JP5085577B2 - Playlist creation device, music playback device, playlist creation method, and playlist creation program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To give a story to a play list by creating the play list considering a feature amount which is specific for a musical piece. <P>SOLUTION: The play list creation device 10 for creating the play list in which a play back order of musical piece data are predetermined, is provided with: a feature amount extraction section 12 which reads the musical data from a database 11, and extracts a plurality of kinds of feature amounts from the read musical piece data; a model construction section 13 for creating a musical piece transitional model by calculating a transitional probability when any one of musical piece data is changed to the other musical piece data, based on the extracted feature amount; a musical piece selection section 14 in which a plurality of musical piece data are selected from the database 11, based on the created musical piece transitional model and input information by a user, to make a list; and an output section 15 for displaying the listed information. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a playlist creation device, a song playback device, a playlist creation method, and a playlist creation program that create a playlist that defines the playback order of song data.

  Conventionally, for example, Patent Document 1 calculates the degree of relevance based on the query music specified by the user and the meta information of the target music in order to allow the user to create a playlist easily and automatically. A technique related to automatic playlist generation using content is disclosed. Specifically, the related word (A) is extracted from the thesaurus dictionary database for the meta information or keywords of the query music, and the music holding the metadata related to (A) is registered in the playlist.

  Further, in Patent Document 2, in order to generate a playlist, the selected music (query) and the music common to the music information (song name, artist name, album name, genre name, etc.) in the database are narrowed down. Thus, a technique for generating or supporting a playlist is disclosed.

  Further, in Patent Document 3, in playlist search, a playlist in a database is associated with story information representing a transition of music when the corresponding playlist is viewed in playback order, and the playlist is searched based on the similarity of the story information. Techniques to do this are disclosed. Here, “story information” includes, for example, the order of genre information of reproduced music, the order of tempo information, and the like.

  Further, Patent Document 4 uses collaborative filtering to search for a playlist in which the listening situation at the time of generation is close to the current listening situation of the listener from playlists generated in the past, and the listener's A technique that can be modified to adapt to current preference information and listening status is disclosed. Here, the “listening situation” is based on annotation collection, and uses information on what scene is sung from the song and lyrics and in what situation.

  Also, in Patent Document 5, in order to register music that the user has never heard in the playlist, the degree of order relation between the songs in the existing playlist is calculated, and based on the degree of order relation between the songs. A technique for automatically generating a playlist is disclosed. Here, the “order relation” is calculated based on a combination and order of specific music appearing in common in a plurality of existing playlists, and a pattern having a combination of music order appearing frequently has a strong order relation. Judging.

  Further, Non-Patent Documents 1 and 2 disclose a music search system based on the acoustic features of music, and Non-Patent Documents 3 and 4 disclose tools for extracting acoustic features. Yes.

JP 2008-217254 A JP 2008-158047 A JP 2007-287256 A JP 2006-268100 A JP 2003-178088 A

K. Hoashi, et al: Feature space modification for content-based music retrieval based on user preferences, Proc. Of ICASSP 2006, pp. 517-520, 2006. Yoshii, K. Goto, M. Komatani, K. Ogata, T. and Okuno, H .: Improving Efficiency and Scal-ability of Model-based Music RecommenderSystem Based on Incremental Training, Proc.ISMIR, pp.89-94, Vienna , Sep. (2007). jAudio: http://sourceforge.net/projects/jaudio/ MARSYAS: http://marsyas.sness.net/

  However, in the techniques disclosed in Patent Documents 1 to 3, music meta information is used for generating a playlist, but it is difficult to apply to a database in which meta information or the like is not maintained. Moreover, even if meta information is added, the accuracy may be a problem. Moreover, in patent document 4, if an existing playlist itself is not provided with an annotation, it cannot be used, and it is difficult for an individual to collect a large number of playlists that have been annotated. Moreover, in Patent Document 5, although it is not necessary to use annotation information or the like by using the order relation of existing playlists, there is a combination of existing playlist collection and a common song order for the collected playlists. It is difficult to generate an effective playlist when there are a small number or when it does not exist.

  As described above, in the prior art, it is difficult to generate the playlist in consideration of the swell and the story in the playlist in consideration of the acoustic characteristics, that is, the reproduction order. On the other hand, in the prior art, a playlist that takes into account the acoustic features can be generated by using the result of the music search system as a playlist, but in such a case, it has similar acoustic features. It is easy to create a playlist composed of songs, and it is difficult to say that the story of the playlist is taken into consideration.

  The present invention has been made in view of such circumstances, and a playlist creation device capable of giving a playlist a story by creating a playlist in consideration of the characteristic amount unique to the music, An object of the present invention is to provide a music player, a playlist creation method, and a playlist creation program.

  (1) In order to achieve the above object, the present invention takes the following measures. That is, the playlist creation device of the present invention is a playlist creation device that creates a playlist in which the playback order of music data is determined, reads music data from a database, and uses a plurality of features from the read music data Based on the extracted feature quantity, a feature quantity extraction unit that extracts the quantity, calculates a transition probability when transitioning from any one piece of music data to another piece of music data in which the playback order is adjacent, A model construction unit that creates a transition model, a music selection unit that selects and lists a plurality of music data from the database based on the created music transition model and user input information, and the list And an output unit for displaying information.

  In this way, based on the feature amount, the transition probability when transitioning from any one piece of music data to another piece of music data adjacent in the playback order is calculated, a music transition model is created, and the music transition model is created. And multiple pieces of music data are selected from the database based on the information input by the user and created as a list, so even if the meta information data is not maintained, a playlist with storylines should be created. Can do. In addition, since the characteristic quantities unique to the music are statistically modeled, it is not necessary to use a plurality of sets of music or many samples including specific music, and it is possible to provide a highly convenient playlist. . For example, there is an advantage that a statistical model can be constructed from the order of albums at hand rather than collecting playlists of others.

  (2) Moreover, the playlist creation device of the present invention is a playlist creation device that creates a playlist in which the reproduction order of music data is determined, reads music data from a database, and a plurality of data from the read music data Based on the extracted feature quantity, a feature quantity extraction unit that extracts types of feature quantities, creates a plurality of clusters composed of at least one piece of music data, and transitions from any one cluster to another cluster Calculating a transition probability at the time of performing, and creating a cluster transition model, and selecting and listing a plurality of music data from the database based on the created cluster transition model and user input information A music selection section and an output section for displaying the listed information are provided.

  In this way, based on the feature amount, the transition probability when transitioning from any one piece of music data to another piece of music data adjacent in the playback order is calculated, a music transition model is created, and the music transition model is created. And multiple pieces of music data are selected from the database based on the information input by the user and created as a list, so even if the meta information data is not maintained, a playlist with storylines should be created. Can do. In addition, since the characteristic quantities unique to the music are statistically modeled, it is not necessary to use a plurality of sets of music or many samples including specific music, and it is possible to provide a highly convenient playlist. . For example, there is an advantage that a statistical model can be constructed from the order of albums at hand rather than collecting playlists of others. Furthermore, since the music data is clustered, it is possible to eliminate the inconvenience when a plurality of combinations of the same music do not exist.

  (3) Further, in the playlist creation device of the present invention, the music selection unit selects music data from each cluster based on a music relation level indicating a relation level between music pieces.

  In this way, music data is selected from each cluster based on the music relationship degree indicating the degree of relation between music pieces, so that it is possible to select a desired music piece when using the cluster transition model. For example, it is possible to select a music piece having a high similarity in the feature amount of the music data, or to select a music piece having a high similarity in meta information.

  (4) In the playlist creation device of the present invention, the music selection unit selects a plurality of music data from the database based on the music specified by the user and the rank of the music in the playlist. It is characterized by listing.

  Thus, since a plurality of music data is selected from the database and listed based on the music specified by the user and the ranking of the music in the playlist, the user specifies the starting point in the playlist. be able to.

  (5) Moreover, the music reproducing device of the present invention includes any one of the playlist creating devices described above, and reproduces music data based on the created list.

  Thereby, even in a situation where meta information data is not maintained, a playlist having story characteristics can be created. In addition, since the characteristic quantities unique to the music are statistically modeled, it is not necessary to use a plurality of sets of music or many samples including specific music, and it is possible to provide a highly convenient playlist. . For example, there is an advantage that a statistical model can be constructed from the order of albums at hand rather than collecting playlists of others.

  (6) Further, the playlist creation method of the present invention is a playlist creation method for creating a playlist in which the reproduction order of the music data is determined. In the feature quantity extraction unit, the music data is read from the database, and the reading is performed. A step of extracting a plurality of types of feature values from the recorded music data, and in the model construction unit, based on the extracted feature values, from one of the music data to another music data having an adjacent playback order Calculating a transition probability at the time of transition, creating a music transition model, and selecting a plurality of music data from the database based on the created music transition model and user input information in the music selection unit; And a step of displaying the listed information in the output unit. It is characterized in that it also comprises.

  In this way, based on the feature amount, the transition probability when transitioning from any one piece of music data to another piece of music data adjacent in the playback order is calculated, a music transition model is created, and the music transition model is created. And multiple pieces of music data are selected from the database based on the information input by the user and created as a list, so even if the meta information data is not maintained, a playlist with storylines should be created. Can do. In addition, since the characteristic quantities unique to the music are statistically modeled, it is not necessary to use a plurality of sets of music or many samples including specific music, and it is possible to provide a highly convenient playlist. . For example, there is an advantage that a statistical model can be constructed from the order of albums at hand rather than collecting playlists of others.

  (7) Further, the playlist creation method of the present invention is a playlist creation method for creating a playlist in which the reproduction order of the music data is determined. In the feature quantity extraction unit, the music data is read from the database, and the reading is performed. Extracting a plurality of types of feature values from the recorded music data, and creating a plurality of clusters composed of at least one song data based on the extracted feature values in the model construction unit, and any one of the clusters Calculating a transition probability when transitioning from one to another cluster, creating a cluster transition model, and in the music selection unit, based on the created cluster transition model and input information by the user, from the database A step of selecting and listing a plurality of music data; and It is characterized by comprising the steps of: displaying a reduction information.

  In this way, based on the feature amount, the transition probability when transitioning from any one piece of music data to another piece of music data adjacent in the playback order is calculated, a music transition model is created, and the music transition model is created. And multiple pieces of music data are selected from the database based on the information input by the user and created as a list, so even if the meta information data is not maintained, a playlist with storylines should be created. Can do. In addition, since the characteristic quantities unique to the music are statistically modeled, it is not necessary to use a plurality of sets of music or many samples including specific music, and it is possible to provide a highly convenient playlist. . For example, there is an advantage that a statistical model can be constructed from the order of albums at hand rather than collecting playlists of others. Furthermore, since the music data is clustered, it is possible to eliminate the inconvenience when a plurality of combinations of the same music do not exist.

  (8) Further, the playlist creation program of the present invention is a playlist creation program for creating a playlist in which the reproduction order of the music data is determined. The playlist creation program reads music data from a database, Based on the process of extracting the type of feature quantity, and based on the extracted feature quantity, the transition probability at the time of transition from any one piece of music data to another piece of music data whose playback order is adjacent is calculated, A process for creating a transition model, a process for selecting and listing a plurality of music data from the database based on the created music transition model and user input information, and displaying the listed information And a series of processes including the processes are converted into commands that can be read and executed by the computer. To have.

  In this way, based on the feature amount, the transition probability when transitioning from any one piece of music data to another piece of music data adjacent in the playback order is calculated, a music transition model is created, and the music transition model is created. And multiple pieces of music data are selected from the database based on the information input by the user and created as a list, so even if the meta information data is not maintained, a playlist with storylines should be created. Can do. In addition, since the characteristic quantities unique to the music are statistically modeled, it is not necessary to use a plurality of sets of music or many samples including specific music, and it is possible to provide a highly convenient playlist. . For example, there is an advantage that a statistical model can be constructed from the order of albums at hand rather than collecting playlists of others.

  (9) Further, the playlist creation program of the present invention is a playlist creation program for creating a playlist in which the reproduction order of music data is determined, and reads music data from a database. Processing for extracting types of feature quantities, and creating a plurality of clusters composed of at least one piece of music data based on the extracted feature quantities, and transitioning from any one cluster to another cluster A process for calculating a transition probability and creating a cluster transition model; a process for selecting and listing a plurality of music data from the database based on the created cluster transition model and user input information; and the list A series of processes, including the process of displaying computerized information, can be read by a computer And it is characterized in executable manner that the command of.

  In this way, based on the feature amount, the transition probability when transitioning from any one piece of music data to another piece of music data adjacent in the playback order is calculated, a music transition model is created, and the music transition model is created. And multiple pieces of music data are selected from the database based on the information input by the user and created as a list, so even if the meta information data is not maintained, a playlist with storylines should be created. Can do. In addition, since the characteristic quantities unique to the music are statistically modeled, it is not necessary to use a plurality of sets of music or many samples including specific music, and it is possible to provide a highly convenient playlist. . For example, there is an advantage that a statistical model can be constructed from the order of albums at hand rather than collecting playlists of others. Furthermore, since the music data is clustered, it is possible to eliminate the inconvenience when a plurality of combinations of the same music do not exist.

  According to the present invention, a playlist having story characteristics can be created even in a situation where meta information data is not maintained. In addition, since the characteristic quantities unique to the music are statistically modeled, it is not necessary to use a plurality of sets of music or many samples including specific music, and it is possible to provide a highly convenient playlist. .

It is a block diagram which shows schematic structure of the playlist production apparatus which concerns on this embodiment. It is a flowchart which shows the operation | movement which produces | generates a play list based on the music data used as the base point selected by the user. It is a flowchart which shows play list creation operation | movement using a music transition probability, when unknown music data are input. It is a flowchart which shows the play list creation operation | movement using a cluster transition model. It is a figure which shows the display and input image in a user input part.

  Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a schematic configuration of a playlist creation apparatus according to the present embodiment. The playlist creation apparatus 10 includes a database 11, a feature amount extraction unit 12, a model construction unit 13, a music selection unit 14, an output unit 15, and a user input unit 16. The database 11 stores music data and feature amounts obtained by a feature amount extraction unit 12 that extracts acoustic feature amounts.

  The feature amount extraction unit 12 extracts the feature amount of the music from the music data stored in the database 11. As the feature amount, for example, parameters based on sound such as tempo, rhythm, tone color, sound pressure, volume, tone color, and tonality can be used as the feature amount. As a specific method, the tools described in Non-Patent Documents 3 and 4 can be used. In addition, the feature quantities are not necessarily limited to those described above, and parameters necessary for estimating them can be used. For example, spectrum information used in frequency analysis, power value for each band, and the like can be given. It should be noted that not only the acoustic features but also other features such as TFIDF can be extracted from the lyrics information and the like, and the characteristic amount unique to the music can be extracted and used. Further, the feature amount extraction unit 12 does not necessarily need to extract the feature amount of the entire music. For example, it is possible to extract in units of structure such as rust and A melody of music, and to extract feature quantities such as the beginning and end sections. It is also possible to simply use a song name, a song identification number, or the like.

  The model construction unit 13 constructs a music transition model based on the feature quantity obtained by the feature quantity extraction unit 12 for the music in the learning data, for example, the order of the music in the album or the existing playlist. For example, a known N-gram model may be generated by calculating a transition probability between songs from the arrangement order of songs in a commercially available album as learning data. N-gram is a model that estimates P (w1w2... Wn) by the following equation.

本発明では、例えば、ある学習データ内で連続したｎ曲目〜（ｎ＋２）曲目の特徴ベクトルをそれぞれ、ｓ＿ｎ、ｓ＿（ｎ＋１）、ｓ＿（ｎ＋２）とすると、２−ｇｒａｍ、３−ｇｒａｍモデルは、次式にて表現することができる。

In the present invention, for example, when the feature vectors of the n-th to (n + 2) -th continuous music in certain learning data are respectively s_n, s_ (n + 1), and s_ (n + 2), the 2-gram and 3-gram models are It can be expressed by the following formula.

ここで、Ｐ（ｓ＿ｎ）とは、学習データ内にｓ＿ｎが出現する確率を表し、
Ｐ（ｓ＿（ｎ＋１），ｓ＿ｎ）は、特徴量ｓ＿ｎ、と、ｓ＿（ｎ＋１）が連続して出現する確率を表している。

Here, P (s_n) represents the probability that s_n appears in the learning data,
P (s_ (n + 1), s_n) represents the probability that the feature quantity s_n and s_ (n + 1) appear successively.

  In this model, N = 2, 3,... Can be used.

  In the above-described model, there are not always a plurality of combinations of the same music pieces. That is, the same problem as in Patent Document 5 exists. However, it is not necessary to build a model for each song, and by applying clustering to the learning data group in advance, it can be improved by using a transition model between song clusters instead of a transition model between songs .

  For example, it can be expressed as follows.

ここで、Ｎ（Ｃｎ）ｈａ学習データ内に該当クラスタに所属する楽曲が現れる頻度として表される。クラスタリングはＫ−ｍｅａｎｓ法などを適用することができる。

Here, it is expressed as the frequency at which the music belonging to the corresponding cluster appears in the N (Cn) ha learning data. For clustering, a K-means method or the like can be applied.

  With such a model, it is possible to generate a playlist that includes a story along the learning data. In addition, since the user can generate a playlist simply by selecting the starting music, it can be applied when meta information or the like is not prepared, and does not require information from others. .

  In addition to estimating the next song feature, it is also possible to estimate the feature that will be the previous song when a certain feature is given. A playlist in which specific music is registered at a position can be generated. In that case, it is possible to cope with the problem by replacing n + 1 n + 2 with n-1 n-2 in all of the above-described equations. Hereinafter, a model using n + 1 n + 2 is referred to as a forward chain model, and a model using n−1 n−2 is referred to as a reverse chain model.

  The user input unit 16 can display and input music and parameters necessary for creating a playlist. For example, the position of the reproduction order, the total number of songs in the playlist, and the like can be designated. It is also possible to modify the automatically generated playlist. FIG. 5 shows a display and input image in the user input unit 16.

  The music selection unit 14 extracts music data from the database 11 based on the music transition model constructed by the model construction unit 13 and the information input by the user input unit 16. For example, music data having the maximum transition probability obtained from the music transition model or music data can be selected from the transition destination cluster obtained from the cluster transition model.

  When a cluster transition model is used, music can be selected from the transition destination cluster by using the degree of music relation. The music relationship degree can be calculated, for example, by selecting a music piece having a high music feature amount similarity or calculating meta information similarity in a transition destination cluster.

  The music feature amount can be extracted not only for the whole music but also for each music head / end section, music structure (A melody, chorus), and the like. Further, as the music feature amount, a feature amount such as timbre, tempo, rhythm, text, etc. can be used, and similarity calculation etc. can be performed by comparing the integrated feature amount groups, Similarity calculation can be performed. The output unit 15 outputs a playlist. It is also possible to play music data.

  FIG. 2 is a flowchart illustrating an operation of generating a playlist based on music data serving as a base point selected by the user. First, when music data and parameters necessary for creating a playlist are input by the user (step S1), whether the ranking in the playlist is the top, the middle, or the end. It is determined (step S2). If it is the head, the music is selected by the forward chain model (step S3). And it is judged whether the music selection by a forward chain model was complete | finished (step S4), and when not complete | finished, it transfers to step S3. On the other hand, when the music selection by the forward chain model is completed in step S4, the process proceeds to step S5, and a playlist is output (step S5).

  In step S2, if the rank in the playlist is at the end, the music is selected by the reverse chain model (step S6). And it is judged whether the music selection by a reverse chain model was complete | finished (step S7), and when not complete | finished, it transfers to step S6. On the other hand, when the music selection by the reverse chain model is completed in step S7, the process proceeds to step S5, and the playlist is output (step S5).

  In step S2, if the rank in the play list is intermediate, music selection by the forward chain model is performed (step S8). And it is judged whether the music selection by a forward chain model was complete | finished (step S9), and when not complete | finished, it transfers to step S8. On the other hand, when the music selection by the forward chain model is completed in step S9, the music selection by the reverse chain model is performed (step S10). And it is judged whether the music selection by a reverse chain model was complete | finished (step S11), and when not complete | finished, it transfers to step S10. On the other hand, when the music selection by the reverse chain model is completed in step S11, the process proceeds to step S5, and a playlist is output (step S5).

  In step S2, when the rank in the playlist is intermediate, the order of the forward chain model and the reverse chain model can be switched. For example, with regard to the forward chain, it is possible to loop from the position serving as the base point to the final music number, and with respect to the reverse chain, it is possible to loop to the starting point music.

  The flowchart of FIG. 2 shows a case where the music specified by the user exists in the database 11 in advance, but for example, a music that does not exist in the database 11 can be input. In that case, a feature amount extraction step may be added before step S2.

  FIG. 3 is a flowchart showing the playlist creation operation using the music transition probability when unknown music data is input. First, when music data (unknown music data) and parameters necessary for creating a playlist are input by the user (step S20), the feature amount of the music data is extracted (step S21). The above-described method can be used as a feature amount extraction method. Next, it is determined whether the rank in the play list is the top, the middle, or the end (step S22).

  In step S22, if it is the head, the music is selected by the forward chain model (step S23). And it is judged whether the music selection by a forward chain model was complete | finished (step S24), and when not complete | finished, it transfers to step S23. On the other hand, when the music selection by the forward chain model is completed in step S24, the process proceeds to step S25, and the playlist is output (step S25).

  In step S22, if the rank in the play list is the end, the music piece is selected by the reverse chain model (step S26). And it is judged whether the music selection by a reverse chain model was complete | finished (step S27), and when not complete | finished, it transfers to step S26. On the other hand, when the music selection by the reverse chain model is completed in step S27, the process proceeds to step S25, and a play list is output (step S25).

  In step S22, if the rank in the play list is intermediate, music selection is performed using the forward chain model (step S28). And it is judged whether the music selection by a forward chain model was complete | finished (step S29), and when not complete | finished, it transfers to step S28. On the other hand, when the music selection by the forward chain model is completed in step S29, the music selection by the reverse chain model is performed (step S30). And it is judged whether the music selection by a reverse chain model was complete | finished (step S31), and when not complete | finished, it transfers to step S30. On the other hand, when the music selection by the reverse chain model is completed in step S31, the process proceeds to step S25, and a play list is output (step S25).

  In step S22, when the rank in the playlist is intermediate, the order of the forward chain model and the reverse chain model can be switched. For example, with regard to the forward chain, it is possible to loop from the position serving as the base point to the final music number, and with respect to the reverse chain, it is possible to loop to the starting point music.

  FIG. 4 is a flowchart showing the playlist creation operation using the cluster transition model. First, when music data and parameters necessary for creating a playlist are input by the user (step S40), the feature amount of the music data is extracted (step S41). The above-described method can be used as a feature amount extraction method. Next, it is determined whether the rank in the play list is the top, the middle, or the end (step S42).

  In step S42, if it is the head, a cluster is selected by the forward chain model (step S43). Next, a music piece is selected based on the degree of relation (music degree of relation) (step S44). Then, it is determined whether or not the music selection based on the degree of relation has been completed (step S45). If it has not been completed, the process proceeds to step S43. On the other hand, when the music selection based on the degree of relationship is completed in step S45, the process proceeds to step S46, and a play list is output (step 46).

  In step S42, if the rank in the playlist is at the end, a cluster is selected by the reverse chain model (step S47). Next, the music is selected based on the degree of relation (music degree of relation) (step S48). Then, it is determined whether or not the music selection based on the degree of relationship has been completed (step S49). If it has not been completed, the process proceeds to step S47. On the other hand, when the music selection based on the degree of relationship is completed in step S49, the process proceeds to step S46, and a play list is output (step S46).

  In step S42, if the rank in the playlist is intermediate, a cluster is selected by the forward chain model (step S50). Next, a music piece is selected based on the degree of relation (music degree of relation) (step S51). Then, it is determined whether or not the music selection based on the degree of relation has been completed (step S52), and if not completed, the process proceeds to step S50. On the other hand, when the music selection based on the degree of relation is completed in step S52, the cluster is selected by the reverse chain model (step S53). Next, music selection based on the degree of relationship is performed (step S54). Then, it is determined whether or not the music selection based on the degree of relation has been completed (step S55). If it has not been completed, the process proceeds to step S53. On the other hand, when the music selection based on the degree of relationship is completed in step S55, the process proceeds to step S46, and a playlist is output (step S46).

  In step S42, when the rank in the playlist is intermediate, the order of the forward chain model and the reverse chain model can be switched. For example, with regard to the forward chain, it is possible to loop from the position serving as the base point to the final music number, and with respect to the reverse chain, it is possible to loop to the starting point music.

  As described above, according to the present embodiment, based on the feature amount, the transition probability at the time of transition from any one piece of music data to another piece of music data whose playback order is adjacent is calculated, and the music transition A model is created, and multiple song data are selected from the database and listed based on the song transition model and user input information, so even in situations where meta information data is not maintained, You can create your own playlist. In addition, since the characteristic quantities unique to the music are statistically modeled, it is not necessary to use a plurality of sets of music or many samples including specific music, and it is possible to provide a highly convenient playlist. . For example, there is an advantage that a statistical model can be constructed from the order of albums at hand rather than collecting playlists of others. Furthermore, since the music data is clustered, it is possible to eliminate the inconvenience when a plurality of combinations of the same music do not exist.

DESCRIPTION OF SYMBOLS 10 Playlist production apparatus 11 Database 12 Feature-value extraction part 13 Model construction part 14 Music selection part 15 Output part 16 User input part

Claims (9)

A playlist creation device for creating a playlist that defines the playback order of music data,
A feature value extraction unit that reads out music data from a database and extracts a plurality of types of feature values from the read music data;
Based on the extracted feature amount, a model construction unit that calculates a transition probability when transitioning from any one piece of music data to another piece of music data in which the reproduction order is adjacent, and creates a music transition model;
A music selection unit that selects and lists a plurality of music data from the database based on the created music transition model and input information by the user;
And an output unit for displaying the listed information. A playlist creation device for creating a playlist that defines the playback order of music data,
A feature value extraction unit that reads out music data from a database and extracts a plurality of types of feature values from the read music data;
Based on the extracted feature amount, create a plurality of clusters composed of at least one piece of music data, calculate a transition probability when transitioning from any one cluster to another cluster, A model building part to be created;
Based on the created cluster transition model and input information by the user, a music selection unit that selects and lists a plurality of music data from the database;
And an output unit for displaying the listed information.   3. The playlist creation apparatus according to claim 2, wherein the music selection unit selects music data from each cluster based on a music relation degree indicating a relation degree between music pieces.   The music selection section selects and lists a plurality of music data from the database based on a music specified by a user and a rank of the music in a playlist. Item 4. The playlist creation device according to any one of items 3 to 4.   5. A music reproducing apparatus comprising the playlist creating apparatus according to claim 1, wherein music data is reproduced based on the created list. A playlist creation method for creating a playlist that defines a playback order of music data,
In the feature quantity extraction unit, reading out music data from a database, and extracting a plurality of types of feature quantities from the read music data;
In the model construction unit, based on the extracted feature amount, the transition probability when transitioning from any one piece of music data to another piece of music data in which the playback order is adjacent is calculated, and a music transition model is created Steps,
In the music selection unit, based on the created music transition model and user input information, selecting and listing a plurality of music data from the database;
The output section includes at least a step of displaying the listed information. A playlist creation method for creating a playlist that defines a playback order of music data,
In the feature quantity extraction unit, reading out music data from a database, and extracting a plurality of types of feature quantities from the read music data;
The model construction unit creates a plurality of clusters composed of at least one piece of music data based on the extracted feature amount, and calculates a transition probability when transitioning from any one cluster to another cluster. Creating a cluster transition model;
In the music selection unit, based on the created cluster transition model and user input information, selecting and listing a plurality of music data from the database;
And a step of displaying the listed information in an output unit. A playlist creation program for creating a playlist that defines the playback order of music data,
A process of reading music data from the database and extracting a plurality of types of feature values from the read music data;
Based on the extracted feature amount, a process of calculating a transition probability when transitioning from any one piece of music data to another piece of music data in which the playback order is adjacent, and creating a music transition model;
Based on the created music transition model and user input information, a process of selecting and listing a plurality of music data from the database;
A playlist creation program characterized in that a series of processes including the process of displaying the listed information is converted into a command that can be read and executed by a computer. A playlist creation program for creating a playlist that defines the playback order of music data,
A process of reading music data from the database and extracting a plurality of types of feature values from the read music data;
Based on the extracted feature amount, create a plurality of clusters composed of at least one piece of music data, calculate a transition probability when transitioning from any one cluster to another cluster, Process to create,
Based on the created cluster transition model and user input information, a process of selecting and listing a plurality of music data from the database;
A playlist creation program characterized in that a series of processes including the process of displaying the listed information is converted into a command that can be read and executed by a computer.

Images