CN111613199B - MIDI sequence generating device based on music theory and statistical rule - Google Patents

Abstract

The invention discloses a MIDI sequence generating device based on music theory and statistical rules, which comprises a computer system, wherein the computer system is configured to: receiving a chord information sequence, a rhythm type and a tone; initializing the initial probability of the pitches of 12 tones in one octave according to the received tone; chord correction, melody motion correction, repeat sound correction and sound threshold correction are carried out on the tone pitch initial probabilities of the 12 tones according to chord restriction, melody motion restriction, repeat sound restriction and sound threshold restriction, and the tone pitch probability distribution of the 12 tones is obtained; selecting a note from the pitch probability distribution in sequence according to the sequence of the pitch from high to low to carry out condition screening, and adding the note meeting the screening condition to the generated melody; the generated melody and rhythm are integrated to generate a MIDI file. The MIDI sequence generating apparatus is capable of generating a multiplicity of MIDI sequences from a given melody.

Description

MIDI sequence generating device based on music theory and statistical rule

Technical Field

The invention relates to the technical field of music science and technology, in particular to a MIDI sequence generating device based on music theory and statistical rules.

Background

MIDI is an industry standard electronic communication protocol that defines different notes and control signals for various types of electronic musical instruments and other playing equipment. Therefore, MIDI is a method for symbolizing music and is widely used in music production.

With the development and popularization of artificial intelligence technology, computer composition becomes a new branch in music production, and symbolic music composition, i.e. MIDI sequence generation, becomes an important component thereof. Many previous MIDI sequence generation efforts incorporate different approaches, which generally have respective generation backgrounds and advantages.

Currently mainstream MIDI sequence methods can be generally divided into two categories: the first is a method of generating a musical composition based entirely on music rules, i.e., a rule for extracting a melody from an existing musical composition and creating a MIDI based on the rule. This approach can generate music that closely resembles existing music. For example, the EMI virtual music generation system of David Cope, the chord bach music generation expert system of Ebciogin. It is very difficult to extract all the music theory rules of a certain style from the music and apply the music theory rules in the generation, and the music theory rules based on the style can be continuously evolved over time.

The second method is a method for generating MIDI by using machine learning in artificial intelligence and deep learning techniques. The method can make the generated music more diverse, and can generate the music end to end. Typical examples are the music creation system MusicVAE multitrack by Adam Robert et al, the MuseGAN symbolized music and accompaniment creation system by Dong Hao-Wen et al, and so on. However, the existing work has many defects, such as the generation quality of music is generally not high, the generated single-track MIDI notes have more wrong tones, and the harmony between the generated multi-track MIDI tracks is not high.

In summary, the conventional method for generating MIDI sequences cannot ensure the diversity of generating MIDI under the controllable melody generation. And the uniqueness of the existing MIDI generation framework also becomes a bottleneck of the MIDI sequence generation problem.

Disclosure of Invention

The invention aims to provide a MIDI sequence generating device based on music theory and statistical rules, which can generate diversified MIDI sequences according to given melodies.

The technical scheme of the invention is as follows:

a MIDI sequence generation apparatus based on music theory and statistical rules, comprising a computer system configured to:

receiving a chord information sequence, a rhythm type and a tone;

initializing the initial probability of the pitches of 12 tones in one octave according to the received tone;

chord correction, melody motion correction, repeat sound correction and sound threshold correction are carried out on the tone pitch initial probabilities of the 12 tones according to chord restriction, melody motion restriction, repeat sound restriction and sound threshold restriction, and the tone pitch probability distribution of the 12 tones is obtained;

selecting a note from the pitch probability distribution in sequence according to the sequence of the pitch from high to low to carry out condition screening, and adding the note meeting the screening condition to the generated melody;

the generated melody and rhythm are integrated to generate a MIDI file.

The chord information sequence contains a sequence of chords and corresponding beat numbers, which often presents a tone, so that the tone can be automatically generated according to the input chord information sequence.

Preferably, the MIDI sequence generating apparatus is provided with a popular music tempo type therein, and directly synthesizes any popular music tempo type to generate a MIDI file.

In initializing the pitch initial probability, the size scale corresponds to different pitch probabilities. Thus, when the key is major, the initial probability of pitch for 12 tones is:

sound	1	#1/b2	2	#2/b3	3	4
							Initial probability of pitch	0.184	0.001	0.155	0.003	0.191	0.109
Sound	#	4/b5	5	#5/b6	6	#6/b7								7
							Initial probability of pitch	0.005	0.214	0.001	0.078	0.004	0.055

When the key is a minor key, the initial probability of the pitches of the 12 tones is:

sound	1	#1/b2	2	#2/b3	3	4
							Initial probability of pitch	0.192	0.005	0.149	0.179	0.002	0.144
Sound	#	4/b5	5	#5/b6	6	#6/b7								7
							Initial probability of pitch	0.002	0.201	0.038	0.012	0.053	0.022

Preferably, the chord modification in accordance with the chord constraint includes:

setting a chord correction coefficient, wherein the chord correction coefficient range is 2.3-3.5;

and multiplying the tone pitch probability of the chord component tone corresponding to the chord information by the chord modification coefficient according to the received chord information to obtain a chord modification result.

Preferably, the melody motion modification according to the melody motion constraint includes:

when the predicted sound and the generated sound have different musical interval differences, multiplying the pitch probability of the sound to be predicted by a melody motion constraint coefficient, wherein the melody motion constraint coefficient is obtained by the following formula:

F(x，root)＝f(x，root+2，0.5)*0.2+f(x，root+4，1)*0.05+f(x，root-2.3，1)*0.1

+f(x，root-5，0.8)*0.06+f(x，root+7，0.8)*0.03+f(x，root+1，0.7)

*0.05+f(x，root-8，3)*0.02

where F (,) represents the melody motion constraint coefficient, x represents the predicted pitch, and root is the generated pitch.

Preferably, the repeating tone modification according to the repeating tone constraint includes:

setting a repeating sound punishment coefficient, wherein the repeating sound punishment coefficient range is 0.7-0.9;

and comparing the predicted tone with all generated tones, and if the predicted tone pitch is contained in the set consisting of all the generated tone pitches, multiplying the tone pitch probability of the predicted tone by the repeat penalty coefficient to obtain a repeat correction result.

Preferably, the threshold modification according to the threshold constraint comprises:

setting a threshold range, a threshold central value and a threshold standard deviation value, and generating a normal distribution with the threshold central value as a mean value and the threshold standard deviation value as a variance according to the threshold range, the threshold central value and the threshold standard deviation value;

and multiplying the pitch probability of the notes with the pitch within the range of the threshold by the probability value of the normal distribution of each note, and normalizing to obtain the output of the range judgment.

Preferably, the following conditional filtering is performed for selecting a note in the pitch probability distribution:

if the predicted note is not in the key, marking the note as wrong note and clearing;

if the predicted note and the chord form a pitch reduction interval, directly removing the note;

if the predicted note and the generated note exceed eight-degree big jumps or form continuous same-direction big jumps with the previous two notes, the note is directly eliminated;

if the predicted note is that the second time the outer tone appears in a certain bar, the note is cleared directly.

Compared with the prior art, the invention has the beneficial effects that:

the MIDI sequence generating device based on the music theory and the statistical rule can generate diversified MIDI sequences under the condition of ensuring controllable melody generation according to the given melody, rhythm and tone.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.

FIG. 1 is a flowchart of MIDI sequence generation based on music theory and statistical rules according to an embodiment of the present invention.

FIG. 2 is a pitch initial probability distribution diagram at major key provided by an embodiment of the present invention;

FIG. 3 is a pitch initial probability distribution diagram for a minor key provided by an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the detailed description and specific examples, while indicating the scope of the invention, are intended for purposes of illustration only and are not intended to limit the scope of the invention.

As shown in FIG. 1, the embodiment provides a MIDI sequence generating apparatus based on music theory and statistical rules, comprising a computer system configured to:

step 1, receiving a chord information sequence, a rhythm type and a tone.

Wherein the chord information sequence is a sequence containing chords and their corresponding beat numbers. The chord progression is consistent with tone, a chord progression list is arranged in a common device and can be customized, the data type of the chord progression list is str and float alternately, str data represents chord types, float data represents the continuous beat number of the chord, such as [ 'C',4.0, 'F',4.0, 'G',4.0, 'C',4.0], C chord duration 4 beat, F chord duration 4 beat, G chord duration 4 beat and C chord duration 4 beat.

The rhythm type defines the number of notes and the duration of each note, i.e., the user can customize the length of each note to be generated, with a minimum unit of one beat in 4/4. The device has built in a tempo type list that records popular music tempo types. Of course, the user may also customize a list for defining the tempo, where each number in the list represents the number of beats of each note generated. The start position of each note defaults to the end position of the last note, so the numbers in the rhythm type may be negative numbers, representing the number of beats continuing forward from the end of the last note.

The tone is the tone in which the melody to be generated is located, and includes the basic tone of big and small, and the tone can be set by the user, or can be automatically generated according to the input chord information sequence.

And 2, initializing the initial probability of the pitches of 12 tones in one octave according to the received tone.

The output of the MIDI sequence generating device is a note sequence of melody, which comprises the pitch and duration information of each note, and finally obtains a MIDI file with piano timbre mounted, wherein the MIDI file comprises a columnar chord proceeding track and a generated melody track which are two tracks in total.

In the melody pitch generation process, a probability model is mainly used, and a probability is set for each note to be generated. The probability is composed of a base probability and an additional probability. The base probability is derived from the user-set key and the probability distribution of each note at the given key is different. The additional probability comes from the high-level characteristics of the motion characteristic, the repeat sound, the chord sound, the range of the sound and the like of the melody, and the additional probability is directly corrected on the basis probability of the corresponding note. And obtaining the probability of each final note to be generated according to the calculation results of the two probabilities.

In the present invention, the minimum unit generated by a MIDI sequence is a note, and its basic attributes are pitch and duration. Therefore, the probability of each pitch of the next note is judged according to the given tone, the size tone corresponds to the probability of different pitches, and the probability is used as the basic probability of the next note.

Fig. 2 and 3 are initial pitch probability distribution diagrams for different progression notes at major key and minor key, respectively. The initial pitch probability is used as a base probability, for example, the base probability corresponding to the 1 st note of major scale is 0.184, and the base probability corresponding to the 1 st note of minor scale is 0.192.

And 3, performing chord correction, melody motion correction, repeat tone correction and tone threshold correction on the tone pitch initial probabilities of the 12 tones according to the chord constraint, the melody motion constraint, the repeat tone constraint and the tone threshold constraint to obtain the tone pitch probability distribution of the 12 tones.

On the basis of the initial pitch probability, chord correction, melody motion correction, repeat sound correction and tone threshold correction are carried out on the initial pitch probability, the correction sequence is not limited, and the four kinds of pitch correction are carried out on one initial pitch probability.

After the key is given, the chord of the section of melody is given in beat unit. On the basis of the basic probability of each tone to be predicted, the probability based on chord tones is increased according to the corresponding chord. Specifically, the chord modification in accordance with the chord constraint includes:

setting the chord correction coefficient to 3;

and multiplying the tone pitch probability of the chord component tone corresponding to the chord information by the chord modification coefficient according to the received chord information to obtain a chord modification result.

The melody exercise usually has the way of proceeding in a horizontal direction, a model advance and the like. The embodiment gives the tone to be predicted an additional probability addition according to the previous tone pitch of the tone to be predicted. Specifically, the melody motion modification according to the melody motion constraint comprises the following steps:

when the predicted sound and the generated sound have different musical interval differences, multiplying the pitch probability of the sound to be predicted by a melody motion constraint coefficient, wherein the melody motion constraint coefficient is obtained by the following formula:

F(x，root)＝f(x，root+2，0.5)*0.2+f(x，root+4，1)*0.05+f(x，root-2.3，1)*0.1

+f(x，root-5，0.8)*0.06+f(x，root+7，0.8)*0.03+f(x，root+1，0.7)

*0.05+f(x，root-8，3)*0.02

where F (,) represents the melody motion constraint coefficient, x represents the predicted pitch, and root is the generated pitch.

If the same sound appears in a melody, it is very unpleasant. Therefore, if a certain pitch appears too much in the generated melody, the probability of the tone to be predicted appearing at the pitch is reduced. Specifically, the repeating tone correction according to the repeating tone constraint includes:

setting a repeating sound punishment coefficient to be 0.8;

and comparing the predicted tone with all generated tones, and if the predicted tone pitch is contained in the set consisting of all the generated tone pitches, multiplying the tone pitch probability of the predicted tone by the repeat penalty coefficient to obtain a repeat correction result.

A melody is usually located in a specific range, and a particularly wide range of ranges makes the melody have no center of gravity in hearing and very difficult to sing. Thus specifying the pitch range of occurrence of the tone to be predicted and adding a penalty to it for the probability of deviating from the centre of the register. Specifically, the threshold modification according to the threshold constraint includes:

setting a threshold range, a threshold central value and a threshold standard deviation value, and generating a normal distribution with the threshold central value as a mean value and the threshold standard deviation value as a variance according to the threshold range, the threshold central value and the threshold standard deviation value;

and multiplying the pitch probability of the notes with the pitch within the range of the threshold by the probability value of the normal distribution of each note, and normalizing to obtain the output of the range judgment.

And 4, selecting a note from the pitch probability distribution in sequence according to the sequence of the pitch from high to low for condition screening, and adding the note meeting the screening condition to the generated melody.

When obtaining the pitch probability distribution and obtaining the maximum pitch probability, comparing the predicted note with the generated note pitch, and performing post-processing, wherein the specific processing is as follows:

(a) clear tone

If the predicted note is not in the key, it is marked as a wrong note, and if there is an even key in the melody, it will increase the interest, but if there is a large number of notes, it will affect the stability of the melody, so the note is set as a high probability to be cleared and regenerated.

(b) Clearing a mute interval

If the predicted note and chord form the attenuation interval, the attenuation interval is directly eliminated.

(c) Melody amplitude control

If the predicted note exceeds the octave jump with the generated note or forms a continuous equidirectional jump with the previous two notes, the predicted note is directly cleared.

(d) Chord tone control

If the predicted note is that the second time the outer tone appears in a certain measure, the note is directly cleared.

And 5, synthesizing the generated melody and rhythm type to generate the MIDI file.

The computer system includes one or more non-transitory computer-readable storage devices storing instructions that, when executed by a processor, the computer system, perform the various computing operations described above. The computer may be a desktop computer, a laptop computer, a workstation, a cloud server, a personal digital assistant, or any other computer system. Computer systems include processors, Read Only Memory (ROM), Random Access Memory (RAM), input/output adapters for connecting peripheral devices (e.g., input devices, output devices, storage devices, etc.), user interface adapters for connecting input devices (e.g., keyboard, mouse, touch screen, voice input), and/or other devices, communication adapters for connecting computers to networks, display adapters for connecting computers to displays, and the like.

The MIDI sequence generating device based on the music theory and the statistical rule can generate diversified MIDI sequences under the condition of ensuring controllable melody generation according to the given melody, rhythm and tone.

The above-mentioned embodiments are intended to illustrate the technical solutions and advantages of the present invention, and it should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention, and any modifications, additions, equivalents, etc. made within the scope of the principles of the present invention should be included in the scope of the present invention.

Claims (8)

1. A MIDI sequence generation apparatus based on music theory and statistical rules, comprising a computer system, wherein the computer system is configured to:

receiving a chord information sequence, a rhythm type and a tone;

initializing pitch initial probabilities of 12 tones within one octave according to the received tone;

chord correction, melody motion correction, repeat sound correction and sound threshold correction are carried out on the tone pitch initial probabilities of the 12 tones according to chord restriction, melody motion restriction, repeat sound restriction and sound threshold restriction, and the tone pitch probability distribution of the 12 tones is obtained;

selecting a note from the pitch probability distribution as a predicted note in sequence from high pitch to low pitch, and carrying out the following condition screening: if the predicted note is not in the key, marking the note as wrong note and clearing; if the predicted note and the chord form a pitch reduction interval, directly removing the note; if the predicted note and the generated note exceed eight-degree big jumps or form continuous same-direction big jumps with the previous two notes, the note is directly eliminated; if the predicted note is that the second time of the outer tone appears in a certain measure, the note is directly cleared; adding notes satisfying the screening condition to the generated melody;

the generated melody and rhythm are integrated to generate a MIDI file.

2. A MIDI sequence generation apparatus as claimed in claim 1 wherein the key is automatically generated based on the inputted chord information sequence.

3. A MIDI sequence generation apparatus as claimed in claim 1 wherein said MIDI sequence generation apparatus has a pop music tempo directly integrated with any one of the pop music tempos to generate a MIDI file.

4. A MIDI sequence generation apparatus as claimed in claim 1, wherein when the key is major, the initial probability of pitches of 12 tones is:

the 12 tones are: 1. #1/b2, 2, #2/b3, 3, 4, #4/b5, 5, #5/b6, 6, # 6/b7, 7, the corresponding pitch initial probabilities are: 0.184, 0.001, 0.155, 0.003, 0.191, 0.109, 0.005, 0.214, 0.001, 0.078, 0.004, 0.055;

when the key is a minor key, the initial probability of the pitches of the 12 tones is:

the 12 tones are: 1. #1/b2, 2, #2/b3, 3, 4, #4/b5, 5, #5/b6, 6, # 6/b7 and 7, wherein the corresponding pitch initial probabilities are respectively as follows: 0.192, 0.005, 0.149, 0.179, 0.002, 0.144, 0.002, 0.201, 0.038, 0.012, 0.053, 0.022.

5. A MIDI sequence generation apparatus as claimed in claim 1 wherein the chord modification in accordance with the chord constraints comprises:

setting a chord correction coefficient, wherein the chord correction coefficient range is 2.3-3.5;

and multiplying the tone pitch probability of the chord component tone corresponding to the chord information by the chord modification coefficient according to the received chord information to obtain a chord modification result.

6. The apparatus of claim 1, wherein the melodic motion modification based on the melodic motion constraint comprises:

when the predicted sound and the generated sound have different musical interval differences, multiplying the pitch probability of the sound to be predicted by a melody motion constraint coefficient, wherein the melody motion constraint coefficient is obtained by the following formula:

F(x，root)＝f(x，root+2，0.5)*0.2+f(x，root+4，1)*0.05+f(x，root-2.3，1)*0.1+f(x，root-5，0.8)*0.06+f(x，root+7，0.8)*0.03+f(x，root+1，0.7)*0.05+f(x，root-8，3)*0.02

where F (,) represents the melody motion constraint coefficient, x represents the predicted pitch, and root is the generated pitch.

7. The apparatus for generating a MIDI sequence based on music and statistical rules of claim 1 wherein the repetitive sound modification according to the repetitive sound constraint comprises:

setting a repeating sound punishment coefficient, wherein the repeating sound punishment coefficient range is 0.7-0.9;

and comparing the predicted tone with all generated tones, and if the predicted tone pitch is contained in the set consisting of all the generated tone pitches, multiplying the tone pitch probability of the predicted tone by the repeat penalty coefficient to obtain a repeat correction result.

8. A MIDI sequence generation apparatus as claimed in claim 1 wherein the threshold modification in accordance with the threshold constraint comprises:

setting a threshold range, a threshold central value and a threshold standard deviation value, and generating a normal distribution with the threshold central value as a mean value and the threshold standard deviation value as a variance according to the threshold range, the threshold central value and the threshold standard deviation value;

and multiplying the pitch probability of the notes with the pitch within the range of the threshold by the probability value of the normal distribution of each note, and normalizing to obtain the output of the range judgment.

Images