresponding to the degree of leadership and the degree of flaking at the level of 0 to 1. Output 16e: Outputs a parameter given at a level of 0 to 1 such as a melody-oriented type and a rhythm-oriented type during playing. Next, the general sensitivity extraction database 38 included in the music listening unit 12 serves to provide a basis for judgment to the listening sensitivity parameter extraction means 32, and how the music primitives output from the music primitive recognition agent group 18 are processed. It is a database that describes how it feels. The general sensitivity extraction database 38 mainly deals with what seems to be common in cultural areas. As an example,
The database for the melody is shown in Table 1 below.

[0036]

[Table 1]

Table 1 shows the emotions that occur between notes on the major scale and diatonic chords derived from the scale. The column shows the diatonic chord, and the row shows the frequency representation of the pitch used for the melody line, for example, I in the C major chord shows the do sound. C, t, a, and d at vertical and horizontal intersecting points express the emotions that occur at that time. The meaning is as follows. (C) chord sound (* is root sound), sense of stability (t) tension punch note, tension feeling (a) void note, prohibition feeling (d) discord, prohibition feeling (c) produces stability feeling, (T) has a tense feeling, and is in a whole tone or chromatic position with respect to 3 degrees at 9 degrees and 7 degrees at 13 degrees, and becomes stronger when this collision occurs. (A) is the sound that each scale has to avoid, which is a very short time to be used. (D) collides with the constituent sounds that determine the type of chord,
It has the same personality as (a) and has a strong personality as a mistone.

As described above, the listening sensitivity parameter extracting means 32 analyzes the melody line based on the output from the chord recognition agent 20 of the music primitive recognition agent group 18 and the MIDI signal from the melody part player. Including agents.
The melody tension agent (not shown) uses the template shown in Table 2 below to describe the emotional parameters described above.
(Emotional Character: EC) is extracted. This is a local tension value generated from each sound of the melody line with respect to the chord.

[0039]

[Table 2]

If attention is paid to the row of rows as a root note, there is similarity between each major and minor. Table 2 shows the types of each diatonic chord as they are and transposes until the root note becomes the same, and further groups into major chords, minor chords, and other chords. For example, for IIm7,
All rows in Table 3 are shifted to the left by the whole note. Table 3
Clearly, the similarity can be read for each group.
Major increased to 4 and increased to 4 and Minor increased to 5 and increased 5.
Only differences can be seen. The template shown in Table 2 is called a melody tension filter. Also,
For the tension value on the melody line, select the line corresponding to the chord from the melody tension filter,
It is interpreted by acting on the melody. Therefore, even if you are not sure about the tonality,
It is one of the features of this embodiment that the interpretation of the tension value is performed. In addition, the chord progression EC agent (not shown) also uses the listening sensitivity parameter extraction means 26.
And its function is to detect the ECs presented by the chord progressions from the chord progressions from the chord progression agents 24 of the music primitive recognition agent group 18 based on their harmonic sense. This EC value is
It shows local emotions within a few measures. For example, Dominant Motion brings a feeling of end, IIm7-
II ♭ 7-Imaj7's chord progression creates a feeling of darkness. By macro recognition, EC is detected from a function code, 4 times ascending, a diatonic code, a dominant motion, a temporary modulation, etc. An example of an EC is shown below.

Et: Stability due to tonic Et0: Loss of tonality Eds: Distance sensed when going from tonic to subdominant Ed: Expectation of tonic due to dominant E4: Dynamic feeling due to 4th ascending Edi: Diamond tonic code Stability due to Ev: Termination due to dominant motion Ei: Development due to false termination E25: Expectation for tonic due to progression of II-V Eb: Dentation due to chromatic progression Ek: Leap feeling due to change in tonality E37: Component sound Seventh code only 3 or 7 degrees Ecc: Code including chromatic or whole-tone collision Each such EC is "0" or from the chord progression EC agent of the listening sensitivity parameter extraction means 32 at the time when chord progression is established. It is output as a parameter value of "1".

As described above, the listening sensitivity parameter extracting means 26 extracts the adjective information of the input performance. Adjective information is a name (tension, depression, etc.)
The musical information (melody, tonality, chords, etc.) that causes it is represented by a tuple of its value, and based on the general emotional extraction database 38, the emotional information that is considered to be culturally common is extracted. I do. In this example, when a minor is detected, a tuple value with the name of darkness is added. On the other hand, not all can be said to be culturally common regarding how music is felt. For example, some people emphasize melody, while others emphasize rhythm. Personality database 1
From 6, the parameter group indicating which music information should be weighted by an individual to listen to music is input. The overall value of the adjective information corresponding to the name is obtained by the product-sum operation of this parameter group and the value extracted by the general affective extraction database 38. Here, tension (TT: TT) is an element that controls the session.
I am paying attention to V). The algorithm will be described below by taking the processing of information corresponding to tension as an example.

Step S1: Receive output from the music primitive recognition agent group 18. Step S2: When each music primitive (for example, melody) is input, the tension parameter corresponding to the melody is calculated by referring to the one registered in the general sensitivity extraction database 38. Step S3: In step S2, the tension (TV: Tension Value) corresponding to each music primitive such as melody, rhythm, and harmony is calculated. In contrast,
The tension parameter corresponding to the entire music at a certain time is calculated by multiplying by the parameter TTV input from the individuality database 16 which priority is given to the primitive and adding them together.

Step S4: The tension parameter TTV is sent from the listening sensitivity parameter extracting means 26 through the line 12. As another example of EC, an example of extracting a parameter of profound feeling will be shown. It is thought that this profound feeling is culturally common, and is born from the next step. Cm-Cm / B-Cm / B ♭-Cm / A-Cm / A ♭-Cm / G The extraction of this emotion requires the continuous extraction of the same chord and the simultaneous extraction of the bass semitone downward progression. Also Cm is Cmad
If it is played with the chord d9 and there is a collision between D and Eb,
The tension from now on emphasizes the profound feeling.

The partner claim parameter extraction means 34
The self-assertiveness of the outside performer is extracted. It is not required to be completely independent of the tension described later, but solo /
The player's self-assertiveness is detected depending largely on whether or not the accompaniment is being played. An example of a self-assertion degree extraction algorithm is shown below. As shown in FIG. 2, this means 34 has three agents (not shown) for extracting melody-likeness, chord-stroke-likeness, and arpeggio-likeness, and uses an eighth note obtained by statistical processing for each measure. The number of notes, the number of changing notes, the average pitch of one bar, the number of chord tones / tension notes, etc. are extracted. If the highest melody among the extracted degrees is a melody, it is recognized as an accompaniment if the solo and chord stroke / arpeggio are high. Then, the partner claim parameter extracting means 34 determines the partner claim parameter PIV.
Is output.

Regarding the extraction of the musical interest parameter, the one recognized macroscopically (macro recognition), the one motivated by the deviation from the tendency to make a deviation (anticipation recognition), and the one when the musical progression can be understood There are three types of things that extract joy (reverse recognition). These will be described below. Viewers are aware of the music they are listening to, looking for similarities to their own musical experience. we,
When you hear a known melody, it reminds you of the moods and emotions that came to your mind when you heard it before. It may be wrong to think that all musical elements (fragments) evoke certain emotions, but there are certain adjectives for the musical elements that are contained in the music that each race has and have a cultural context. Has implications. Poplar song composers may intentionally use such events as tags.

The above macro recognition method can be realized on a computer by using dynamic pattern matching. So far, there is an example of research that analyzes global information that indicates the personality of a performer and uses it in the composition generator. In this system, the emotional parameter E
Extraction of part of C, the partner claim parameter PIV, and the musical interest parameter MIV is thus realized macroscopically. In the extraction of EC, in addition to the above-mentioned melody analysis method, there are things that are considered to be almost common for cultural reasons such as a feeling of solution by dominant motion or a feeling of depression in cliche. The recognition of the musical interest parameter MIV is triggered when a so-called favorite pattern comes.

Therefore, the macro recognizing means 40 extracts the interest in the music primitive by template matching according to the information from the personality database 16. That is, the macro recognizing means 40 determines whether or not the user is interested in a particular music element or pattern. As an example, we show the extraction of preference for the collision of semitones and fulltones when inferring chord progression. This collision creates tension. The code information has the following structure in the database to improve search efficiency.

Example Cmaj7 constituent tone data string (1) C, E, G, B 0, 4, 3, 4 (2) E, G, B, C 8, 3, 4, 1 (3) G, B, C, E 5,4,1,4 (4) B, C, E, G 1,1,4,3 All chord data does not record the actual pitch name, but the root note to improve search efficiency. The decision routine is divided into other parts, and the pitch difference between adjacent parts is stored as the number of semitones. With such code information, the number of searches is reduced to one-twelfth, which speeds up the search, and it is possible to search for all expanded forms, including those in which the root sound is omitted. As shown in the example, the first number in the data string indicates the position of the root note, which is the difference from the lowest note. For example, in Cmaj7 (1), the first constituent sound is C, there is no pitch difference, and it is 0. In (2), the number of semitones starting from E to the root C is 8, and the second and subsequent numbers are numerical sequences showing pitch differences. The collision is extracted by investigating the existence of 1 or 2 indicating the pitch difference between the semitone and the whole tone in the data string of the chord database selected as a candidate.

As one of the foundations of music theory,
Implication and realization with musical delays and deviations
There is something in which emotion is created by (Realization) and music is recognized. In order to handle gradually increasing types of information, it is possible to predict what the musical meaning and result should be at a point in time when delays and deviations are known, and whether the viewer knows the style of the music. It is necessary to know the result of reaching the song. Although there is some debate as to whether or not a completely known development creates emotions, it is a model that explains the appreciation of music in which the format can be expected to some extent.

In order to realize this model, a method of grasping a musical tendency, making a prediction, and outputting a musical interest parameter MIV in response to the deviation can be considered. There are various levels of expectations, but this system implements a part that requests a tonic when a dominant, which is considered to be the basis of Western music, is detected. Basically, the MIV is continuously output until the realization is detected, and a large MIV is output at the time of realization, but when the realization is not detected for a long time, the forgetting process is performed. Moreover, although it may be slightly different from the musical emotion, the recognition of the change point (the continuity) of the input data is also based on the intention and the realization.

Therefore, the anticipation recognizing means 42 handles anticipation in music. For example, expect a tonic four times above when dominant 7 is detected. Output while expecting (less than forgetting time) and when the expectation is realized. This expectation increases during expectation, but conversely, if there is no realization during that time, output is not performed. Further, in the next-phrase prediction agent 28 described above, the output is also performed when a theme part appears. Below is the dominant 7
An algorithm for the case of expecting a tonic up 4 times when is detected is shown.

Step S1: Code recognition agent 2
The output of dominant 7 is received by 0. Memorize the time. Step S2: The elapsed time from when the dominant 7 is entered is detected. Step S3: If the elapsed time exceeds the set time (forgetting time), the process ends.

Step S4: It is judged whether or not a tonic comes from the code recognition agent. When a tonic arrives, an evaluation function having the elapsed time and the target element as variables, for example, a value obtained by integrating the former and the latter is output to the musical interest parameter extraction means 36. If the tonic does not come, the process returns to step S2. While the above method is attractive, it is unlikely that all viewers will perceive all musical implications and understand the progression. We may perceive the realization without intent so that we can enjoy unknown music without any knowledge of style. Especially in jazz, which is mainly for improvisational performance, sometimes chord progressions or melody are performed in which the tonality is intentionally vague. In this case, the viewer naturally feels joy, but this joy does not gradually increase, but has the character of being released at the moment of understanding. Moreover, when there is a long time gap between intention and realization, it is difficult to handle emotional changes by the above method.

In order to extract such a musical interest parameter, the retrograde recognition means 44 uses a model of performing retrograde recognition of the music primitive and extracting the musical interest parameter at that time. Retrograde recognition is a method of recognizing musical primitives that leave local ambiguity.
Basically, recognition is performed in the following steps. (1) Put a hypothesis where you cannot understand and keep it in temporary memory.

(2) Evaluation is performed retroactively from the time when a new understanding is possible. At the stage of (1), local interpretation is performed, but the overall connection is not understood. In (2), the overall connection (meaning) is evaluated. At this time, the sensitivity information is extracted in the form of outputting the musical interest parameter MIV when the meaning is successfully evaluated. At present, the above-mentioned processing is realized regarding recognition of chord progression and extraction of musical interest parameter MIV.

That is, the retrograde recognizing means 44 accumulates music primitives when there are a plurality of possible interpretations, and outputs hit points when one pass is confirmed. The macro recognition means 40, the anticipation recognition means 42, and the retrograde recognition means 44 in the music listening unit 12 are realized in each agent.

Then, the musical interest parameter extraction means 3
Reference numeral 6 receives the parameters indicating the musical interest from the macro recognizing means 40, the anticipation recognizing means 42 and the retrograde recognizing means 44, and performs the calculation (sum) to calculate the parameters for the musical interest, and Output. It is used to output a measure of whether or not it is musically understandable.

The logical constraint extraction means 46 selects one of the states of the musical primitive recognition agent group 18 which becomes a musical constraint at the time of performance generation, and the music generation unit 1
4 to the accompaniment pattern creating means 48 and the solo performance pattern creating means 50. The logical constraint extraction means 46 also functions as a monitoring mechanism for checking whether the entire performance is proceeding well. For example, in the swing, when the bass part changes from a performance centering on chord constituent sounds to a performance heavily using scale sounds, it is known that the session has shifted from the theme to ad lib.

The solo / accompaniment / stop selection means 52 included in the music generation section 14 is the partner claim parameter extraction means 3
4 and the output PIV and MIV from the musical interest parameter extraction means 36, the selection of solo / accompaniment / stop is performed. For example, the musical interest parameter extraction means 36
If the output MIV from MV continues to be lower than the threshold value, the performance is stopped because the music cannot be understood. Otherwise, depending on the partner assertion degree PIV from the partner claim parameter extracting means 34 and the self-assertion parameter from the personality database 16, selection is made as to whether to perform solo or accompaniment. From each output of the solo / accompaniment / stop selection means 52, the accompaniment pattern creation means 48,
Solo performance pattern creating means 50 and performance synthesizing means 54
To a solo / accompaniment / stop switch signal. The accompaniment pattern creating means 48 and the solo performance pattern creating means 5
The output from the personality database 16 is also sent to 0. This parameter is used as a parameter for interaction in the solo performance pattern creating means 50.

The performance sensitivity parameter generating means 56 receives the sensitivity parameter EC in the listening from the listening sensitivity parameter extracting means 32 and expresses the accompaniment and solo expressions in the accompaniment pattern creating means 48 and the solo performance pattern creating means 50. Determine the parameters to determine. The sensitivity parameter from the listening sensitivity parameter extraction means 32 handles the sensitivity parameter at a past time, but here, it is converted into a sensitivity parameter for deciding future performance by observing the transition tendency. From the individuality database 16, the output 16d gives a parameter corresponding to the degree of leadership or the degree of misunderstanding. The degree of leadership is used, for example, in order to facilitate the leadership by increasing the performance tension parameter corresponding to the function of trying to raise the tension when the tension parameter is low among the listening sensitivity parameters. The degree of flair has a function of converting a listening sensitivity parameter into a performance sensitivity parameter by its negative logic. Regarding the conversion from the listening sensitivity parameter to the performance sensitivity parameter, various control parameters are not unique and are used as control parameters of conversion probability.

The accompaniment pattern creating means 48 receives the output from the performance sensitivity parameter creating means 56 and creates an accompaniment pattern. For example, when the performance tension parameter is large, a musical performance with large excitement is generated. With regard to the accompaniment generation method, two kinds of sequence data calling or real-time accompaniment generation can be selected. In the latter, for example, a mechanism for increasing the number of strokes (the number of notes in the accompaniment) when the performance tension parameter is large is realized.
Further, the table as shown in Table 1 is reversely used from the general affective expression database 58.

An example of an algorithm in the accompaniment pattern creating means 48 will be shown below. Step S1: The performance sensitivity parameter is received from the performance parameter generating means 56. Step S2: A tension value corresponding to each primitive is calculated by multiplying each of the primitives (sound number, harmony) by the performance-oriented parameter obtained from the individuality database 16.

Step S3: Select an element controlled by sensitive information from the elements limited by the logical constraint output from the logical constraint extraction means 46,
Determine the accompaniment. As an example of the determination method, a normal distribution of parameters corresponding to each primitive is prepared, and the appearance rate of the primitive when a certain parameter is given is read as the value of the normal distribution. The primitive to be used is selected according to the appearance rate for each primitive.

The accompaniment pattern creating means 48 can also create an accompaniment pattern as follows. (1) Calling sequence data The database has accompaniment phrases from two to several chord progressions. In addition, a plurality of phrases are prepared that are correlated with the performance tension parameter.
A phrase is selected together with the data in the personality database 16 to generate an accompaniment.

(2) Real-time accompaniment generation Based on an output from a random number generator (not shown), a note having a rest, a note length and a pitch is obtained. Normalization of the pitch is added to this string by the chordal constraint from the chord progression at that time, and the string is added to the upper or lower part of the obtained string based on the constraint to generate an accompaniment phrase. To do. Also, after normalizing the phrases already generated in the song in the same way,
The method of reuse is also used together.

It should be noted that which of the above methods (1) and (2) should be used may be changed by a performance tension parameter or the like. The general affective expression database 58 stores the same database as the general affective extraction database 38. As for the logical constraint, for example, the walking base is considered. When chord progression is specified, the range in which the next chord can be taken is limited. At this time, it is necessary to avoid arranging a bass pattern such that the root sound of the limited chord becomes the fourth beat, and such processing is realized by the rule base.

In addition, the tension parameter of each primitive is stored in the form shown in FIG. The solo performance pattern creation means 50 is the performance sensitivity parameter generation means 5
The output from 6 is received and a solo performance pattern is created.
For example, when the performance tension parameter is large, a musical performance with large excitement is generated. As for the solo generation method, as in the case of accompaniment pattern generation, it is possible to select two types of sequence data calling and real-time accompaniment generation. In the latter, for example, a mechanism for increasing the number of strokes (the number of notes in the accompaniment) when the performance tension parameter is large is realized. Further, the table as used in Table 1 is reversely used from the general affective expression database 58.

In this solo performance pattern creating means 50,
A performance pattern is generated as follows. (1) Calling sequence data The database has accompaniment phrases from two to several chord progressions. In addition, a plurality of phrases are prepared that are correlated with the performance tension parameter.
A phrase is selected together with the data in the personality database 16 to generate a performance.

(2) Real-time solo performance generation Based on an output from a random number generator (not shown), a note having a rest, a note length, and a pitch change is obtained. The pitch is normalized to this string by the chordal restrictions from the chord progression at that time to generate a solo performance. Also, the phrase already generated in the song is reused after being normalized in the same manner.

Which of the above methods (1) and (2) should be used may be changed by a performance tension parameter or the like. The performance synthesizing means 54 synthesizes the accompaniment and solo performance outputs from the accompaniment pattern creating means 48 and the solo performance pattern creating means 50 to generate a performance.
When the CPU is limited, the solo / accompaniment / stop selection means 52 switches the accompaniment pattern creation means 4
8 or the solo performance pattern creating means 50 is selectively executed. When multiple CPUs can be used, both perform processing in parallel and perform solo / accompaniment /
According to the output from the stop selection means 52, the performance synthesis means 5
Select one in 4.

Experimental Results The music session apparatus 10 of the above-described embodiment is configured as a distributed processing by a plurality of personal computers, but it goes without saying that this may be processed by a single computer. However, in the experiment, each agent was formed in the same computer, MIDI signals were given from the Hammond organ, and the upper keyboard, lower keyboard,
And the bass keyboard are the melody, chord, and bass parts, respectively. In addition, the current system does not have a separate database for the songs that are input, and starts to interpret the tones when the tones are unknown.

Experimental Example 1 In FIG. 4, information from the base agent was not given in order to confirm the capabilities and effectiveness of the chord recognition agent and the chord progression agent. In FIG. 4, E
C, TTV, PIV, and MIV only show results related to chord progression. The TTV is obtained by processing and summing the EC values as described above. PIV is activated by non-diatonic chords, chromatic collisions of chord constituent tones, and continuation of identical bass tones, and MIV is activated by the appearance of V7, which is expected to be dominant motion.

In the example of FIG. 4, 8 beats and a tempo are about 100.
Is. This song has an odd number of repetitions of 7 bars up to Sabi, and it is a strange song, but it has a beautiful melody that you can't even notice. It is a song that is coded on the basis of four ascending progressions and incorporates the counterpoint method, which is a classic technique in Sabi. Figure 4 shows after the theme performance
These are the results of analysis of several measures that have been improvised. The horizontal axis represents time, and the vertical axis represents candidate codes output by the chord recognition agent, and broken lines and solid lines show how the chord progression agent interprets the chord progression. Below that, E
The analysis result of C agent is shown. The dashed line attached to the EC value indicates that the EC value has been determined for the codes at the start and end thereof. In addition, the T due to the effects of macro recognition, anticipation recognition, and retrograde recognition in the extraction of emotional information.
TV, PIV, and MIV are activated. The musical notation shown in the notation is not the actual performance data, but is normalized by the action of the chord agent.

First, since the system does not know the tonality, Fm
I chose two, aj7 and Dm9. Next, we have four candidates, but Gm9 and Em7-5 have exactly the same constituent sounds,
C9, G ♭ 7-9-13 are candidates for which the root note is omitted. By interpreting the next A7 ascending 4 times, Em7-5 was selected from the previous candidates. Then, Dm9, Gm9, C7, and Fmaj7 are selected as candidates from the ascending four times. Here, the tonality is decided to be F major, and the first Fmaj7
I have chosen. Also, an appropriate EC is output according to the chord progression decision.

Looking at the results of interpretation of the progress thus far, basically, the progress of SD-DT is repeated twice. Up to 7 / 8th of the code is tied up four times,
It shows a natural flow without difficulty. The third A7 is not a diamond tonic code for the F major scale,
Including C # that is not on the scale in the constituent sounds, where Dmaj
I expected a new development such as, but as a result, it progressed to the relational Dm7, giving a calm atmosphere of a diatonic. In this way, the chord recognition agent and the chord progression agent were able to interpret fairly accurately in the songs in which the ascending progression of four times was varied.

Experimental Example 2 Next, an experimental example of two music pieces interpreted by all agents including the base agent will be shown. 5 is 4
The song is a variety of ascending progression, the original song is Chanson. It is a song often played in jazz due to its beautiful measure and simple structure. The experiment shows the results of several bars from a theme played on a swing (2 beats) with a tempo of about 120. As in Experiment 1, the chord recognition agent uses E ♭ based on the possibility that the tonality is G minor.
Although maj7 and Cm9 are selected, based on the information of the base agent, the chord progression agent selects Cm9, and thereafter, ascending 4 times (including 4th increase), F9, B ♭
It is correctly interpreted as maj7, E ♭ maj7, Am7-5, D7-9, Gm9. According to the interpreted results, only the ascending progression of the diatonic four times is shown, and the dark and calm thoughts that are repeated SD-DT without leaving G minor are shown.

There are some notable points in the interpretation of the chord progression agent. It was not selected because it used the information of the base agent, but in the first half and the second half, we found a chromatic descending progression starting from C and A, which the player did not even notice. The following progression is created by changing the bass sound to B ♭ at E ♭ maj7. Cm9-B7-7-9-13 -B ♭ maj9-E ♭ maj9 / B ♭-Am7-5 -A ♭
7-9-Gm9 These long chromatic descents have the effect of creating a calm atmosphere and a solemn mood, especially in Bossa Nova. From the above, it can be seen that the system of the embodiment is also useful as an arrangement support device.

Experimental Example 3 FIG. 6 was composed as an insertion song of the movie “Casablanca”. In the original, Fm7 -B ♭ at the beginning
7 is repeated twice, but when played as jazz, various progressions have been devised to escape redundancy. In the experiment, I played as follows.

Fm7 -B ♭ 7 -D ♭ 6 -B ♭ 7 / D -E ♭ maj7-Fm7-Gm9-C9 By using D6 with the same two constituent sounds as the substitute chord of Fm7, the bass sound is D ♭ -D -E ♭ -F -G is intended to be a powerful chromatic descent. Each agent provided sufficient knowledge about chromatic progression as individuality information, but D ♭ 6 could not be selected. The reason for this lies in the knowledge of surrogate chords given to agents and special chord progressions. The proxy code was limited to those with the same function, such as the proxy relationship between G7 and D ♭ 7. This is because if the same chord is recognized as a substitute for up to three constituent tones, the number of candidates increases abnormally. Also, by giving the chord progression agent knowledge such as VII ♭ 6-V7 / VII-I, it is possible to interpret this song, but there are many standard songs with such peculiar chord progressions. , If all the songs are targeted, the information will be huge and the agent will be overloaded.

As described above, according to the music session device 10 of the embodiment, as shown in Table 3, the chord progressions intended by the performer were extracted in many songs.

[0082]

[Table 3]

A plurality of the music session devices 10 described above can be used to construct a session system as shown in FIG. That is, by preparing a unique individuality profile (individuality database 16 in FIG. 1) of each music session device 10, a music session according to each individuality can be realized. Furthermore, by configuring the system using a plurality of devices 10,
Since music is generated in real time without the score, B
It will be effective as GM (Back Ground Music).

[Brief description of drawings]

FIG. 1 is a functional block diagram showing an embodiment of the present invention.

FIG. 2 is an illustrative view showing each agent in a partner claim parameter extraction means included in the embodiment in FIG.

FIG. 3 is an illustrative view showing a general affective expression database included in the embodiment in FIG. 1;

FIG. 4 is an illustrative view showing a performance situation and each parameter in Experimental Example 1.

5 is an illustrative view showing a performance situation and each parameter in Experimental Example 2. FIG.

FIG. 6 is an illustrative view showing a performance situation and each parameter in Experimental Example 3;

FIG. 7 is an illustrative view showing one example of a music session system using an embodiment.

[Explanation of symbols]

 10 ... Music session system 12 ... Music listening unit 14 ... Music generation unit 16 ... Personality database 18 ... Music primitive recognition agent group 32 ... Listening sensitivity parameter extraction means 34 ... Partner assertion parameter extraction means 36 ... Music interest parameter extraction means 38 ... General sensitivity extraction database 40 ... Macro recognition means 42 ... Anticipation recognition means 44 ... Retrograde recognition means 46 ... Logical constraint extraction means 48 ... Accompaniment pattern creation means 50 ... Solo performance pattern creation means 52 ... Solo / accompaniment / stop selection means 54 ... Performance synthesizing means 56 ... Performance affective parameter generating means 58 ... General affective expression database

Claims (8)

[Claims] 1. A music listening means for receiving a music signal and outputting a listening parameter, a music producing means for producing music by receiving the listening parameter, and an individuality for giving a personality profile to the music listening means and the music producing means. A music session device comprising profile assigning means. 2. The music listening means includes a music primitive extracting means for extracting a music primitive based on the music signal, and a listening sensitivity parameter extracting means for extracting a listening sensitivity parameter based on the music primitive and the personality profile. The music session apparatus according to claim 1, further comprising: the listening sensitivity parameter is provided to the music generating means. 3. The music generation means generates a performance parameter based on the listening sensitivity parameter and the personality profile, and a performance parameter generation means that generates a performance pattern according to the performance sensitivity parameter and the personality profile. Claim 2 including
The described music session device. 4. The generation means includes accompaniment pattern generation means and solo performance pattern generation means, and the music generation means includes solo / accompaniment control means for controlling the accompaniment pattern generation means and solo performance pattern generation means. Item 3
The described music session device. 5. The music listening means includes claim parameter extracting means for extracting a partner claim parameter representing a claim level of a partner, and the solo / accompaniment control means according to the partner claim parameter, and the accompaniment pattern creating means and the solo accompaniment pattern. The music session apparatus according to claim 4, which controls the performance pattern generation means. 6. The music listening means includes a music interest parameter extraction means for extracting a music interest parameter based on the music primitive and the individuality profile, and the solo / accompaniment control means further includes the music interest parameter. The music session apparatus according to claim 5, wherein the accompaniment pattern generation means and the solo performance pattern generation means are controlled accordingly. 7. The music listening means includes a logical constraint output means for outputting a logical constraint based on the music primitive, and the accompaniment pattern generation means and the solo performance pattern generation means take the logical constraint into consideration. 7. The music session device according to claim 4, which generates an accompaniment pattern and a solo performance pattern. 8. A music session system comprising a plurality of music session devices according to claim 1, each of which is provided with a different personality profile.