KR20030034656A - Method for constructing a music database and retrieving a music information using the statistical properties of musical speech - Google Patents

Abstract

PURPOSE: A method for constructing a music database and searching music information using the statistics property of music is provided to construct the music database by classifying the music into the white, pink and brown music according to the statistics property. CONSTITUTION: After sampling a song by receiving the music as the song from a user, a pitch of the sampled song is measured(604). The sampled song is vectorized by using the measured pitch and a vector row is generated(606). The inputted song is classified into one of the white, pink and brown music according to the statistics property by using the generated vector row(608). After searching the most similar vector row to the vector row of the inputted song from the database of the classified music style, a title of the inputted song is provided to the user by using an index of the searched vector row(616).

Description

Method for constructing a music database and retrieving a music information using the statistical properties of musical speech}

The present invention relates to a music database construction method and a music information retrieval method. In particular, according to the statistical properties of music, music is classified into white, pink, and brown music to construct a music database. Music database construction method for efficiently reducing the number of music files, and the search target song inputted by the user as a song is classified into one of the music types of white / pink / brown music and then the classified music types. The present invention relates to a music information retrieval method that enables quick retrieval by only retrieving a database, and a computer-readable recording medium having recorded thereon a program for realizing the above methods.

In the United States, since the early 1990s, Rutgers University has begun researching Distance Metrics as a basic research project to implement digital music libraries for MIDI files. In the UK, Southampton University studied content-based retrieval using pitch contours. In New Zealand, a digital library called MELody inDEX has been implemented.

In Asia, research is being conducted on the implementation of digital music libraries and the use of voice to effectively search for them. At the Tsing Hua University in Taiwan and Hong Kong University in Hong Kong, research using rhythm, indexing technology of music databases, selection of melody lines, A study on extracting representative patterns from a music database (DB) is underway.

In Korea, as a study on computerization of music information processing, we presented melody input method and melody output method to solve I / O problem of melody, and designed music database to efficiently manage a large amount of music and music information. We designed and implemented a music information processing system to retrieve, insert, change, and delete the necessary information from the music database according to the user's request. However, this is a computerized approach, which assigns a special code to each of the numerous music symbols. Therefore, it takes a lot of time and manpower to construct and operate a database (DB), there was a problem that requires special equipment.

In addition, since the conventional music information retrieval system is constructed by focusing on the reduction of the search time from a computational point of view, there is a problem in that the signal processing concept is weak and the accuracy of the search is reduced and the search time is very long.

The present invention has been made to solve the above problems, and builds a music database by classifying music into white, pink, and brown music according to the statistical properties of the music. It is an object of the present invention to provide a computer-readable recording medium having recorded thereon a method and a program for realizing the method.

In addition, the present invention, by first classifying the search target song input as a song from the user to any one of the music type of white / pink / brown music, and then quickly search only the database for the classified music type It is an object of the present invention to provide a music information retrieval method for retrieving music information and a computer readable recording medium storing a program for realizing the above methods.

1 is a block diagram of an embodiment of a music information retrieval system according to the present invention;

2 is a diagram illustrating an embodiment of a search unit of FIG. 1 according to the present invention;

3A to 3C are diagrams illustrating one embodiment of a statistical distribution in a frequency domain showing characteristics of white / pink / brown music applied to the present invention.

Figures 4a to 4c is an illustration of the pattern of the representative vector of the white (pink) / pink (brown) music applied to the present invention.

5 is a flowchart illustrating a method for constructing a database using statistical properties of music according to the present invention.

6 is a flowchart illustrating a music search method using statistical properties of music according to the present invention.

* Explanation of symbols for main parts of the drawings

100: sampling unit 102: pitch measurement unit

104: vectorization unit 106: search unit

108: vector sequence database 110: music title database

112: music file information database 200: music type classifier

202: Vector Heat Finder

In accordance with another aspect of the present invention, a music information retrieval method applied to a music information retrieval system includes receiving a song to be searched by a user as a song, sampling the input song, and then Measuring a pitch of the sampled song; Generating a vector sequence by vectorizing the sampled song using the measured pitch; Using the vector sequence generated in the second step, the input song is one of white music, pink music, and brown music classified according to the statistical properties of music. Classifying the third step; And searching for a vector sequence most similar to the vector sequence of the input song generated in the second stage using a database of music types classified in the third stage in the music database as a search range, and then using the index of the searched vector sequence. And a fourth step of searching for a title of a song input as the song and providing the title to the user.

The method may further include a fifth step of retrieving music file information of a song corresponding to the title retrieved in the fourth step and providing the same to the user according to the user's request.

On the other hand, the present invention, in the music database construction method, according to the statistical properties of the music (white), pink (pink), and brown (brown) music, the first step to build a database by large classification; And a second step of sampling a music to be stored, measuring a pitch of the sampled music, and vectorizing the sampled music using the measured pitch to generate a vector sequence. And classifying the music type into one of white music, pink music, and brown music according to the characteristics of the vector string generated in the second step, and then storing the vector string of the music to be stored in the database for the classified music types. A third step of storing.

Meanwhile, in order to search for music information, the present invention receives a song to be searched from a user as a song in a music information retrieval system having a processor, samples the input song, and then samples the received song. A first function of measuring a pitch of the song; A second function of vectorizing the sampled song using the measured pitch to generate a vector sequence; Using the vector sequence generated in the second function, the input song is one of white music, pink music, and brown music classified according to the statistical properties of music. A third function classified as; And searching a vector string most similar to a vector string of an input song generated in the second function using a database of music types classified in the third function as a search range in a music database, and then using the index of the searched vector string. A computer-readable recording medium having recorded thereon a program for realizing a fourth function provided to the user by searching for a title of a song input as the song is provided.

The present invention also provides a program for further realizing a fifth function of retrieving music file information of a song corresponding to a title searched by the fourth function and providing the same to the user according to a request of the user. Provide a computer-readable recording medium for recording.

On the other hand, the present invention, in order to build a music database, the music information system having a processor, according to the statistical characteristics of the music (white), pink (pink), and brown (brown) music is largely classified into a database A first function of building a; And a second function of sampling a music to be stored, measuring a pitch of the sampled music, and vectorizing the sampled music using the measured pitch to generate a vector sequence. And classifying the music string into one of white music, pink music, and brown music according to the characteristics of the vector string generated by the second function, and then storing the vector string of the music to be stored in the database for the classified music types. A computer readable recording medium having recorded thereon a program for realizing a third function of storing is provided.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a configuration diagram of an embodiment of a music information retrieval system according to the present invention.

The music information retrieval system is composed of a sampling unit 100, a pitch measuring unit 102, a vectorizing unit 104, a retrieving unit 106, and a music database 108, 110, 112. Here, the music database (DB: Database) is a vector sequence database 108 for storing a vector sequence of music, a music title database 110 for storing titles of music, and music file information (music file, URL for music file). Music file database 112, which is stored in the database, and the vector sequence database 108 is logically divided into a white database, a pink database, and a brown database. . It is also possible to physically divide the white database, the pink database, and the brown database.

When the sampling unit 100 receives and samples a part / all of a song to be searched from a user as a song (including a humming sound), the pitch tracker 102 may determine the pitch of the sampled song. pitch).

That is, the pitch tracker 102 creates a pitch curve for the entire melody of the input song, and then forms a note vector string record of the music title database 110, and the length of the note vector string constituting the record. Is determined based on the number of notes. An index is assigned to the note vector string of each record, and the header of one record contains access point information of the music database (DB) 112 corresponding to the title of the song. Here, the pitch tracker 102 may be configured by various methods such as autocorrelation, AMDF, and three level center clipping methods.

The vectorizer 104 generates a vector sequence by vectorizing a song sampled by the sampling unit 100 using the pitch measured by the pitch measuring unit 102.

Then, the search unit 106 analyzes the vector sequence generated by the vectorization unit, and music of any one of white music, pink music, and brown music classified according to the statistical properties of the music. After classifying by type, the vector sequence database 108 restricts the virtual database corresponding to the classified music type to the search range and represents the representative vector pattern most similar to the vector pattern of the input song vector sequence generated by the vectorization unit 104. Search for.

Then, the search unit 106 searches among the vector strings belonging to the similar representative vector pattern having the vector string most similar to the vector string of the song input by the user, and divides the music title by using the index for the retrieved vector string. Obtain an access point of the database 110.

The search unit 106 searches for the title of the song input by the user as the song in the music title database 110 using the access point, and provides the title of the song to the user. If requested, the music file information database In 112, music file information corresponding to the title is searched for and provided.

2 is a block diagram of an embodiment of a search unit of FIG. 1 according to the present invention, and the search unit 106 includes a music type classifier 200 and a vector string searcher 202.

Xn represents a note vector string generated by vectorizing the input song, and the output is an index (or codeword index) of the note vector string to be used as an access point. Becomes

When Xn is input, the music type classifier 200 determines the statistical distribution of the note vector sequence Xn of the input song ( , Where f is the frequency of the power spectrum), which determines whether to use a white database, a brown database, or a pink database.

That is, the music type classifier 200 calculates a vector distribution of the vector sequence of the input song, and determines whether the input song corresponds to white, brown, or pink music. To judge. Here, the criterion is an inclination observed when the magnitude spectrum is shown on a log scale after calculating the power spectrum of the input vector song sequence (see FIGS. 3A to 3C).

Then, the vector string searcher 202 searches for a note vector string similar to the note vector string of the input song in any one of a white database, a brown database, or a pink database. At this time, the relationship between the required search time and the storage space of the vector string database (the storage space of the codebook) is shown in Equation 1 below.

Where k, N and r represent the dimension of the note vector string of the input song, the size (codebook size) and the bit rate of the vector string database, respectively.

3A to 3C are diagrams illustrating an exemplary embodiment of a statistical distribution in a frequency domain showing characteristics of white / pink / brown music applied to the present invention.

FIG. 3A is a diagram illustrating a statistical distribution of white music, showing that the white music has a nearly horizontal slope.

3B is a diagram illustrating a statistical distribution of pink music, indicating that the pink music is a straight line having a slope of −10 dB / dec (1 / f).

FIG. 3C illustrates a statistical distribution of brown music, which indicates that the brown music is a straight line having a slope of −20 dB / dec (1 / f ² ).

By using the statistical distribution characteristics as described above, it is possible to classify which type of music the input song is white music, pink music, and brown music.

4A to 4C are exemplary views of patterns of representative vectors of white music, pink music, and brown music applied to the present invention.

FIG. 4A shows an exemplary diagram of a representative vector pattern for white music, FIG. 4B shows an exemplary diagram of a representative vector pattern for pink music, and FIG. 4C shows a representative vector for brown music. An illustration of the pattern is shown.

Each point in the figure represents a representative vector.

5 is a flowchart illustrating a method for constructing a database using statistical properties of music according to the present invention.

The music information system is largely classified into white music, pink music, and brown music according to the statistical properties of music to build a database. Here, the database is divided into a white database, a pink database, and a brown database, which are physically or logically independent databases.

The music information system receives the music to be stored (502), samples the music (504), and measures the pitch of the sampled music (506).

By vectorizing the sampled music using the measured pitch, a vector sequence of input music is generated (508), and classified into one of white music / pink music / brown music according to the characteristics of the generated vector sequence. (510).

Subsequently, a pattern of the representative vector is extracted by applying a vector quantization algorithm such as a Linde Buzo Gray (LBG) algorithm to the generated vector sequence (512).

The vector of the music to be stored based on the pattern of the representative vector extracted from "512" in the database for the music types classified in "510" (for example, the white database if the music type is classified as white music). Store the heat (514)

6 is a flowchart illustrating a music search method using statistical properties of music according to the present invention.

When the sampling unit 100 receives a part / all of the song to be searched from the user as a song (including a humming sound) (600), and when sampling (602), the pitch tracker (102) The pitch of the sampled song is measured (604).

Thereafter, the vectorizer 104 generates a vector sequence by vectorizing the sampled input song using the pitch measured by the pitch measurer 102 (step 606).

Then, the searching unit 106 analyzes the vector sequence generated by the vectorizing unit 104, and among the white music, pink music, and brown music classified according to the statistical properties of the music. The music is classified into one music type (608).

In addition, the search unit 106 limits the virtual database corresponding to the classified music types in the vector sequence database 108 to the search range (610). That is, the search range is limited to any one of the white / pink / brown database according to the classified music type (610).

Also, the search unit 106 searches for the representative vector pattern most similar to the vector pattern of the vector sequence for the input song generated through vectorization within the database range limited to "610" (612). In other words, the most similar representative vector pattern is first found among the representative vector patterns.

Thereafter, the search unit 106 searches for a vector string most similar to the vector string of the song input by the user among the vector strings belonging to the most representative representative vector pattern, and obtains an index for the retrieved vector string (614). Here, the header of the subvector column is assigned an access point value of the music title database to which the note vector string belongs (multiple subvectors can have the same access point value), so that the text of the music from the music title database can be set. (text) Information (e.g., music title, etc.) is obtained.

Subsequently, the search unit obtains an access point of the music title database 110 divided by using the acquired index, and then the song input by the user as a song in the music title database 110 using the access point. The title of the song is provided to the user by searching the title of the music (616). That is, after the text information of the music is obtained, the music search is performed using the obtained text information to provide the result to the user. .

Meanwhile, if requested by the user, the music file information database 112 retrieves and provides music file information corresponding to the title (618).

The present invention described above is capable of various substitutions, modifications, and changes without departing from the spirit of the present invention for those skilled in the art to which the present invention pertains, and the above-described embodiments and accompanying It is not limited by the drawings.

As described above, the present invention classifies the types of music using the statistical distribution of the inputted vector strings, and then searches the database corresponding to the classified music types by limiting the search range. It is effective to make searching efficiently.

In addition, the present invention, it is possible to operate a large database, to create a music database for each genre has the effect of increasing the search speed and accuracy.

In addition, the present invention has the effect of maintaining the usefulness of the existing text-based music search method.

Claims (8)

In the music information search method applied to the music information search system, A first step of receiving a song to be searched from a user as a song, sampling the input song, and measuring a pitch of the sampled song; Generating a vector sequence by vectorizing the sampled song using the measured pitch; Using the vector sequence generated in the second step, the input song is one of white music, pink music, and brown music classified according to the statistical properties of music. Classifying the third step; And Searching for a vector sequence most similar to the vector sequence of the input song generated in the second stage using a database of music types classified in the third stage as a search range, and using the index of the retrieved vector sequence. A fourth step of searching for the title of the song input as the song and providing the title to the user Music information retrieval method using the statistical properties of the music, including. The method of claim 1, A fifth step of retrieving music file information of a song corresponding to the title retrieved in the fourth step and providing the same to the user according to the user's request; Music information search method using the statistical properties of the music further comprising. The method according to claim 1 or 2, The fourth step, A sixth step of searching for a representative vector pattern most similar to the vector pattern of the vector sequence generated in the second step by limiting a database of the music types classified in the third step of the music database to a search range; And The vector sequence most similar to the vector sequence generated in the second step is searched among the vector sequences belonging to the representative vector pattern searched in the sixth step, and the title of the song is searched by using the index of the searched vector sequence. 7th step to give to Music information retrieval method using the statistical properties of the music, including. In the music database construction method, A first step of constructing a database by largely classifying into white music, pink music, and brown music according to the statistical properties of the music; And A second step of sampling a music to be stored, measuring a pitch of the sampled music, and vectorizing the sampled music using the measured pitch to generate a vector sequence; And After classifying the music type into one of white music, pink music, and brown music according to the characteristics of the vector string generated in the second step, the vector string of the music to be stored is stored in a database of the classified music types. The third step Music database construction method using the statistical properties of the music, including. The method of claim 4, wherein The third step, A fourth step of classifying the music type into any one of white music, pink music, and brown music according to the characteristics of the vector sequence generated in the second step; A fifth step of extracting a pattern of the representative vector by applying a vector quantization algorithm to the generated vector sequence; And A sixth step of storing the vector sequence of the music to be stored based on the extracted representative vector pattern in a database of the music types classified in the fourth step; Music database construction method using the statistical properties of the music, including. In order to retrieve music information, a music information retrieval system having a processor, A first function of receiving a song to be searched from a user as a song, sampling the input song, and measuring a pitch of the sampled song; A second function of vectorizing the sampled song using the measured pitch to generate a vector sequence; Using the vector sequence generated in the second function, the input song is one of white music, pink music, and brown music classified according to the statistical properties of music. A third function classified as; And Searching for a vector string most similar to a vector string of an input song generated in the second function, using a database of music types classified in the third function as a search range, and then using the index of the searched vector string. A fourth function of searching for the title of a song input as a song and providing the same to the user A computer-readable recording medium having recorded thereon a program for realizing this. The method of claim 6, A fifth function of retrieving music file information of a song corresponding to a title searched by the fourth function and providing the same to the user according to a request of the user; A computer-readable recording medium that records a program for further realization. In order to build a music database, in a music information system having a processor, A first function of constructing a database by roughly classifying into white music, pink music, and brown music according to the statistical properties of the music; And A second function of sampling a music to be stored, measuring a pitch of the sampled music, and then vectorizing the sampled music using the measured pitch to generate a vector sequence; And After classifying the music type into one of white music, pink music, and brown music according to the characteristics of the vector string generated by the second function, the vector string of the music to be stored is stored in a database of the classified music types. Third function A computer-readable recording medium having recorded thereon a program for realizing this.