KR100484902B1 - Method for accumulation and search of contour by arithmatization - Google Patents

Abstract

The present invention relates to a method for storing and retrieving songs by quantifying music melody, and its purpose is to compare not only the pitch but also the height of the pitch as a result of comparing the level of the pitch in more detail and to database the human body. The number of pitches is digitized according to the pitch change, and then a matching song is searched in the database. Therefore, only the humming is input to extract the song more accurately.

In the music storing process according to the object of the present invention, the step of dividing the height direction and size of each pitch in the music melody by the predetermined grade, divided by the predetermined grade, the total melody displayed by the predetermined grade overlap each other to a certain length The step of dividing into a plurality of different sub-line rate, the step of digitizing each divided sub-line rate to a value within a certain range and storing the database, and the song search process extracts the high and low of the input sound Converting into a grade comparing the adjacent pitch to a set region, generating a predetermined length of subline string converted to grade, digitizing the subline and performing alignment, and a quantized song Searching for, and displaying a list of the found similar songs.

Description

Music storage and retrieval method by melody quantification {Method for accumulation and search of contour by arithmatization}

The present invention relates to a system for retrieving a musical melody and a search for the same. More specifically, the musical pitch is digitized by various grades and stored as a database, and the user's humming is inputted to define the pitch. The present invention relates to a method of storing and retrieving a song by digitizing a musical melody that converts a height into a certain grade and then searches a previously stored database to inform the user of a corresponding song when a matching song is detected.

In general, database searches have been done in a textual environment. Recently, multimedia data according to the development of various contents are similarly made by searching a text environment.

When audio data is databased or searched, it is stored as an audio data file, but when it is databased or searched, search keys are all made in a text environment such as title, singer, composer, and song theme.

On the other hand, while encountering a lot of songs, you need to know the title, singer, etc. of all songs, you can find the song. For example, in order to find a song to sing in a karaoke room, the user needs to check a coded list of information such as a song title, a singer, and the first verse of the lyrics.

If the user did not know the above information and knew only one or two melodies, it was almost impossible to find the song in the list.

Therefore, in order to solve such a problem, conventionally, a single note in a melody is classified into three types according to the change direction thereof to search for a song.

That is, if the note is higher than the adjacent previous note, 'u'(up); if the note is lower than the adjacent note, 'd'(down); In the case of 'r' (repetition) respectively. Also, each pitch in the tune is represented by one element of the alphabet 'Σ ₃ = {u, d, r}', so that one tune is composed of the string 'S' Σ 'consisting of the elements of' Σ ₃ '. _It can be expressed as ₃ ^* '. For example, as the sheet music showing the first four words of agitation 'school bell' as shown in FIG. 1, when the elements of 'Σ ₃ ' are used, as shown in FIG. It can be expressed as = rurdrdurdrd '. In this way, the entire 'school bell' song can be expressed by the stitching of the elements of 'Σ ₃ '. Conversely, most songs can only be identified by the entire string of songs.

In order to identify a song with a query string consisting of elements of Σ _3, the length of the query string must be 15 or more.

However, such a prior art, in consideration of the database size in reality, the length of the query string to prevent the search collision (collision) should be longer than this. There is a problem that it is not practical for the user to create a query string of this length for the actual song to be searched.

In addition, since a song is composed of hundreds of notes, the length of the string representing the tune of the song is also as long. Therefore, it is a considerable time consuming problem to sequentially search individual songs in the entire database using a general string matching technique to find a song including a melody string introduced into a query.

The present invention is to solve the above-mentioned problems of the prior art, the object of which is not only high pitch as compared to the adjacent pitches, but also to separate the degree of the height in more detail into a multi-step database, and humming people It provides a method of storing and retrieving songs by quantifying the music melody, which digitizes a number of pitch grades according to the change of pitch, and searches for matching songs in the database, so that only the humming is input to extract the song more accurately and quickly. have. By quantifying the storage in the form of a tree or a hash table, the conventional string or vector matching method is required to significantly improve the disadvantage that the search time takes in proportion to the amount of music stored in the database.

 According to the present invention, in the method of databaseting each pitch in the tune of a song, the music storing process by the quantification of the music tune according to the present invention is performed. A first step of dividing a direction and a magnitude into predetermined grades, a second step of dividing the entire melody divided by the predetermined grade into a plurality of different sublines to overlap each other with a predetermined length, and each of the divided sublines And a third step of digitizing the database to a value within a predetermined range and storing the database.

In the song search process by quantifying the music melody to achieve another object of the present invention, the song search method by the melody that detects the height of the user's voice to search for a matching song in the database, the height of the input sound After extracting the first step of converting the level of high and low to a grade compared to the adjacent pitch in the set region, the second step of generating a predetermined length of the sub-line string converted to the grade, and the sub-line And a fourth step of performing a sort, a fourth step of searching for a song including the digitized sub-line, and a fifth step of displaying a list of the found similar songs.

The present invention made as described above will be described in detail with reference to the accompanying drawings.

FIG. 3 is a block diagram of a music search apparatus using numerical melody for embodying the present invention. The pitch extracting unit 102 extracts the height of each sound by inputting a user's song input through the input unit 101. And a search engine 110 for converting the pitch data extracted from the pitch extracting unit 102 into various sub-rates and sorting each sub-rate to search for a matching song, and the search engine unit ( According to the control of the search engine 110 and the database 122 for extracting the approximate songs that are matched from the pre-stored songs through the index unit 121 according to the search request of the 110; The display unit 131 sorts and displays a list.

In this case, the search engine unit 110 includes a sub-line rate generator 111, a sub-line rate sorter 112, and a searcher 113.

When described in more detail with reference to the accompanying drawings, the operation according to the embodiment of the present invention configured as described above are as follows.

First of all, the present invention divides the result of comparing the previous syllables into multiple stages for all the pitches of the song, and then digitizes the separated grades and stores them in the database. When the user inputs a humming in the future, the stored database extracts the height of each pitch thereof, digitizes it within the set area, searches for a matching song, and displays the found song.

3 is a block diagram of a music retrieval apparatus using melody digitization for implementing the present invention. When a user hums a song, the input unit 101 detects it and converts it into a digital signal.

Here, the input unit can be both an analog signal including human humming and playing, and a digital signal such as an output of a digital keyboard instrument or other digital audio data.

The pitch extraction unit 102 extracts the high and low of the converted digital voice signal. That is, it plays a role of converting the digitized melody through the input unit into a query that can be recognized by the search engine.

The search engine 110 searches for the database 122 using the query input from the pitch extraction unit 102 and displays the search results through the display unit 131.

That is, the search engine 110 performs insertion, deletion and substitution operations, which are interval based operations, using the pitch of the pitch extracted from the pitch extractor 102. To produce each set of substring strings.

The sub-line rate aligning unit 112 generates the adjacent sub-line rate numerical values with respect to a calculation or combination of adjacent sub-lines from the sub-line string.

The search unit 113 searches the database 122 through the index unit 121 with the adjacent sub-line rate values as primary keys.

4 is a flowchart of a music storing process by quantifying a music tune according to an embodiment of the present invention, in which each pitch of the tune is higher than the adjacent previous pitch in order to store the original tune in the database 122; Low (d) and identical (r) are identified and expressed as a string of elements of each alphabet (S101).

The given melody is as low as Σ ₁₆ = {d ₇ , d ₆ , ‥‥, d ₁ , r ₀ , u ₁ , u ₂ , ‥‥, u ₈ } ' u) and the degree of change is represented by a string of numbers. That is, as a result of comparing the pitch with the adjacent previous pitch, 'd ₇ , d ₆ , ‥‥, d ₁ ' if the previous note is low, 'r ₀ ' if the same note is high, and 'u if the previous note is high ₁ , u ₂ , ..., u ₈ '.

Here, if the note is higher than the previous note (u), all the pitches with a difference greater than or equal to 8 are expressed as 'u ₈ ', and if the note is lower than the previous note (d) if the difference is greater than or equal to 7 The pitch is represented by 'd ₇ '. This is because, when the size of the alphabet exceeds a certain level, the efficiency of search and storage space is greatly reduced.

For example, the melody of FIG. 1 may be represented by the melody string 'S' = r ₀ u ₂ r ₀ d ₂ r ₀ d ₃ u ₃ r ₀ d ₃ r ₀ d ₂ 'according to the height and grade. .

As shown in FIG. 6, the melody may be divided into sub-strings to be overlapped by a predetermined length (K-segment) of the sub-line. (S102)

As such, the substring strings respectively perform digitization (S103).

In other words, the melodies 'Σ ₁₆ = {d ₇ , d ₆ , ..., d ₁ , r ₀ , u ₁ , u ₂ , ..., u ₈ }' converted to grades are digitized. Since all are separated by 16 grades, the grades matched by 0, 1, 2, ..., 15 are digitized. More specifically, the numerical value is expressed as 'd ₇ = 0, d ₆ = 7, ..., d ₁ = 6, r ₀ = 7, u ₁ = 8, u ₂ = 9, ..., u ₈ = 15'.

If a set of all substring strings of length 'K'is' Σ ₁₆ ^k 'and a set of assorted non-negative integers is Z ₀ ⁺ , an arbitrary substring string of length' K '' s = s ₁ s ₂ ‥‥ s _k

'(s _i ∈Σ ₁₆ , 1≤i≤k) is encoded into a value within a range by the one-to-one function f = Σ ₁₆ ^k → Z ₀ ⁺ as shown in Equation 1 below. Can be.

On the other hand, if the substring string 's = s ₁ s ₂ ‥‥ s _k is a hexadecimal number consisting of the numbers v (s ₁ ), v (s ₂ ), ‥‥, v (s _k ), f (s) is It is easy to see that the decimal value corresponds to this hexadecimal number.

delete

Further, when the length of the sub-line is k, the range of the value of f (s) is expressed by the following equation (2).

If the length of the sub-line is set to 8, the range of the value of f (s) is expressed by Equation 3.

This means that there are a total of 16 ⁸ = 2 ³² different floats.

This is large enough to keep search collisions low per float. This also means that all 8-length substring strings can be represented with only 4 bytes (32 bits). This allows you to take full advantage of the power of a 32-bit machine.

As such, each substring rate string that has been digitized is converted into a database and stored in the database 122 (S104).

FIG. 5 is a flowchart of a music search process by digitizing a music tune according to an embodiment of the present invention. If a user hummings through an input unit 101 such as a microphone, the pitch extraction unit 102 is converted into a digital voice signal. Output as (S110)

In the pitch extraction unit 102, user input may be distorted in the pitch extraction process, and problems in the database may be generated such as incorrectly inputting a song in a database or a variety of variations exist due to one song being played in various ways. This can interfere with your search. For this reason, the search should be performed approximately while allowing some error.

The pitch of the pitch extracted by the pitch extraction unit 102 generates the sub-line rate in the sub-line rate generation unit 111. (S111)

For the sub-line string 's', the sub-line string within a predetermined distance from the 's' is called an adjacent sub-line string of 's'. Here, the distance between the strings is determined by the type and number of operations required to convert from one string to another string.

In addition, a set of adjacent sub-line strings of the sub-line string 's' is expressed as 'N (s)'.

Given a query subline string 's', to generate an approximate search for the query, first create an adjacent subline string set 'N (s)' of the string from the string 's'. Among the operations used in this generation process, there are three basic operations of string matching theory: insert, delete and replace operations. Since the three basic operations are performed in units of pitch from the viewpoint of Mars, this is called interval based operations.

Such a pitch-based operation is performed in units of a single pitch in the melody irrespective of the overall melody, and thus often generates adjacent sub-lines which are not probable from the viewpoint of Mars.

For example, as shown in FIG. 7A, the fourth sub-line string 'd ₂ r ₀ d ₃ u ₃ r ₀ d ₃ r ₀ d ₂ ' is shown in FIG. 6, and FIG. 7B is the first in FIG. 7A. This is a diagram showing the subline rate with the pitch 'u ₄ ' inserted between the second and second pitches. The subline string after insertion is 'd ₂ u ₄ r ₀ d ₃ u ₃ r ₀ d ₃ r ₀ '. .

If the notes before the insertion operation are referred to as notes 1, 2, ..., 9 from the front, the magnitude of the pitch between the 1st note and the remaining notes is significantly different before and after the insertion operation as shown in FIG. 7A. In addition, each of the above two sub-line is a musically different feeling.

In contrast, note-based operations perform operations in units of notes from the viewpoint of harmony, and there are three operations of insertion, deletion, and substitution, and note-based insertion is a substring string 's. Replace one note 's _i ' with two notes 'xy' that satisfy the following equation (4).

As shown in Figs. 8A and 8B, this operation is equivalent to inserting a new note between the second and third notes. This serves to compensate for the user who missed a note in the query (S113).

The note-based delete operation replaces two pitches s_i s_i + 1 in the substring string 's' with a pitch 'x' as long as the following equation (5) is satisfied.

9A and 9B are pitch display diagrams based on a note-based deletion operation, which is equivalent to deleting a third note, which serves to correct an error in which a user incorrectly inserts a note in a query (S114).

Note-based imputation involves two pitches in the substring string 's' Is replaced by two pitches xy satisfying equation (6).

delete

10A and 10B are diagrams showing a note-based substitution operation, in which a note-based substitution operation is equivalent to replacing a third note with another note. This serves to correct a user's mistake of putting one note into another in the query (S112).

The substitution, insertion, and deletion subline strings generated through the above process are respectively digitized and aligned. (S115) (S116) The numerical process refers to the numerical process for storing the music melody database.

When searching the database, the numerical values of the subline rate are used as the primary key and sorted according to their size. The number of the song including the sub-line and the positional information of the sub-line in the tune are stored along each numerical value of the sub-line.

Sorting of adjacent sublines allows for a more efficient database search. The search method traverses the melody tree and uses the method of searching for adjacent sublines.

If the number of songs found in the search process is large, it is difficult for the user to search. Therefore, it is necessary to help the user search by assigning scores to the songs searched according to a predetermined criterion and arranging songs by scores.

The criteria for assigning scores to the retrieved songs are calculated based on the weight ratio based on the pitch-based operation or the tone-based operation, and the weight ratio based on the insertion, deletion, and substitution operations. Different weights for the number of inserts, deletes, weights according to the negative or pitch subject to the calculation in the substitution operation, weights according to the number of sublines in the tune of the query to generate the corresponding sublines for each adjacent sublines Optimizing the ranking according to how the weight of the floats in the database depends on the extent to which the floats characterize the songs that contain the corresponding float or how the weights of successive floats in the query are retrieved from successive positions of a song (ranking optimization) is determined. (S119)

Through this process, only the humming of the user is detected, the song stored in the database is searched and displayed to the user.

Although the preferred embodiment according to the present invention has been described above, it will be understood by those skilled in the art that various modifications and changes can be made without departing from the scope of the claims of the present invention.

As described above, the song storage and retrieval method by the quantification of the music melody according to the present invention, even if you do not know the title or code number of the song that you want to find in karaoke, etc. After searching for a song, the searched song is displayed to the user so that the song can be easily found. In addition, compared to the conventional matching method, the numerical method enables the use of a tree or hash table data structure, which significantly improves the search time and significantly improves the practicality.

1 is a sheet music showing a typical four-melody melody,

FIG. 2 is a diagram illustrating a change direction of pitch in FIG. 1;

3 is a block diagram of an apparatus for retrieving music by melody digitization for implementing the present invention;

4 is a flowchart illustrating a process of digitizing a musical tune and storing it in a database according to an embodiment of the present invention.

5 is a flowchart of a music retrieval process by digitizing a musical tune according to an embodiment of the present invention.

6 is a diagram showing a string of four sub-rates according to an embodiment of the present invention,

Figure 7a, 7b is a view showing an embodiment of the pitch-based insertion operation according to the present invention,

8A and 8B show an embodiment of a sound-based interpolation operation according to the present invention,

9A and 9B illustrate an example of a sound-based deletion operation according to the present invention.

10A and 10B illustrate an example of a sound-based substitution operation according to the present invention.

<Description of Symbols for Major Parts of Drawings>

101: input unit 102: pitch extraction unit

110: search engine unit 111: floating rate generation unit

112: floating line sorting unit 113: search unit

121: index portion 122: database

131: display unit

Claims (8)

In the method of database each pitch in the song melody, A first step of dividing the pitch and the height direction of each pitch in the melody by a predetermined grade; A second step of dividing the displayed total melody divided by the predetermined grade into a plurality of different sub melodies so as to overlap each other with a predetermined length; And And a third step of digitizing each of the divided sub-line rates to a value within a predetermined range and storing the database. According to claim 1, wherein in the first step the height direction compared to the adjacent pitch of the shear is divided into high (u), the same (r), low (d), characterized in that the size is at least 4 grades or more Music storage method by melody quantification. The music storage method according to claim 1 or 2, wherein the grade of each pitch is digitized in ascending order. In the song search method by the melody that detects the height of the user's voice to search for a matching song in the database, A first step of extracting the height of the input sound and converting the degree of the height into a grade comparing the adjacent pitch to a set region; Generating a substring string having a predetermined length converted to the grade; A third step of performing alignment after digitizing the sub line rate; A fourth step of searching for a piece of music including the digitized subline rate; And And a fifth step of displaying the searched list of similar songs. 5. The tune rate according to claim 4, wherein the input high and low grades separate and convert high (u), same (r), and low (d) to at least 4 grades or more in comparison with the adjacent adjacent pitches. Search method by digitization. 5. The method of claim 4, wherein the generation of the sub-line string in the second step generates a plurality of approximate sub-line strings for correcting the insertion, omission, and substitution of the pitches generated by the user's input error and the deformation of the song. A music retrieval method by digitizing a musical tune. 7. The music according to claim 4 or 6, wherein the sub-line string generation generates adjacent string sets of the sub-line strings by independently / combining a pitch / tone based operation according to a preset operation weight. Song search method by digitization of melody. 5. The method of claim 4, wherein the fifth step optimizes the ranking by assigning weights according to the insertion, omission, and substitution operations of the plurality of songs to be searched and displayed.