JP5521094B2 - Music classification device and music search system - Google Patents

Description

  The present invention relates to a music classification device and a music search system.

  As conventional music search techniques, for example, techniques described in Patent Documents 1 and 2 and Non-Patent Document 1 are known. In the prior art 1 of Patent Document 1, a song including a character string input by a user in the lyrics is searched, and the song is reproduced from the character string. In Prior Art 2 of Patent Document 2, music is searched based not only on the apparent lyrics but also on the pronunciation (ruby) of the lyrics. In Prior Art 3 of Non-Patent Document 1, annotation information is collected in advance for each piece of music to be searched, and the similarity between songs is estimated from the lyrics and annotation information.

JP 2001-125583 A JP 2008-243155 A

Katsuhiko Tsuji, Junji Hirata, Satoshi Nagao, “Online Music Recommendation System Based on Annotations on Situation and Preference”, Information Processing Society of Japan Research Report, Information Processing Society of Japan, December 12, 2004, Vol. 2004, No. 127, pp.33-38

  However, in the related arts 1 and 2 described above, if the character string of the search keyword and the actual lyrics do not match, the music search is not successful, so it is difficult to conceptually search for the music. In the prior art 3, since annotation information is collected for each piece of music to be searched, there is a problem that the cost for collecting the annotation information is high.

  The present invention has been made in view of such circumstances, and its purpose is to enable conceptual search for music and to suppress the cost of collecting additional information in music search. An object of the present invention is to provide a music classification device and a music search system that can be used.

(1) A music classification device according to the present invention is a category comprising a set of search target music data storage means for storing lyrics data of a search target music, and a category name which is a conceptual expression of the lyrics data and music. A category learning data storage means for accumulating learning data, and a music category for generating a classifier for classifying the music based on the lyrics of the music to the classification destination specified by the category name by learning processing using the category learning data Classifying learning means; music category classification means for applying the lyrics data of the search target music to the classifier to classify the search target music; and music category classification result data for storing data of the classification result of the search target music For each category name, the search target music is stored in the storage means and the music category classification result data stored in the music category classification result data storage means. In-category music re-classification means for performing detailed re-classification, and in-category music re-classification result data storage means for storing the result of re-classification of the search target music, the in-category music re-classification means, For each reclassification, a subcategory important word is extracted based on a feature vector of each search target music data, the importance of the subcategory important word is calculated, and the subcategory important word having the highest importance is set as a subcategory name, The in-category music reclassification result data storage means stores search target music data associated with the subcategory name for each category name, and sets the importance of the subcategory important word and the subcategory important word for each subcategory name. It memorizes.

(2) The music classification device according to the present invention includes feature vector generation means for generating a feature vector of lyrics data, and uses the feature vector of lyrics data to generate the classifier and classify the search target song. It is performed.

(3) The music search system according to the present invention uses the music classification device according to (1) above and the result of classification of the music to be searched by the music classification device to conceptually express the music specified by the user. A music search device that searches for music that matches the subcategory name selection screen of the category name designated by the user on the screen of the display device, and the music search device displays the subcategory name selection screen A display area for each subcategory name for the specified category name is provided above, and a subcategory important word related to the subcategory name is displayed on the display area for the subcategory name. Highlighted according to the importance of subcategory important words.

(4) In the music search system according to the present invention, the music search device supports means for presenting a search result including a plurality of songs to the user, and purchase of the plurality of songs by continuous reproduction or electronic commerce. And means.

  According to the present invention, it is possible to conceptually search for music, and it is possible to obtain an effect that the cost for collecting additional information in searching for music can be suppressed.

It is a block diagram which shows the structure of the music search system which concerns on one Embodiment of this invention. It is a flowchart which shows the procedure of the music classification process which concerns on the embodiment. It is an example of composition of category name selection screen 100 concerning the embodiment. It is an example of composition of subcategory name selection screen 110 concerning the embodiment. It is an example of composition of music list display screen 120 concerning the embodiment.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of a music search system according to an embodiment of the present invention. In FIG. 1, the music search system includes a music classification device 1 and a music search device 2. The music search device 2 supplies a music search result to the music playback device 3. The music playback device 3 plays back music data for playing back music with the speaker 4. For example, the music playback device 3 is a portable music playback device that can be connected to the music search device 2 via a communication line. Alternatively, the music playback device 3 may be provided in a mobile phone and connected to the music search device 2 using the communication function of the mobile phone. The speaker 4 can be of various shapes such as a stationary type, a headphone type, and an earphone type.

  First, the music classification apparatus 1 will be described.

  The music classification device 1 includes a search target music data storage unit 11, a category learning data storage unit 12, a music category classification learning module 13, a music category classification module 14, a music category classification result data storage unit 15, and an in-category music reclassification module 16. And an in-category music reclassification result data storage unit 17.

  The search target music data storage unit 11 stores data of music to be searched (hereinafter referred to as search target music data). The search target music data includes at least text data of the lyrics of the music (hereinafter referred to as lyrics data). In the present embodiment, the search target music data includes music data that can be played back by the music playback device 3 together with the lyrics data. Thereby, the music playback device 3 can play back music using the music data of the music classification device 1.

  The category learning data storage unit 12 accumulates category learning data. The category learning data is data used for learning processing for classifying music using lyrics. The category learning data is composed of a set of lyrics data and a category name. The category name is the name of the music classification destination. A plurality of category names may be paired for one lyric data.

  The category name is a conceptual expression of the music such as the music scene or the world view. For example, seasonal names such as “spring”, “summer”, “autumn”, and “winter”, words representing emotions such as “love” and “youth”, and the like can be used as category names.

  The music category classification learning module 13 generates a classifier for classifying the music based on the lyrics of the music to the classification destination specified by the category name by the learning process using the category learning data. For example, a support vector machine (SVM) can be used to generate the classifier. When using a support vector machine, a classifier is generated for each classification destination specified by a category name. A classifier generated by a support vector machine is referred to as a “category classifier”.

  In the generation of a category classifier relating to a certain category name, first, a feature vector is generated for each lyric data paired with the category name in the category learning data. Next, a category classifier associated with the category name is generated by the support vector machine using the feature vector group.

  When a feature vector of song lyrics data is input to a category classifier associated with a certain category name, the appropriateness of classifying the song to the classification destination specified by the category name is output. Based on this degree of appropriateness, it is possible to determine whether or not to classify the music piece into the classification destination specified by the category name.

  As a method for generating the feature vector of the lyrics data, for example, the lyrics are divided into words by morphological analysis, and a vector having the appearance frequency of each word or the TF * IDF value as an element is generated.

  Further, a characteristic section of the music may be strengthened with respect to the feature vector of the lyrics data, or a section other than the characteristic section of the music may be weakened. For example, when information indicating the most prominent part (rust) of the music is added to the lyrics data, a weight is given to the appearance frequency or TF * IDF value of the word included in the chorus section in the lyrics data. The importance may be increased by a coefficient. Thereby, the feature of the lyrics in the chorus section of the music can be emphasized with respect to the feature vector of the lyrics data. Or, conversely, the importance may be lowered by a weighting factor with respect to the appearance frequency or TF * IDF value of a word included in a section other than the characteristic section of the music (for example, the section of Rap). Thereby, the feature of the lyrics in the sections other than the characteristic section of the music can be weakened with respect to the feature vector of the lyrics data.

  Hereinafter, in this embodiment, description will be made assuming that a category classifier is used.

The music category classification learning module 13 supplies a category classifier relating to each category name to the music category classification module 14.
The music category classification module 14 applies the lyrics data included in the search target music data to the category classifier associated with each category name, and classifies the search target music data. Specifically, first, a feature vector of lyric data included in search target music data is generated. Next, the feature vector is input to the category classifier to obtain the appropriateness. Next, based on the appropriateness level, it is determined whether or not the search target music data is to be classified to the category name classification destination related to the category classifier. As a result, when it is determined to be classified, the search target music data is stored in the music category classification result data storage unit 15 in association with the category name.

  This music category classification process is performed for every category name. That is, for one search target music data, the category classifier for all category names is used, the appropriateness level for each category name is acquired, and it is determined whether or not to classify each category name into the classification destination. The music category classification result data storage unit 15 stores them in association with all the category names determined to be classified. Therefore, a plurality of category names may be associated with one search target music data and stored in the music category classification result data storage unit 15.

  The music category classification result data storage unit 15 stores search target music data associated with the category name as data of the classification result of the search target music data by the music category classification module 14 (hereinafter referred to as music category classification result data). .

  The in-category music reclassification module 16 performs detailed reclassification of search target music data for each category name on the music category classification result data. Specifically, first, for each search target music data associated with the same category name, the lyrics are divided into words by morphological analysis for the lyrics data included in the search target music data, and the frequency of occurrence of each word or A feature vector whose elements are TF * IDF values is generated. Next, clustering processing is performed based on the feature vector of each search target music data, and the search target music data is reclassified. Next, for each reclassification, an important word (hereinafter, referred to as a subcategory important word) is extracted based on the feature vector of each search target music data, the importance of the subcategory important word is calculated, and the subcategory having the highest importance Use important words as subcategory names.

  For example, as a result of performing clustering processing based on the lyrics data, each piece of search target music data associated with the category name “love” is reclassified to the subcategory name “love”, “farewell”, or “unrequited love”. This makes it possible to classify the search target music data into more detailed subcategory names “love”, “parting”, or “unrequited” even if they have the same category name “love”.

  Note that, as a result of the clustering process, with respect to search target music data for which an appropriate reclassification destination has not been obtained, there is no reclassification destination of the search target music data.

The in-category music reclassification module 16 stores the reclassification result of the search target music data in the in-category music reclassification result data storage unit 17.
The in-category music reclassification result data storage unit 17 is a search object associated with a subcategory name for each category name as data of a result of reclassification of search object music data (hereinafter referred to as in-category music reclassification result data). Store music data. Further, for each subcategory name, a set of subcategory important words and importance is stored.

  FIG. 2 is a flowchart showing a procedure of music classification processing according to the present embodiment. In FIG. 2, in step S1, the music category classification learning module 13 generates a category classifier through learning processing using category learning data. In step S2, the music category classification module 14 classifies search target music data using a category classifier. This classification process is performed for all search target music data. In step S3, it is determined whether or not the classification process in step S2 has been completed for all search target music data. When the classification process of step S2 is completed for all search target music data, the process proceeds to step S4.

  In step S4, the in-category music reclassification module 16 reclassifies the search target music data for each category name with respect to the music category classification result data. This reclassification process is performed for all corresponding search target music data for each category name. In step S5, it is determined whether or not the reclassification process in step S4 has been completed for all search target music data corresponding to the category name to be reclassified. If the reclassification process in step S4 has been completed for all corresponding search target music data, the process proceeds to step S6. In step S6, it is determined whether the reclassification process in step S4 has been completed for all category names. When the reclassification process in step S4 is completed for all category names, the process in FIG. 2 is terminated.

  The music classification process by the music classification device 1 described above is performed as a preparation stage before a music search is actually performed.

  Next, the music search device 2 will be described.

  The user operates the music search device 2 to search for music. FIG. 3 is a configuration example of the category name selection screen 100 displayed on the screen of the display device of the music search device 2. On the category name selection screen 100, a display area 101 for each category name is provided. The music search device 2 acquires all category names included in the music category classification result data from the music classification device 1 and generates a category name selection screen 100.

  The user can specify the category name display area 101 to be searched on the category name selection screen 100 by operating the pointer 102. The music search device 2 acquires, from the music classification device 1, a set of all subcategory names and subcategory important words and importance related to each subcategory name from the music classification device 1 for the category name related to the display area 101 designated by the pointer 102. Generate name selection screen.

  FIG. 4 is a configuration example of the subcategory name selection screen 110 displayed on the screen of the display device of the music search device 2. The configuration example of the subcategory name selection screen 110 is for the category name “love”, and “love”, “parting”, and “unrequited” are set as the subcategory names. On the subcategory name selection screen 110, a display area 111 for each subcategory name is provided. In the example of FIG. 4, a display area 111 a for the subcategory name “farewell”, a display area 111 b for “love”, and a display area 111 c for “unrequited love” are provided. Further, on each display area 111, a subcategory important word related to the subcategory name is displayed. For example, the sub-category important words “like”, “love”, “emotion”, and “you” are displayed on the display area 111b of “love”. Furthermore, the subcategory important words are highlighted according to their importance. The highlighting method includes changing the character size, character color, and the like. In the example of FIG. 4, the larger the importance, the larger the character size.

  Although not illustrated in FIG. 4, if there is search target music data without a subcategory name (no reclassification destination), a display area 111 without a subcategory name is provided on the subcategory name selection screen 110. .

  The user can operate the pointer 102 to specify the display area 111 of the subcategory name to be searched or the display area 111 without the subcategory name on the subcategory name selection screen 110. The music search device 2 creates a list of music of search target music data for the subcategory name or no subcategory name related to the display area 111 designated by the pointer 102. This music list is displayed on the screen as a music search result.

  FIG. 5 is a configuration example of the search result music list display screen 120 displayed on the screen of the display device of the music search device 2. On the music list display screen 120, for the music to be searched and searched, an identification number (ID) field, a music name (Title) field, an artist name (Artist) field, and an instruction to play music are displayed. A play button (PLAY) field is provided.

  The user can operate the pointer 102 (not shown in FIG. 5) to specify a desired music playback button on the music list display screen 120. The music search device 2 acquires the music data of the music related to the playback button designated by the pointer 102 from the music classification device 1 and supplies the music data to the music playback device 3. The music playback device 3 plays back the music data. Thereby, the user can listen to the music of the search result.

  Note that a list of songs of search target song data related to the category name specified by the user may be created as a list of songs as a search result.

  In addition, the music search device 2 sequentially acquires the music data of the music included in the search result music list from the music classification device 1 and supplies the music data to the music playback device 3, and the music playback device 3 sequentially supplies the music data. Alternatively, it may be reproduced continuously.

  Alternatively, the music search device 2 may present a list of music as a search result to the user so that the user can purchase all or part of the music included in the music list by electronic commerce.

  As described above, according to the present embodiment, the user can obtain a search result of music by designating a conceptual expression (category name, subcategory name) that represents the music that the user wants to listen to. Therefore, even if the user does not know the lyrics of the music to be searched or does not want to search for a specific music, the user can search for the music by an ambiguous expression such as the atmosphere of the music.

  Further, according to the present embodiment, it is not necessary to collect annotation information for each piece of search target music data, and the cost for collecting annotation information can be suppressed.

  Further, by enhancing the feature of the lyrics in the chorus section of the song with respect to the feature vector of the lyrics data, it is possible to perform song classification that properly captures the musical expression feature of the song. Thereby, it can be expected that the search result of the music further matches the search intention of the user.

  Note that the music classification device 1 and the music search device 2 according to the present embodiment may be realized by dedicated hardware, or may be configured by a computer system such as a personal computer, and each shown in FIG. You may implement | achieve the function by running the program for implement | achieving the function of each part of the apparatuses 1 and 2. FIG.

In addition, an input device, a display device, and the like (none of which are shown) are connected to the music classification device 1 and the music search device 2 as peripheral devices. Here, the input device refers to an input device such as a keyboard and a mouse. The display device refers to a CRT (Cathode Ray Tube), a liquid crystal display device or the like.
The peripheral device may be connected directly to the devices 1 and 2 or may be connected via a communication line.

Further, a program for realizing the function of the music classification device 1 or the music search device 2 shown in FIG. 1 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system, By executing, music classification processing or music search processing may be performed. Here, the “computer system” may include an OS and hardware such as peripheral devices.
Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used.
“Computer-readable recording medium” refers to a flexible disk, a magneto-optical disk, a ROM, a writable nonvolatile memory such as a flash memory, a portable medium such as a DVD (Digital Versatile Disk), and a built-in computer system. A storage device such as a hard disk.

Further, the “computer-readable recording medium” means a volatile memory (for example, DRAM (Dynamic DRAM) in a computer system that becomes a server or a client when a program is transmitted through a network such as the Internet or a communication line such as a telephone line. Random Access Memory)), etc., which hold programs for a certain period of time.
The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line.
The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

As mentioned above, although embodiment of this invention was explained in full detail with reference to drawings, the specific structure is not restricted to this embodiment, The design change etc. of the range which does not deviate from the summary of this invention are included.
For example, in the above-described embodiment, the search target music data includes the music data. However, when the search target music data does not include the music data, the music playback device 3 is connected via the communication network. The music data may be acquired by accessing the music database and downloading the music data, or by reading the music data from a recording medium on which the music data is recorded.

DESCRIPTION OF SYMBOLS 1 ... Music classification apparatus, 2 ... Music search apparatus, 11 ... Search object music data storage part, 12 ... Category learning data storage part, 13 ... Music category classification learning module, 14 ... Music category classification module, 15 ... Music category classification result Data storage unit, 16 ... in-category music reclassification module, 17 ... in-category music reclassification result data storage unit

Claims (4)

Search object music data storage means for storing lyrics data of the music to be searched;
Category learning data storage means for storing category learning data consisting of a set of lyric data and a category name that is a conceptual expression of music;
Music category classification learning means for generating a classifier for classifying the music based on the lyrics of the music to the classification destination specified by the category name by learning processing using category learning data;
Applying lyric data of search target music to the classifier, and music category classification means for classifying search target music;
Music category classification result data storage means for storing the classification result data of the search target music;
In-category music reclassification means for performing detailed reclassification of the search target music for each category name with respect to the music category classification result data stored in the music category classification result data storage means;
In-category music reclassification result data storage means for storing the result of reclassification of the search target music,
The in-category music reclassification means extracts, for each reclassification, a subcategory important word based on a feature vector of each search target music data, calculates the importance of the subcategory important word, and has the highest importance Subcategory important words are subcategory names,
The in-category music reclassification result data storage means stores search target music data associated with the subcategory name for each category name, and sets the importance of the subcategory important word and the subcategory important word for each subcategory name. Remember,
The music classification apparatus characterized by this. A feature vector generating means for generating a feature vector of the lyrics data;
The music classification apparatus according to claim 1, wherein the classifier is generated and the search target music is classified using a feature vector of lyrics data. A music classification device according to claim 1;
Using the result of classification of the music to be searched by the music classification device, and a music search device that searches for music that matches the conceptual expression of the music specified by the user,
The music search device displays a subcategory name selection screen of a category name designated by the user on the screen of the display device,
On the sub category name selection screen, a display area for each sub category name for the designated category name is provided,
On the display area of the subcategory name, a subcategory important word related to the subcategory name is displayed,
The subcategory important words are highlighted according to the importance of the subcategory important words.
A music search system characterized by this. The music search device includes:
Means for presenting search results including a plurality of songs to the user;
Means for supporting purchase by continuous reproduction or electronic commerce for the plurality of songs;
The music search system according to claim 3, further comprising:

Images