KR20080082022A - Likelihood measurement apparatus and method based on music characteristics and music recommendation system and method using its - Google Patents

Abstract

A device and a method for measuring similarity based on feature of music, and a music recommendation system and method using the same are provided to calculate the similarity in an Internet or digital broadcasting environment based on the feature of the preferred music of a user and recommend the music effectively to the user based on the similarity. A music feature extractor(131) extracts feature information, which is at least one of rhythm or tone feature information, reflecting characteristics of music. A music feature information database(132) stores the extracted feature information in a feature vector row. A similarity calculator(133) calculates similarity between music by comparing a music list(16) extracted based on genre/singer information of the current queried music(14) of a user with the feature information of the current queried music or accumulative feature information of the past queried music. An updater updates the feature information of the current queried music to the accumulative feature information of the past queried music. A music recommender generates and provides a recommendation list in high similarity order depending on a similarity calculation result.

Description

Likelihood measurement apparatus and method based on music characteristics and music recommendation system and method using its

1 is a configuration diagram of an apparatus for measuring similarity based on music characteristics according to the present invention and a music recommendation system using the same;

2 is an explanatory diagram illustrating a music characteristic-based similarity measuring method and a music recommendation process using the same according to the present invention;

3 is a flow chart of an embodiment of a music recommendation method according to the present invention;

4 is an exemplary configuration diagram of a text file in which music feature information used in the present invention is stored;

5 is an explanatory diagram showing a list format of user preference query music used in the present invention.

* Explanation of symbols on the main parts of the drawing

11: list extraction unit 12: music DB

13: similarity measuring device 14: quality original music

15: user past query history information 16: music recommendation song list

The present invention relates to an apparatus and method for measuring similarity based on music characteristics, and a music recommendation system and method using the same. More specifically, the present invention relates to a music preference system based on the characteristics (rhythm, tone, etc.) of user preferred music in the Internet or digital broadcasting environment. The present invention relates to a music characteristic-based similarity measuring apparatus and method thereof, and a music recommendation system using the same, and method for calculating music similarity and effectively recommending music based thereon.

Digital music databases are growing and the demand for effective digital music retrieval, automatic music summaries, and music recommendation systems is growing, and in particular, the demand for automated music recommendation systems that can generate commercial value is growing. .

As a general method for recommending music, a user may recommend using a distribution pattern of previously listened music, or simply compare similarities between music.

In the former case, the music download list of the users may be compared with each other to extract a user group in which downloads of the same songs overlap a lot. Based on this, a complete list of downloads / streamings of a similar user group is constructed, and then compared to this list, the song is not recommended to the user list. In this case, the group may be configured based on the user's music download list, but the user's music mood may be simply classified into joy, sadness, and cheerfulness by using the signal processing technology. The distribution can be calculated. Based on this, users' tastes may be grouped to recommend music held by a group having the same taste to other users of the same group.

Such a method of recommending music through user grouping through a user's existing listening song or a list of downloaded songs is, for example, a user's possession of A or a distribution of existing listening songs is maintained by User B or distribution of existing listening songs. Assuming that is similar to, a new listening song of user B is recommended to user A, or simply classified based on the mood of music.

Thus, this approach may reflect some of the user's preferences for music, but ultimately, rather than recommend it as a user's preferred music characteristic (eg, rhythm, timbre, etc.), the user has already listened or downloaded. Since the music selected by the other user is recommended by using the distribution of one music, the accuracy is very low, and there is a problem in that the characteristics of the music preferred by the user (eg, rhythm, tone, etc.) cannot be properly reflected. In addition, the manner in which the user-owned music is simply classified into four or five categories and the users are grouped in the distribution thereof to recommend the music held by each other is highly accurate according to the ambiguity and diversity of the music atmosphere. There is a problem falling.

In the latter case, the similarity between music is calculated by extracting feature vectors of lower levels (for example, physical levels of signals such as energy and peak values) from the music signal. As described above, in the case of recommending similar music by finding similarity between music, there is an advantage that the user's preference can be reflected by recommending music similar to user's query, but the lower level (for example, energy, peak) Since a feature vector of a simple physical level of a signal such as a value is extracted and processed, the accuracy is lowered.

However, since the user's preferred intention is often based on the song characteristics itself, it may be more effective to grasp the music characteristics and to search for and recommend music similar to the characteristic patterns of the user's query music or the user's past queries. Accordingly, music characteristics based on music characteristics, which can extract music characteristics related to rhythm and tone of digital music, etc. into feature vector sequences, and search for and recommend music similar to cumulative characteristic patterns of music of user query or user's past query in music database. Recommendation system is desperately needed.

The present invention has been proposed in order to meet the above demands, and based on the characteristics of music that can calculate the similarity based on the characteristics (rhythm, tone, etc.) of the user's preferred music in the Internet or digital broadcasting environment It is an object of the present invention to provide a similarity measuring apparatus and method, a music recommendation system using the same, and a method thereof.

Other objects and advantages of the present invention can be understood by the following description, and will be more clearly understood by the embodiments of the present invention. Also, it will be readily appreciated that the objects and advantages of the present invention may be realized by the means and combinations thereof indicated in the claims.

According to an aspect of the present invention, there is provided an apparatus for measuring similarity between music, comprising: music feature extraction means for extracting feature information reflecting characteristics of music; Music feature information storage means for storing the extracted feature information as a feature vector sequence; And similarity calculation means for calculating the similarity between the music by comparing the extracted music list based on genre / singer information on the user query music with the feature information of the user's current query music and / or the cumulative feature information of the user's past query music. Characterized in that consisting of.

The present invention may further comprise updating means for updating the feature information of the user's current query music to the cumulative feature information of the user's past query music.

The method may further include music recommendation means for generating and providing a recommendation list by a predetermined number in order of the song having the highest similarity according to the similarity calculation result.

On the other hand, the present invention, a method for measuring the similarity between music, the step of extracting feature information reflecting the characteristics of the music; And comparing the extracted music list with the feature information of the user's current query music and / or the cumulative feature information of the user's past query music based on the genre / singer information on the user query music, and calculating the similarity between the music. Characterized in that made.

The present invention may further include accumulating and storing feature information of the user's current query music in cumulative feature information of the user's past query music.

The present invention may further include generating and providing a recommendation list for a predetermined number of songs in the order of high similarity among the music lists according to the similarity calculation result.

On the other hand, the present invention provides a music recommendation system comprising: music feature extraction means for extracting feature information reflecting a characteristic of music; Music feature information storage means for storing the extracted feature information as a feature vector sequence; List extracting means for extracting a music list based on genre / singer information on user query music; Similarity calculation means for comparing the extracted music list with the feature information of the user's current query music and / or the cumulative feature information of the user's past query music to calculate the similarity between the music; And music recommendation means for generating and providing a recommendation list by a predetermined number in the order of high similarity among the music lists according to the similarity calculation result.

The present invention may further comprise updating means for updating the feature information of the user's current query music to the cumulative feature information of the user's past query music.

On the other hand, the present invention, in the music recommendation method, the step of extracting feature information reflecting the characteristics of the music; Extracting a music list based on genre / singer information about a user query music; Calculating similarity between music by comparing the extracted music list with feature information of user current query music and / or cumulative feature information of user past query music; And generating and providing a recommendation list by a predetermined number in the order of the song having the highest similarity among the music lists according to the result of the similarity calculation.

The present invention may further include accumulating and storing feature information of the user's current query music in cumulative feature information of the user's past query music.

The present invention analyzes the characteristics of the music data from the original query or the user's past query history in order to effectively recommend the music to the user in the Internet or digital broadcasting environment, and calculates the similarity, based on this recommend the music (song recommendation) )

That is, the present invention calculates the similarity based on the upper level music characteristics such as rhythm, timbre, etc., and builds a music recommendation system based on this, and recommends music effectively to the user. In addition, by accumulating and storing user preferences, a recommendation service in consideration of user preference (preferred) is possible.

In this way, by extracting and using the characteristics of the music expressed by the user's preferred intention (that is, recommending the music according to the user's preferred music characteristics), a more effective and personalized personalized music recommendation is possible. This can be an effective tool for personalized music recommendation services in the Internet or digital broadcasting environment, and is expected to play a good role in the coming customized service era.

The above objects, features and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, whereby those skilled in the art may easily implement the technical idea of the present invention. There will be. In addition, in describing the present invention, when it is determined that the detailed description of the known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. 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 apparatus for measuring similarity based on music characteristics according to the present invention and a music recommendation system using the same.

As shown in FIG. 1, the apparatus for measuring a similarity based on music characteristics according to the present invention includes a music feature for extracting feature information (which is one of rhythm or timbre feature information indicating a music characteristic of a higher level) reflecting the music characteristic. A music feature information database 132 that stores the extracted feature information (rhythm, timbre, etc.) as a feature vector sequence, a music list extracted based on genre / singer information about user query music; Compare characteristic information of the user's current query music (music characteristic information extracted from the original query music 14) and / or cumulative characteristic information of the user's past query music (music characteristic information extracted from the user's past query history information 15) And a similarity calculator 133 for calculating the similarity between the music.

Here, 'characteristic information of the user's current query music' refers to a song title of the corresponding music A, or a user directly inputting a music recommendation after the user listens to the music A (click the music recommendation icon). A feature vector representing a rhythm or timbre characteristic of music A having a song title of one music A. FIG.

In addition, the "cumulative feature information of user's past query music" means that the number of user's queries or the number of times of listening / downloading a specific song, etc. of the user's cumulative stored music (A, B, C, D ...) A feature vector representing rhythm or timbre characteristics. Therefore, when the similarity is calculated based on the 'characteristic information of the user's current query music', the 'characteristic information of the user's current query music' is accumulated in the 'cumulative characteristic information of the user's historical query music'. It is accumulated and stored for each artist.

When the music recommendation system is constructed based on the similarity measuring device that calculates the similarity based on the high level music characteristics such as rhythm and tone, the feature information reflecting the characteristics of the music (rhythm indicating the high level music characteristics) Or a music feature extraction unit 131 for extracting one of the tone feature information, a music feature information database 132 storing the extracted feature information (rhythm, tone, etc.) as a feature vector sequence; A list extracting unit 11 for extracting a music list based on the genre / singer information for the user, feature information of the extracted music list and the user's current query music (music characteristic information extracted from the query original song music 14), and / Or compares the cumulative characteristic information of the user's past query music (music characteristic information extracted from the user's past query history information 15), Similarity calculation unit 133 for calculating the similarity between and generates a recommendation list (list of music recommendation songs by genre / singer by genre / singer 16) by a predetermined number of songs in the order of high similarity according to the similarity calculation result. A music recommendation portion (not shown) to provide. In this case, if the similarity is calculated based on the feature information of the user's current query music, the feature information of the user's current query music is accumulated and stored by the music genre / artist in the cumulative feature information of the user's past query music. And an updater (not shown) for (update).

Then, the operation of the music recommendation system according to the present invention will be described in more detail.

In the music recommendation service of the present invention, when a user requests music recommendation (click the music recommendation icon) after listening to music A in a digital broadcasting or Internet environment, the music title, genre, and singer of the music A Metadata composed of artist information or the like is transmitted to the server, or the song title, genre, or artist information of music A directly input by the user as a query is transmitted to the server.

In other words, in a digital broadcasting or Internet environment, a user may express his / her preferred intention (genre / artist input or song title selection / entry) for a certain music, and metadata of the song may be Can be delivered to the server. In this case, the metadata includes information such as music genre, singer, song name, and the like. The metadata of the user's favorite song may be delivered to the server whenever a user's favorite intention occurs. May be passed to the server. In this case, the list format of the user preference query music is shown in FIG. 5 and is composed of a user ID, an artist, a song title, a genre, a user preference, and the like.

However, as described above, it is also possible for a user to directly input a song name or to input information regarding a music genre or a singer as a query condition. In this case, only the query (search condition) will be sent to the server, not the metadata from the client to the server. However, in the case where the user directly inputs the song name, if the metadata of the song exists in the client, the metadata of the song is transmitted to the server. Otherwise, the query (search condition) is transmitted to the server.

Therefore, even when a user directly enters genre or artist information of a music, or selects or directly enters a song title of a music, the server may generate a user's favorite genre or artist. Information is available.

Thereafter, the list extracting unit 11 of the server extracts songs corresponding to the genre or artist from the music DB 12 to generate a music list (a collection of comparison target music). For example, all of the songs corresponding to the 'dance' genre or all the songs of the singer 'Hyo-ri' are selected from the music DB 12 to generate a music list. Therefore, the music list is a list of songs of the genre or singer of the corresponding song name that the user selects or directly inputs through the music recommendation icon, or the songs of the genre or singer that the user directly enters as a query (search condition). It means the entire list.

That is, the list extracting unit 11 selects only songs corresponding to the genre of the query music or the singer of the query music through the genre / singer information of the metadata from the music DB 12, or the query word (the song name or The music DB 12 selects only the songs corresponding to the genre of the quality music or the singer of the quality music.

Thereafter, the similarity measuring apparatus 13 converts the music list (collection of comparison target music) extracted by the list extraction unit 11 into the feature information of the user's current query music (music characteristic information extracted from the original song music 14 of the query) and / Or similarity between the music is calculated by comparing the cumulative characteristic information of the user past query music (music characteristic information extracted from the user past query history information 15). At this time, the feature information such as rhythm, tone, etc. representing the characteristics of the music is basically extracted in advance and stored in the music feature information DB 132 as a text file, it should be noted that may be extracted in real time.

Therefore, when the user selects or directly inputs a song title of the music, feature information of the corresponding music (that is, music characteristic information extracted from the original query music 14) and feature information of each song on the music list are displayed. The similarity is calculated (measurement of similarity between music).

If the user directly enters genre or artist information of the music, the characteristic information of the music based on the user's ID and preference (refer to the user's past query history DB), that is, the user The similarity is calculated (measurement of similarity between music) by comparing the feature information of each piece of music on the music list with music feature information extracted from past query history information (DB) 15.

That is, when the user selects or directly inputs a song title as a query (search condition), the feature information of the music (the feature information of the user's current query music) is compared with the feature of each song on the music list to be compared. Compared with the information, and when the genre or artist information of the music is directly entered as a query (search condition), the cumulative characteristic information of the user's past query music based on the user ID and the preference is compared. The similarity between the music is calculated by comparing with the feature information of each song on the list.

Then, according to the result of the similarity calculation, the music recommendation list 16 is finally generated as a text file in order of high similarity for each genre and artist, and the feature vector information of the user's recent query music is obtained from the user's past query feature vector information ( It accumulates and stores it in the DB of past query history by user.

This series of steps is shown in FIG.

Referring to FIG. 3, when metadata (genre of music, song name, and singer) of user query music exists (301), the music list is selected by genre and singer in the music DB 12 using the metadata. If there is no metadata (304), the searched song name or singer or genre information is searched for by the existing music search algorithm (303). The music DB 12 selects music lists by genres and singers ( 304). At this time, the server's ID information is basically known to the server. Therefore, the server can use the historical query history information (DB) for each user when calculating the similarity.

Then, the similarity and the cumulative characteristics of the selected music (list of comparison target music) and the original music of the query or the user's past query music is calculated (305), and the recommendation list (the predetermined number of songs starting from the highest similarity music) The music recommendation list) is generated and provided to the user (307). In this case, the most recent user query music characteristics are accumulated and stored in the user past query music characteristics (306).

Next, a process of extracting feature information reflecting music characteristics will be described in more detail.

All sound sources are pre-extracted and saved as text files. In this case, as shown in FIG. 4, the text file stores corresponding song information and a feature vector indicating a music characteristic of the song.

In the present invention, the rhythm characteristic and the timbre characteristic important in music psychology are reflected in the feature vector in order to represent the musical characteristic. Many studies of music psychology say that the characteristics of rhythm and timbre are the most fundamental and important when determining the mood of music when humans listen to music. Therefore, in the present invention, the similarity between the music is calculated by extracting the feature vector representing the music characteristic.

First, in order to extract a rhythm feature, an acoustic signal may be digitized to perform a discrete Fourier transform, and then an onset sequence for each frequency band may be detected, and then the rhythm feature may be extracted using an autocorrelation function. . Here, the onset sequence is an important measure of the strength and weakness of the bit of the audio signal, and can be regarded as a sequence showing the flow of the bit strength and weakness of the signal.

For reference, in the process of detecting an onset sequence from an audio signal, first, the audio signal is divided into predetermined frames to detect the onset sequence, and if a discrete Fourier transform is performed for each frame, the energy for each frequency index for each frame Will be available. The low frequency band filtering is performed on the time axis for each frequency index, and then filtered again using a "Canny operator", which is a kind of onset curve detection filter. The "Canny operator" used at this time is well known as a filter for edge detection in the field of image signal processing.

In addition, one onset sequence is obtained by adding all onsets having positive values for each frequency index. This results in an onset sequence through median filtering. This sequence extracts the rhythm feature vector.

The value of the first dimension means the average value of onset per entire frame. The value of the second dimension means the degree of the number of onsets, and the value of the 3rd and 4th dimension means the average value of the local maxima and the local minima of the autocorrelation result using the onset sequence using the autocorrelation function. Is obtained.

In addition, various methods may exist as a rhythm feature vector extraction method.

On the other hand, in order to extract the tone characteristics of music, the music signal is divided into frames having a predetermined length, and feature vectors such as mel-frequency cepstral coefficients (MFCC), linear predictive coding (LPC), and linear predictive cepstral coefficients (LPCC) are obtained for each frame. It can be extracted and used. MFCC, LPCC, LPC and the like are well known in the field of digital signal processing or speech recognition.

In the present invention, a feature vector such as MFCC, LPCC or LPC is used as a feature vector representing a tone of music.

In the case of MFCC, the usefulness of the technique is well known to model human auditory characteristics to reflect the characteristics of speech or musical signals. LPCC and LPC are also well known for their effects by modeling the rough frequency spectrum of signals. After the feature vectors are obtained for each frame, the average and variance of the vector columns may be used as feature vectors, or the average of the vector columns may be used as feature vectors.

The feature vector reflecting the characteristics of the music may be stored as a text file as shown in FIG. 4. In this case, the storage format may be modified in various ways, but information on a corresponding song name and singer of the feature vector should be known. The feature vectors obtained as described above can be collected in a single vector sequence and calculated on the Euclidean distance scale or the cosine distance scale to calculate the similarity between music. The Euclidean or cosine distance scale method used at this time is well known in various fields of engineering and mathematics, and is useful for measuring similarity or distance between vector columns.

Meanwhile, the method of accumulating the most recent user query music characteristics to the user past query music characteristics will be described in detail.

In the user query music feature, the feature of each query music is accumulated and stored for each user. At this time, as shown in FIG. 5, the user's past query music feature vector is cumulatively stored by using the number of user queries or the number of times a specific song is listened to. For example, by adding the number of user queries to the feature vector elements of each song in the list (FIG. 5) and adding them, the vector values of all the songs are added by element, and then each value is divided by the number of songs in the list. May be accumulated and stored. In addition, the user's past query music features may be accumulated and stored for each music genre / singer.

As described above, the method of the present invention may be implemented as a program and stored in a recording medium (CD-ROM, RAM, ROM, floppy disk, hard disk, magneto-optical disk, etc.) in a computer-readable form. Since this process can be easily implemented by those skilled in the art will not be described in more detail.

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

As described above, the present invention enables more effective and personalized personalized music recommendation by extracting and using the characteristics of the music level expressed by the user's preferred intention in the Internet and digital broadcasting environment. However, there is an effect that can be usefully used in the T-commerce field in a digital broadcasting environment.

Claims (24)

In the device for measuring the similarity between music, Music feature extraction means for extracting feature information reflecting the characteristics of the music; Music feature information storage means for storing the extracted feature information as a feature vector sequence; And Similarity calculation means for calculating the similarity between music by comparing the extracted music list based on genre / singer information about the user query music with the feature information of the user's current query music and / or the cumulative feature information of the user's past query music Music characteristic-based similarity measuring device comprising a. The method of claim 1, Updating means for updating the feature information of the user's current query music to the cumulative feature information of the user's past query music Music characteristic-based similarity measuring device further comprising. The method of claim 1, Music recommendation means for generating and providing a recommendation list by a predetermined number of songs in the order of high similarity according to the similarity calculation result; Music characteristic-based similarity measuring device further comprising. The method according to any one of claims 1 to 3, The feature information, And at least one of rhythm or tone characteristic information indicating a higher level music characteristic. The method of claim 4, wherein The rhythm feature information, After performing a discrete Fourier transform by digitizing an acoustic signal, detecting an onset sequence for each frequency band, and using the autocorrelation function, it is a feature vector representing the rhythm characteristics of the extracted music. A similarity measurement apparatus based on music characteristics. The method of claim 4, wherein The tone feature information, The music signal is divided into frames having a predetermined length, and the characteristics of the tone characteristics of one of music, such as Mel-Frequency Cepstral Coefficients (MFCC), Linear Predictive Coding (LPC), or Linear Predictive Cepstral Coefficients (LPCC), extracted for each frame Music characteristic-based similarity measuring device, characterized in that the vector. The method of claim 4, wherein The similarity calculation means, A music characteristic-based similarity measuring device, characterized in that the similarity between music is calculated by collecting feature vectors into one vector column and calculating one of the Euclidean distance scale or the cosine distance scale. In the method of measuring the similarity between music, Extracting feature information reflecting the characteristics of music; And Calculating the similarity between the music by comparing the extracted music list based on genre / singer information on the user query music with the feature information of the user's current query music and / or the cumulative feature information of the user's past query music; Music characteristic-based similarity measurement method comprising a. The method of claim 8, Accumulating and storing feature information of the user's current query music in cumulative feature information of the user's past query music; Music characteristic-based similarity measurement method further comprising. The method of claim 9, The cumulative storage process, A method of measuring similarity based on music characteristics, characterized by adding vector values of all songs by elements and dividing each value by the number of songs in the list to accumulate and store music characteristics of past queries for each user. The method of claim 8, Generating and providing a recommendation list by a predetermined number in order of songs having high similarity among the music lists according to the similarity calculation result; Music characteristic-based similarity measurement method further comprising. The method according to any one of claims 8 to 11, The feature information, And at least one of rhythm or tone characteristic information indicating a higher level music characteristic. The method of claim 12, The rhythm feature information, After performing a discrete Fourier transform by digitizing an acoustic signal, detecting an onset sequence for each frequency band, and using the autocorrelation function, it is a feature vector representing the rhythm characteristics of the extracted music. Music characteristic-based similarity measurement method. The method of claim 13, The process of detecting the onset sequence, After dividing the audio signal into predetermined frames, performing Discrete Fourier Transform on each frame to find the energy per frequency index for each frame, low frequency band filtering is performed on the time axis for each frequency index, and then filtered using the onset curve detection filter. A method of measuring similarity based on music characteristics, characterized in that a single onset sequence is obtained by adding all onsets having a positive value for each frequency index, and again obtaining an onset sequence through median filtering. The method of claim 12, The tone feature information, The music signal is divided into frames having a predetermined length, and the characteristics of the tone characteristics of one of music, such as Mel-Frequency Cepstral Coefficients (MFCC), Linear Predictive Coding (LPC), or Linear Predictive Cepstral Coefficients (LPCC), extracted for each frame Music characteristic-based similarity measurement method characterized in that the vector. The method of claim 15, The tone feature vector is A method of measuring similarity based on music characteristics, characterized in that it is one of an average and a variance of a vector sequence, or an average value of a vector sequence. In the music recommendation system, Music feature extraction means for extracting feature information reflecting the characteristics of the music; Music feature information storage means for storing the extracted feature information as a feature vector sequence; List extracting means for extracting a music list based on genre / singer information on user query music; Similarity calculation means for comparing the extracted music list with the feature information of the user's current query music and / or the cumulative feature information of the user's past query music to calculate the similarity between the music; And Music recommendation means for generating and providing a recommendation list for a predetermined number of songs in the order of high similarity according to the similarity calculation result; Music recommendation system comprising a. The method of claim 17, Updating means for updating the feature information of the user's current query music to the cumulative feature information of the user's past query music Music recommendation system that includes more. The method of claim 17 or 18, The feature information, Music recommendation system, characterized in that at least one of the rhythm or tone feature information indicating the upper level music characteristics. The method of claim 19, The rhythm feature information is obtained by performing digital Fourier transform by digitizing an acoustic signal, detecting an onset sequence for each frequency band, and then using the autocorrelation function to extract the rhythm characteristic of the extracted music. Is a feature vector that represents The tone characteristic information may include at least one piece of music: Mel-Frequency Cepstral Coefficients (MFCC), Linear Predictive Coding (LPC), or Linear Predictive Cepstral Coefficients (LPCC) extracted by dividing a music signal into frames having a predetermined length. A music recommendation system, characterized in that it is a feature vector indicative of the tone characteristics. In the music recommendation method, Extracting feature information reflecting the characteristics of music; Extracting a music list based on genre / singer information about a user query music; Calculating similarity between music by comparing the extracted music list with feature information of user current query music and / or cumulative feature information of user past query music; And Generating and providing a recommendation list by a predetermined number in order of songs having high similarity among the music lists according to the similarity calculation result; Music recommendation method comprising a. The method of claim 21, Accumulating and storing feature information of the user's current query music in cumulative feature information of the user's past query music; Music recommendation method further comprising. The method of claim 21 or 22, The feature information, Music recommendation method characterized in that at least one of the rhythm or sound feature information indicating the music characteristics of the upper level. The method of claim 23, The rhythm feature information is obtained by performing digital Fourier transform by digitizing an acoustic signal, detecting an onset sequence for each frequency band, and then using the autocorrelation function to extract the rhythm characteristic of the extracted music. Is a feature vector that represents The tone characteristic information may include at least one piece of music: Mel-Frequency Cepstral Coefficients (MFCC), Linear Predictive Coding (LPC), or Linear Predictive Cepstral Coefficients (LPCC) extracted by dividing a music signal into frames having a predetermined length. A music recommendation method characterized in that it is a feature vector representing the tone characteristics of the music.

Images