JP6672707B2 - Playlist generation method, playlist generation device, program, and list generation method - Google Patents

Description

  The present invention relates to a playlist generation method, a playlist generation device, a program, and a playlist generation method.

  In recent years, types of electronic devices connectable to a communication network have been increasing, and services using the communication network have been provided to various electronic devices. In particular, as one of services using a communication network such as the Internet, music distribution services for various electronic devices have been widely spread.

  As one of the music distribution services, a user (operator) of an electronic device specifies a keyword representing an image of a song to be heard by voice, and the distribution server generates a playlist in which songs matching the image are extracted. Services for delivering to electronic devices are known. As a method of generating a playlist, there is a method in which a feature amount representing an image is set for each of the playlist and the song, and a song having a feature amount having a high degree of matching with the feature amount of the playlist is registered in the playlist. It is known (see, for example, Patent Document 1).

JP-A-2003-132850

  When creating a playlist of music data based on the feature amount of an image, a song having a feature amount within a predetermined range centered on the feature amount of the playlist in the image feature space is registered in the playlist. Therefore, when there are playlists having similar image feature amounts, there may be a plurality of playlists in which a certain song is to be registered.

  When a playlist is generated, a keyword indicating a condition (attribute) different from the image of the song is specified by voice, thereby narrowing down the songs to be registered in the playlist and generating the playlist. It is also possible. Therefore, when the playlist is generated by narrowing down the songs, the number of songs registered in the playlist may be small and the feature amount of the image may be biased.

  As described above, when the number of songs registered in the playlist is small and the feature amount of the image is biased, a discrepancy occurs between the image of the playlist and the image held by the user only from the registered songs, and a sense of incongruity occurs. May be embraced. The same applies to a case where a reproduction list of data that can be classified based on a feature amount representing an image, such as document data for reading and reproducing a novel or the like, image data, etc., based on an image, is not limited to a song (music). I can say that.

  In one aspect, an object of the present invention is to reduce a sense of discomfort given to a user due to a discrepancy between an image of a presented playlist and an image of a song included in the playlist.

In one embodiment, the computer executes the following processing. First, the computer narrows down songs based on attribute information indicating attributes of songs to be registered in the playlist included in the playlist generation request from the user. Then, the computer, the feature of the song for the feature value representing the image of the song narrowed, and registered in a range that has been correlated to each play list on the database to a playlist is distributed Based on the range of the play list defining the range , a play list in which the tunes whose features representing the images are included in the range is generated as a recommended play list candidate . Then, the feature quantity distribution of the songs registered in the generated recommendation playlist candidates, biased against the corresponding range is assumed about the requested playlist is a playlist generation is requested by the generation request If you are, based on the corresponding distributions for the song to be included in the recommendation playlist candidate, and the range assumed for recommended playlist candidate, and the range assumed with the requirements playlist recommendation playlist From the candidates , a recommended playlist that is a playlist to be presented to the user is determined .

  According to the above aspect, it is possible to reduce the discomfort given to the user due to the discrepancy between the image of the presented playlist and the image of the music included in the playlist.

FIG. 2 is a diagram illustrating a functional configuration of a playlist generation device according to the first embodiment. The figure which shows the example of a structure of a priority calculation part and a narration preparation part It is a figure showing an example of a keyword DB. It is a figure showing an example of a narrowing down attribute DB. It is a figure showing the example of music DB. It is a figure showing an example of a play list DB. FIG. 6 is a diagram illustrating an example of an image feature amount. FIG. 3 is a diagram illustrating an image feature space. 9 is a flowchart (part 1) illustrating a playlist generation process according to the first embodiment. 10 is a flowchart (part 2) illustrating a playlist generation process according to the first embodiment. FIG. 9 is a diagram for describing an example of a playlist change method. It is a flowchart explaining the content of the process of extracting a search keyword and a refinement attribute. It is a flowchart explaining the content of the process which calculates recommendation priority. FIG. 7 is a diagram illustrating a required shift amount. FIG. 4 is a diagram illustrating a recognition shift amount. It is a figure showing another example of the range of a play list. FIG. 4 is a diagram illustrating a relationship between a range of a playlist and a probability density distribution. It is a figure showing the functional composition of the play list generation device concerning a 2nd embodiment. It is a figure showing the example of composition of an adaptive processing part. 9 is a flowchart illustrating a process performed by an adaptive processing unit. FIG. 9 is a diagram for describing an example of a method of modifying a range of a playlist. FIG. 9 is a diagram illustrating an example of a method of expanding a range of a playlist. FIG. 9 is a diagram for explaining one of effects obtained by expanding the range of the playlist. It is a figure showing the composition of the play list reproduction system concerning a 3rd embodiment. It is a figure showing the functional composition of the server in the play list reproduction system concerning a 3rd embodiment. It is a figure showing the composition of the play list reproduction system concerning a 4th embodiment. It is a figure showing the functional composition of the server in the playlist reproduction system concerning a 4th embodiment. It is a figure showing the composition of the play list reproduction system concerning a 5th embodiment. FIG. 2 is a diagram illustrating a hardware configuration of a computer.

[First Embodiment]
FIG. 1 is a diagram illustrating a functional configuration of the playlist generation device according to the first embodiment.

  As shown in FIG. 1, the playlist generation device 1 according to the present embodiment includes a required playlist input unit 101, a keyword / attribute extraction unit 102, and a music narrowing unit 103. Further, the playlist generation device 1 includes a recommended playlist candidate generation unit 104, a priority calculation unit 105, and a playlist determination unit 106. Further, the playlist generation device 1 includes a narration creation unit 107, a playlist reproduction unit 108, and an audio output unit 109. Further, the playlist generation device 1 includes a keyword DB 121, a narrow-down attribute DB 122, a song DB 123, and a playlist DB 124.

  The request playlist input unit 101 receives an input of a playlist generation request from a user by voice or text. When the input of the generation request is received by voice, the requested playlist input unit 101 includes a voice input unit such as a microphone and a voice recognition unit. When accepting an input by voice, the request playlist input unit 101 converts a surrounding sound picked up by the microphone into a voice signal, converts the voice into a text by voice recognition, and outputs the text. When accepting text input, the request playlist input unit 101 detects a character input from a keyboard (including a software keyboard) and outputs the text.

  The keyword / attribute extraction unit 102 extracts a search keyword from the keyword DB 121 based on the text obtained by the voice recognition unit, and extracts a song narrowing attribute from the narrowing attribute DB 122. The keyword DB 121 is a database in which a plurality of correspondences between texts and search keywords are registered. The search keyword is a keyword used for searching for a playlist. The narrowing down attribute DB 122 is a database in which a plurality of correspondences between texts and song narrowing down attributes are registered. The song narrowing down attribute is attribute information used to narrow down songs to be registered in the playlist.

  The song narrowing unit 103 narrows down the candidate songs to be registered in the playlist to songs having attributes matching the narrowing attribute from all the songs registered in the song DB 123 based on the extracted narrowing attributes. . The song DB 123 is a database in which a plurality of correspondences between song audio data, song attributes, and image feature amounts of the song are registered. The song attribute is attribute information used to narrow down song audio data. The music image feature amount is information obtained by digitizing the music image.

  The recommended playlist candidate generation unit 104 extracts a playlist including the songs narrowed down by the song narrowing unit 103 from the playlist DB 124 and creates a recommended playlist candidate. The recommended playlist candidate is a playlist candidate that is recommended (presented) to a user (operator) of the playlist generation device 1. The recommended playlist candidate includes a playlist name, a song ID, and an assumed distribution. The playlist DB 124 is a database in which playlist names, assumed distributions, and playlist ranges are registered in association with each playlist. The assumed distribution of the playlist is the probability density distribution of the playlist in the image feature space used when expressing the distribution of the song based on the image feature amount of the song.For example, when the probability density distribution is a normal distribution, the average (Center) and variance. The range of the playlist is a range that defines the feature amount of a music piece to be registered in the playlist in the image feature space. That is, the recommended playlist candidate generating unit 104 extracts a playlist including the songs narrowed down by the song narrowing unit 103 within the range of the playlist from the playlists registered in the playlist DB 124, and Create

  The priority calculation unit 105 calculates the priority of each recommended playlist candidate when the recommended playlist candidate generation unit 104 creates a plurality of recommended playlist candidates. The priority calculation unit 105 calculates the priority of each recommended playlist candidate based on the assumed distribution and range of the recommended playlist candidate and the search keyword.

  The playlist determination unit 106 determines a playlist to be adopted from among a plurality of recommended playlist candidates, that is, a playlist to be presented to the user, based on the priority calculated by the priority calculation unit 105.

  The narration creating unit 107 creates narration audio data representing an image or the like of the playlist determined by the playlist determining unit 106.

  The playlist playback unit 108 plays back the narration audio data created by the narration creation unit 107 and plays back the song audio data included in the playlist determined by the playlist determination unit 106.

  The audio output unit 109 is, for example, a speaker, and outputs the narration audio data and the music audio data output by the playlist reproduction unit 108 as music audio to a space outside the playlist generation device 1.

  As described above, the playlist generation device 1 according to the present embodiment also functions as a playlist reproduction device that reproduces and outputs the generated playlist.

FIG. 2 is a diagram illustrating a configuration example of a priority calculation unit and a narration creation unit.
As illustrated in FIG. 2, the priority calculation unit 105 includes a recognition shift amount calculation unit 105a, a request shift amount calculation unit 105b, a playlist search unit 105c, and a shift amount integration unit 105d.

  The recognition shift amount calculation unit 105a calculates the distribution of the feature amount of the song in the playlist obtained from the image feature amount of the song corresponding to the song ID included in the recommended playlist candidate, and the song assumed by the recommended playlist candidate. The difference (recognition shift amount) from the distribution of the feature amount is calculated. Here, the distribution of the feature amount of the music assumed as the recommended playlist candidate (playlist) is, for example, the case where a playlist is created with all the music registered in the music DB 123 as a registration target. This is the distribution of the feature amounts of all the songs included in the list. In the following description, the distribution of the feature amount of a music piece is also referred to as “music piece distribution” or simply “distribution”.

  The request deviation amount calculation unit 105b calculates the difference between the distribution of music assumed as a recommended playlist candidate and the distribution of music assumed in a playlist corresponding to the search keyword extracted by the keyword / attribute extraction unit 102 ( Required deviation). The request deviation amount calculation unit 105b acquires a playlist corresponding to the search keyword from the playlist search unit 105c. In the following description, a playlist corresponding to a search keyword, that is, a playlist requested by a user is also referred to as a “requested playlist”.

  The playlist search unit 105c searches the playlist DB 124 using the search keyword extracted by the keyword / attribute extraction unit 102 as a key, and acquires a playlist (requested playlist) that matches the search keyword.

  The deviation amount integration unit 105d calculates the priority of each recommended playlist candidate based on the recognition deviation amount and the required deviation amount for each recommended playlist candidate.

  Further, as shown in FIG. 2, the narration creating unit 107 includes a narration text creating unit 107a and a narration voice creating unit 107b. The narration sentence creation unit 107a creates a sentence (text) including information on the determined playlist based on the search keyword and the playlist determined by the playlist determination unit 106. The narration sound creation unit 107b creates sound data including information on the determined playlist based on the sentence created by the narration sentence creation unit 107a.

  FIG. 3 is a diagram illustrating an example of the keyword DB. The keyword DB 121 is a database in which a plurality of correspondences between texts and search keywords are registered as described above. Each of the text and the search keyword in the keyword DB 121 is a character string representing an image of a playlist (song), as shown in FIG. Further, as shown in FIG. 3, the keyword DB 121 can register a plurality of texts in association with one search keyword.

  FIG. 4 is a diagram illustrating an example of the narrow-down attribute DB. As described above, the narrowing down attribute DB 122 is a database in which a plurality of correspondences between texts and song narrowing down attributes are registered. The song narrowing-down attribute is classification information of a song different from an image, and includes an artist name, gender, age, and the like as shown in FIG. For example, when the text includes a character string representing “male”, the keyword / attribute extracting unit 102 extracts attribute information “sex: male” as one of the narrowing-down attributes, and sends the attribute information to the song narrowing-down unit 103. Output. In this case, the song narrowing unit 103 narrows down the songs to be registered in the playlist from all the songs registered in the song DB 123 to songs of a male artist.

  If the text includes a character string representing the artist name and a character string representing the date of publication, the attribute information associated with those character strings is extracted and notified to the song narrowing unit 103. .

  FIG. 5 is a diagram illustrating an example of the music DB. The song DB 123 is a database in which a plurality of correspondences between song audio data, song attributes, and song image feature amounts are registered as described above. In the song DB 123, a plurality of song audio data is identified by the song ID. One song ID is associated with one piece of song audio data, a song attribute of the song audio data, and an image feature amount of the song audio data.

  The song attribute of the song DB 123 is classification information of a song different from the image, and includes information corresponding to the narrowing down attribute in the narrowing down attribute DB 122. In the song DB 123 shown in FIG. 5, the song attributes include the title, artist name, gender, and age.

  The song image feature amount in the song DB 123 is information obtained by digitizing the song image. The image feature amount of the music is, for example, information obtained by quantifying the extent to which one or more images are applied to one music by multi-stage evaluation.

  FIG. 6 is a diagram illustrating an example of the playlist DB. As shown in FIG. 6, the playlist DB 124 is a database in which a playlist name, an assumed distribution, and a range of the playlist are registered in association with each playlist. As described above, the assumed distribution of the playlist is the average (center) and variance of the playlist in the image feature space when expressing the distribution of the song based on the image feature amount of the song. Further, the range of the playlist is a range that defines the image feature amount of the music to be registered in the playlist in the image feature space. When the range of the playlist is a circle or a sphere (super sphere), the range of the playlist in the playlist DB 124 is designated by a radius from the center as shown in FIG.

  FIG. 7 is a diagram illustrating an example of the image feature amount. FIG. 8 is a diagram illustrating an image feature space.

  As shown in FIG. 7, the image feature amount of a music piece is represented by a value obtained by digitizing the degree of one or more images. By using this image feature amount, each song registered in the song DB 123 is represented by one point in the image feature space. The range of each playlist is specified by a circle or a sphere (super sphere) in the image feature space, for example. Then, when a certain playlist is designated, the songs included in the range of the playlist in the image feature space are the songs to be registered in the playlist.

  For example, when there are two image feature amounts, a first image (image H) and a second image (image K), the image feature space is a two-dimensional space (that is, a plane) as shown in FIG. It is represented by

  In the image feature space shown in FIG. 8, the horizontal axis is the axis representing the feature amount of the image H, and the vertical axis is the axis representing the feature amount of the image K. Further, in the image feature space shown in FIG. 8, the feature amount of each image is shown in a range from -1.0 to +1.0. In this image feature space, for example, the horizontal axis (image H) is an image of "pleasant / unpleasant", and the vertical axis (image K) is an image of "excitation / sedation". At this time, the abscissa indicates, for example, +1.0 as “pleasant” and −1.0 as “displeased”, so that a song having a stronger pleasant image is mapped to the right of the image feature space. On the vertical axis, for example, +1.0 is set to "excitation" and -1.0 is set to "sedation", so that a song having a stronger excitement image is mapped to an upper portion of the image feature space. In the image feature space, for example, the horizontal axis (image H) may be an image of "joy / sorrow" and the vertical axis (image K) may be an image of "anger / ease".

  All the songs registered in the song DB 123 can be represented by one point (one black circle) in the image feature space in FIG. For example, a song in which the feature amount of the image H is 0.2 and the feature amount of the image K is 0.7 is represented by a point 201. Also, a tune in which the feature amount of the image H is -0.4 and the feature amount of the image K is 0.2 is represented by a point 202.

  The range of the playlist is represented by a circle centered on one point in the image feature space. Therefore, as shown in FIG. 8, when a range 301 of a certain playlist is represented by a circle and a song narrowed down based on the narrowing down attribute is represented by a black circle, an area surrounded by the circumference of the range 301 Are to be registered in the play list. The songs to be registered in the playlist represented by the range 302 are three songs corresponding to the three points included in the area surrounded by the circumference of the range 302. When the text obtained by the voice recognition unit includes a song narrowing attribute together with the image of the playlist, the registration target for the playlist is a song having a feature amount within the range of the playlist and matching the narrowing attribute. Only songs with attributes are registered. For example, if the text includes a narrowing attribute together with a playlist image in which the range of the feature amount is represented by a circle in the range 301, only songs having a song attribute that matches the narrowing attribute among the five songs in the range 301 are included in the text. It becomes a registration target for the playlist.

  FIG. 9A is a flowchart (part 1) illustrating a playlist creation process according to the first embodiment. FIG. 9B is a flowchart (part 2) illustrating the playlist creation process according to the first embodiment.

  The playlist creation device 1 of the present embodiment performs a playlist creation process including a playlist playback process as shown in FIGS. 9A and 9B. 9A and 9B show a playlist creation process in the case of receiving an input of a playlist creation request by voice.

  In the playlist creation process, the playlist creation device 1 first performs a process of acquiring an audio signal and converting it into text (step S1). The process of step S1 is performed by the microphone and the voice recognition unit of the request playlist input unit 101. The microphone collects sound propagating around the playlist creation device 1, converts the sound into a sound signal, and outputs the sound signal to the sound recognition unit. The voice recognition unit performs a known voice recognition process on the voice signal. For example, the voice recognition unit first determines whether a significant voice is included in the voice signal. Then, when a significant voice is included in the voice signal, the voice recognition unit converts the significant voice portion into text.

  After the above-described speech recognition processing, the request playlist input unit 101 determines whether or not a text has been obtained (step S2). If the text has not been obtained (Step S2; No), the request playlist input unit 101 repeats the process of Step S1 until the text is obtained.

  On the other hand, when the text is obtained (Step S2; Yes), the requested playlist input unit 101 outputs the obtained text to the keyword / attribute extraction unit 102 as a playlist generation request from the user. Upon receiving the text from the request playlist input unit 101, the keyword / attribute extraction unit 102 performs a process of extracting a search keyword and a narrowing attribute based on the received text (step S3). In step S3, the keyword / attribute extraction unit 102 first searches the keyword DB 121 shown in FIG. 3 to check for a text that matches the received text. Then, when a text that matches the received text is registered in the keyword DB 121, the keyword / attribute extraction unit 102 extracts a search keyword associated with the matched text from the keyword DB 121. The keyword / attribute extraction unit 102 outputs the extracted search keywords to the priority calculation unit 105 and the narration creation unit 107.

  Also, in step S3, the keyword / attribute extraction unit 102 searches the narrowed-down attribute DB 122 shown in FIG. 4 and checks whether there is a narrowed-down attribute that matches the received text. Then, when a text that matches the received text is registered in the narrow-down attribute DB 122, the keyword / attribute extraction unit 102 extracts a narrow-down attribute associated with the matched text from the narrow-down DB 122. The keyword / attribute extraction unit 102 outputs the extracted refinement attributes to the music refinement unit 103.

  When the processing of step S3 is completed, the playlist generation device 1 then narrows down the song IDs of the songs to be registered in the playlist to song IDs having song attributes matching the narrowing attributes from all song IDs. Processing is performed (step S4). The process of step S4 is performed by the music narrowing unit 103. Upon receiving the narrowing down attribute from the keyword / attribute extracting unit 102, the song narrowing down unit 103 extracts a song ID having a song attribute matching the narrowing down attribute by referring to the song DB 123, and narrows down the song ID. The song narrowing unit 103 outputs the narrowed song ID and the image feature amount of the song associated with the song ID to the recommended playlist candidate generation unit 104.

  Upon receiving the song ID and the image feature amount from the song narrowing unit 103, the recommended playlist candidate generation unit 104 generates a recommended playlist candidate based on the song ID and the image feature amount and the playlist DB 124 ( Step S5). In the process of step S5, the recommended playlist candidate generation unit 104 compares the range of each playlist registered in the playlist DB 124 with the image feature amount, and determines a playlist including the image feature amount in the range. Extract. That is, the recommended playlist candidate generation unit 104 indicates the image feature amount of each song and the range of the playlist in the image feature space as illustrated in FIG. A playlist including points is set as a recommended playlist candidate. The recommended playlist candidate generation unit 104 outputs the generated recommended playlist candidates to the priority calculation unit 105.

  Although not shown, when the number of generated recommended playlist candidates is one, the recommended playlist candidate generating unit 104 outputs the generated recommended playlist candidates to the playlist determining unit 106.

  Upon receiving the search keyword from the keyword / attribute extraction unit 102 and receiving the recommended playlist candidate from the recommended playlist candidate unit 105, the priority calculation unit 105 calculates a recommendation priority for the recommended playlist candidate (step S6). ). In the process of step S6, the priority calculation unit 105 calculates the recognition error amount by the recognition error amount calculation unit 105a, and calculates the required error amount by the required error amount calculation unit 105b. Then, the priority calculation unit 105 calculates the recommendation priority of each recommended playlist candidate by integrating the recognition shift amount and the requested shift amount by the shift amount integrating unit 105d. The priority calculation unit 105 outputs the calculated recommendation priority to the playlist determination unit 106.

  Upon receiving the recommendation priority calculated by the priority calculation unit 105, the playlist determination unit 106 determines a playlist to be recommended based on the recommendation priority (Step S7). In the processing of step S7, the playlist determination unit 106 determines the playlist having the highest recommendation priority as the recommended playlist. At this time, the play list determination unit 106 may change only the name of the play list, or may change the play list itself to a recommended play list and change the tune to be registered. The playlist determination unit 106 outputs the name of the determined recommended playlist to the narration creation unit 107 (narration text creation unit 107a), and outputs the song ID included in the recommended playlist to the playlist playback unit 108. If the recommended playlist candidate includes the required playlist and the recommended priority of the required playlist is the highest, the recommended playlist is the required playlist (that is, the playlist requested by the user).

  Next, as shown in FIG. 9B, the playlist generation device 1 creates a narration voice for the recommended playlist (step S8). The process of step S8 is performed by the narration creation unit 107. The narration creation unit 107 first creates a narration sentence in the narration sentence creation unit 107a. The narration sentence creating unit 107a determines whether or not the name of the recommended playlist matches the previously acquired search keyword (or the name of the playlist specified by the search keyword). Based on the determination result, the narration text creating unit 107a creates a narration text to be notified to the user. If the name of the recommended playlist matches the search keyword, the narration sentence creation unit 107a creates a text notifying that the playlist of the search keyword is to be played. On the other hand, if the name of the recommended playlist does not match the search keyword, the narration sentence creating unit 107a creates a text notifying that the recommended playlist is to be reproduced because, for example, the playlist of the search keyword could not be created. I do. The narration text creation unit 107a outputs the text of the created narration text to the narration voice creation unit 107b. The narration speech creation unit 107b creates narration speech data from the text of the received narration sentence using a known speech synthesis technique. The narration sound creation unit 107b outputs the created narration sound data to the playlist reproduction unit 108. Note that the narration text created by the narration text creation unit 107a may be displayed on a display device such as a liquid crystal display without being converted into voice data.

  Thereafter, in the playlist generation device 1, the playlist playback unit 108 that has received the song ID and the narration voice included in the recommended playlist plays back the narration voice data and the song voice data with the song ID included in the recommended playlist ( Step S9). In the process of step S9, the playlist reproduction unit 108 first reproduces narration audio data. Thereafter, the play list reproducing unit 108 sequentially reproduces the music sound data of the music IDs included in the recommended play list in a predetermined order. The voice signal of the narration voice and the voice signal of the tune voice obtained by the playback of the playlist playback unit 108 are output as audio to the outside of the playlist generation device 1 by the audio output unit (speaker) 109.

FIG. 10 is a diagram illustrating an example of a method of changing a playlist.
The inclusion relationship between the recommended playlist and the requested playlist determined in step S7 is, for example, the relationship shown in (a) to (c) of FIG. In the example of FIG. 10 (a), the range 312 recommendation playlist _{L 2} is encapsulated in the range 311 requests the playlist _{L 1.} In such cases, the recommendation songs included in the playlist L ₂ of the range 312 shown by black circles are included within the scope 311 of all requests playlist L1. Therefore, if the range 312 recommendation playlist L ₂ is encapsulated in the range 311 requests the playlist L _1, playlist determination unit 106 may be changed only names of the playlist to be presented to the user , it may change the play list itself on the recommendation playlist L _2.

On the other hand, as shown in (b) of FIG. 10, when only a part of the recommended playlist L ₂ in the range 312 overlaps the range 311 requests the playlist L _1, only songs overlap region indicated by black circles It is also possible to create a playlist in which is registered. However, in this case, because there is a bias in the distribution of music, which is registered in the play list is also within the scope 312 of the recommendation playlist L _2, when changing only the name of the playlist, the playback of the recommended playlist L ₂ Image May be inconsistent with the image of the song. Therefore, when only a part of the recommended playlist L ₂ in the range 312 overlaps the range 311 requests the playlist L _1, the play list itself, as shown in (c) of FIG. 10, the range of the recommended playlist L ₂ All the songs included in 312 may be changed to a registered playlist. In the example shown in (c) of FIG. 10, a small degree of deviation of the songs in the range 312 recommendation playlist L _2. For this reason, and eliminates the discrepancy between the image of the song to be played back with the recommendation playlist L ₂ image, it is possible to reduce the uncomfortable feeling to the user.

  As described above, in the playlist generation processing according to the present embodiment, a playlist candidate to be recommended to the user is generated based on the song narrowing down attribute requested by the user (operator), and the playlist is generated in response to the request from the user. And compare it with the requested playlist generated. Then, when a playlist having a higher priority than the request playlist exists among the candidates for the playlist to be recommended, the name of the playlist having a higher priority is presented to the user. As a result, the user can listen to the tune to be reproduced after grasping the name of the playlist. Also, the image of the playlist with a higher priority is closer to the image of the music to be reproduced than the image of the playlist requested by the user. Therefore, according to the playlist generation processing according to the present embodiment, it is possible to reduce a sense of discomfort given to the user due to a discrepancy between the image of the playlist and the image of the music reproduced according to the playlist. Further, by changing not only the name of the playlist but also the playlist itself, as shown in FIG. It is possible to reduce the discomfort given to the person. Also, in the playlist generation process according to the present embodiment, in order to present to the user the name of the playlist closer to the image of the song to be reproduced, the user must perform trial and error on the keyword for reproducing the song that matches the image. It is possible to save the trouble of performing.

  Further, according to the playlist generation processing according to the present embodiment, the user can know a more appropriate name for the playlist name requested by the user. Therefore, the user can easily grasp the keyword for obtaining the playlist suitable for the image held by the user. For example, when a playlist of “cool songs” with a similar image is presented in response to a request for a playlist of “bright songs”, the user can select a song to be played when the playlist of “cool songs” is requested. You can know the image of. For this reason, the user can know which of the playlists “bright song” and “cool song” is played, and which song is closer to the image desired by the user. Therefore, it becomes easy to request generation of a playlist closer to the image desired by the user from the next time.

  FIG. 11 is a flowchart illustrating the content of the process of extracting the search keyword and the narrowing down attribute.

  The keyword / attribute extraction unit 102 performs the process (step S3) of extracting a search keyword and a narrowing attribute in the playlist generation process according to the present embodiment. As shown in FIG. 11, the keyword / attribute extraction unit 102 first obtains a text from the request playlist input unit 101 (step S301).

  Next, the keyword / attribute extraction unit 102 searches the keyword DB 121 using the obtained text as key information, and extracts a search keyword (step S302). As shown in FIG. 3, a plurality of correspondences between texts and search keywords are registered in the keyword DB 121. In the process of step S302, the keyword / attribute extraction unit 102 checks whether or not the acquired text is registered in the keyword DB 121. At this time, for example, the keyword / attribute extraction unit 102 performs a regular expression search using the acquired keyword for each keyword registered in the keyword DB 121, and checks whether there is a keyword that matches the acquired keyword. If the acquired text is registered in the keyword DB 121, a keyword associated with the text is extracted and used as a search keyword. When a plurality of texts are obtained from the voice recognition unit, the keyword / attribute extraction unit 102 checks whether or not each of the obtained texts is registered in the keyword DB 121.

  Next, the keyword / attribute extraction unit 102 searches the narrowing-down attribute DB 122 using the obtained text as key information, and extracts the narrowing-down attribute of the music (step S303). As shown in FIG. 4, a plurality of correspondences between texts and song narrowing-down attributes are registered in the narrowing-down attribute DB 122. In the process of step S303, the keyword / attribute extraction unit 102 checks whether or not the acquired text is registered in the narrow-down attribute DB 122. At this time, for example, the keyword / attribute extraction unit 102 performs a regular expression search using the acquired keyword for each keyword registered in the narrowed-down attribute DB 122, and checks whether there is a keyword that matches the acquired keyword. Then, when the acquired text is registered in the narrow-down attribute DB 122, a song attribute associated with the text is extracted. When a plurality of texts are obtained from the voice recognition unit, the keyword / attribute extraction unit 102 checks whether or not each of the obtained texts is registered in the narrow-down attribute DB 122.

FIG. 12 is a flowchart illustrating the content of the process of calculating the recommendation priority.
The process of calculating the recommendation priority in the playlist generation process according to the present embodiment (step S6) is performed by the priority calculation unit 105. As shown in FIG. 12, the priority calculation unit 105 first obtains a search keyword and a recommended playlist candidate (step S601). The priority calculation unit 105 acquires a search keyword from the keyword / attribute extraction unit 102, and acquires a recommended playlist candidate from the recommended playlist candidate generation unit 104. The acquired search keyword is input to the playlist search unit 105c. The acquired recommended playlist candidate is input to the recognition shift amount calculation unit 105a and the request shift amount calculation unit 105b.

  Next, the priority calculation unit 105 performs a process of calculating the recognition shift amount (Step S602) and a process of calculating the required shift amount (Steps S603 and S604).

The process of calculating the recognition shift amount (step S602) is performed by the recognition shift amount calculation unit 105a. The recognition shift amount calculation unit 105a calculates, as the recognition shift amount, the shift amount between the normal distribution based on the image feature amount corresponding to the song ID in the recommended playlist candidate and the normal distribution of the song assumed by the recommended playlist candidate. Is calculated. For example, the recognition shift amount calculation unit 105a calculates the normal distribution P _i based on the image feature amount of the song corresponding to the song ID in the i-th recommended playlist candidate and the normal distribution P _i of the song assumed by the recommended playlist candidate. calculating a Q _i and the KL divergence _{D KL (P i || Q i} ). KL divergence is a measure for measuring the difference between two distributions, and KL divergence D _KL (P || Q) for two normal distributions P and Q is represented by the following equation (1).

Μ _P and Σ _P in Equation (1) are the mean and variance of the d-dimensional normal distribution P, respectively. Further, μ _Q and Σ _Q in the equation (1) are the average and the variance of the d-dimensional normal distribution Q, respectively.

When calculating the KL divergence D _KL (P _i || Q _i ) as the recognition shift amount, the recognition shift amount calculation unit 105a uses the song DB 123 based on the song ID in the i-th recommended playlist candidate. , The image feature amount of the song associated with each song ID is obtained.

  The recognition shift amount calculation unit 105a outputs the calculated recognition shift amount to the shift amount integration unit 105d.

  On the other hand, the process of calculating the required shift amount (steps S603 and S604) is performed by the playlist search unit 105c and the requested shift amount calculation unit 105b.

  The playlist search unit 105c searches the playlist DB 124 and extracts a playlist corresponding to the acquired search keyword (step S603). When the playlist search unit 105c extracts the playlist corresponding to the search keyword, the request deviation amount calculation unit 105b, at this time, the playlist search unit 105c, for example, searches for the name registered in the playlist DB 124. A regular expression search is performed using the obtained search keyword, and an assumed distribution and range of a playlist whose name matches the search keyword are extracted. The playlist search unit 105c outputs the name (search keyword), the assumed distribution, and the range of the extracted playlist to the request deviation amount calculation unit 105b.

The request deviation amount calculation unit 105b calculates, as the request deviation amount, a deviation between the normal distribution assumed for the recommended playlist candidate and the normal distribution assumed for the playlist received from the playlist search unit 105c (assumed distribution). Calculate the amount. For example, the request deviation amount calculation unit 105b calculates the KL divergence of the normal distribution P _{i of the} music assumed in the i-th recommended playlist candidate and the normal distribution R of the music assumed in the playlist based on the search keyword. _DKL (P _i || R) is calculated. The KL divergence D _KL (P _i || R) for the normal distribution P _i , R is calculated using equation (1).

  The request deviation amount calculation unit 105b outputs the calculated request deviation amount to the deviation amount integration unit 105d.

After calculating the recognition shift amount and the required shift amount in this way, in the priority calculating unit 105, the shift amount integrating unit 105d calculates the recommendation priority based on the recognition shift amount and the required shift amount (step S605). In step S605, the deviation amount integration unit 105d integrates the recognition deviation amount and the required deviation amount, and calculates a recommendation priority represented by a weighted sum of the two values. Shift amount integration unit 105d calculates, for example, the recommendation priority PP _i of the i-th recommendation playlist candidate by the following equation (2).

PP _i = −w1 · D _KL (P _i || Q _i ) −w 2 · D _KL (P _i || R) (2)

  W1 and w2 in Expression (2) are weighting values for the respective shift amounts (KL divergence). The combination of the weight values w1 and w2 is arbitrary. However, there is a possibility that usability will deteriorate if the user's request is not met. For example, w1 is reduced and w2 is increased, for example, w1 = 0.4, w2 = 0.6. When it is desired to suppress a large change in the operation before and after the application of the playlist generation processing of the present embodiment, for example, the value of w1 is further reduced, such as w1 = 0.1, w2 = 0.9, and the value of w2. May be increased. Further, the combination of the weight values w1 and w2 may be changed according to, for example, the elapsed time (number of days) from the previous playlist generation processing. Further, the combination of the weight values w1 and w2 may be appropriately adjusted by the user himself.

  Here, the required shift amount and the recognition shift amount will be described again with reference to FIGS. 13A and 13B. FIG. 13A is a diagram illustrating the required shift amount. FIG. 13B is a diagram illustrating the recognition shift amount.

  As described above, the required shift amount is a shift amount between the normal distribution of the music assumed as the recommended playlist candidate and the normal distribution of the music assumed in the playlist specified by the search keyword. The normal distribution of the music assumed by the playlist is the assumed distribution in the playlist DB 124.

For example, as shown in (a) of FIG. 13A, there are five recommended playlist candidates, and five playlists L ₁ , L ₂ , L ₃ , L ₄ , and L ₄ in the image feature space (HK plane). and the _{L 5} ranges 311, 312, 313, and 315 are represented by circles, respectively. It is also assumed that the positions of the songs (211, 212, 213,...) Narrowed down by the narrowing down attribute in the image feature space are represented by the positions of the black circles shown in FIG. Furthermore, five play lists _{L 1, L 2, L 3} , L 4, and the play list _{L 1} of _{which L 5} and were playlist identified (requested playlist) by the search keyword. In this case, the required shift amount calculating unit 105b for each playlist _L i of (i = 1,2,3,4,5), a normal distribution _{P i} which assumes the playlist _{L i,} playlist The KL divergence D _KL (P _i || R) with the assumed normal distribution R of L ₁ is calculated.

The assumed normal distribution (assumed distribution) P _i of the five playlists L _i ( _i = 1, 2, 3, 4, 5) shown in (a) of FIG. b).

The KL divergence D _KL (P _i || R) is a measure for measuring the difference between the two normal distributions P _i , R as described above. When P _i = R, D _KL (P _i || R) = It becomes 0. The KL divergence D _KL (P _i || R) increases as the difference between the two normal distributions P _i , R increases. Therefore, in the example shown in FIG. 13A (b), the KL divergence D _KL (P ₁ || R) for the playlist L ₁ , which is also the requested playlist, becomes 0, and the position farthest from the playlist L ₁ KL divergence D _KL (P ₅ || R) for the playlist L _{5 in} the first item has the largest value.

  In other words, a song included in a playlist with a small required deviation amount is located at a position within or close to the range of the playlist (requested playlist) specified by the search keyword in the image feature space. For this reason, it is preferable that the recommendation priority at the time of determining the recommended playlist from the recommended playlist candidates be such that the playlist with the smaller request deviation amount has the higher recommended priority.

However, in the example shown in (a) of FIG. 13A, the song to be included within the scope 311 of the play list _{L 1} is only tracks are also included within the scope 312 of the play list _{L 2,} and the play list _{L 1} The bias in the range 311 is large. Therefore, if a playlist to be recommended is determined by calculating the recommendation priority based on only the requested deviation amount, there is a discrepancy between the image of the playlist requested by the user and the image of the song played according to the determined playlist. This may cause the user to feel uncomfortable. Therefore, in the playlist generation processing according to the present embodiment, the recognition shift amount is calculated together with the request shift amount, and the recommendation priority obtained by integrating these is calculated. As described above, the recognition shift amount calculates the shift amount between the normal distribution based on the image feature amount of the song corresponding to the song ID in the recommended playlist candidate and the assumed normal distribution of the recommended playlist candidate. .

The normal distribution based on the image feature amount of the song corresponding to the song ID in the recommended playlist candidate is the normal distribution (average and average) calculated based only on the image feature amount of the songs included in the range of the playlist among the narrowed-down songs. (Dispersion). For example, it contains three pieces of music 211, music 212, and tune 213 to a playlist _{L 1} in the range 311 shown in (a) of FIG. 13A. Therefore, the normal distribution P ₁ based on the image feature amount of music in i = 1 playlist (recommended playlist candidate) L ₁ is calculated on the basis of the image feature amount of the three songs. On the other hand, the normal distribution _{Q 1} which assumes playlist _{L 1} of i = 1 is a registered envisaged distributed playlist DB 124. Therefore, when the two normal distributions P ₁ and Q ₁ for the playlist L ₁ are expressed in the image feature space, the two normal distributions P ₁ and Q ₁ have a relationship as shown in FIG. expressed.

Similarly, when the two normal distributions P _i and Q _i are expressed in the image feature space for the other four playlists L _i (i = 2, 3, 4, 5), (b) in FIG. To (e).

The relationship (difference) between the two normal distributions P _i and Q _i in each playlist L _i (i = 1, 2, 3, 4, 5) is quantified by KL divergence D _KL (P _i || Q _i ). can do. When the difference between the two normal distributions P _i and Q _i is small, the KL divergence D _KL (P _i || Q _i ) has a small value close to zero. On the other hand, when the difference between the two normal distributions P _i and Q _i is large, KL divergence D _KL (P _i || Q _i ) has a large value. That, KL divergence _{D KL (P i || Q i} ) for a playlist _{L 1} shown in FIG. 13 (a), the KL-divergence _{D KL (P i || Q i} ) for other playlist This is a large value. Therefore, if you play a song 211, 212 and 213 included in the scope 311 of the play list _{L 1} as songs in a playlist _{L 1,} the user, the image of the reproduced music does not match the image playlist _{L 1} It is highly likely that you will feel uncomfortable.

3 songs included in the playlist _{L 1} (request playlist) shown in (a) of FIG. 13B it is all included in the playlist _{L 2} shown in (b). Moreover, the KL divergence D _KL (P ₂ || Q ₂ ) for the playlist L ₂ shown in (b) of FIG. 13B is the KL divergence D _KL (P ₁ || for the playlist L ₁ of (a). Q ₁ ). Therefore, if you want to play the only song 211, 212, and 213 that are included in the play list _{L 1,} rather than as a music play list _{L 1,} better to play as a song playlist _{L 2} it is less likely to occur discomfort for the song of the image .

  Therefore, in the playlist generation processing according to the present embodiment, as described above, the recommendation priority obtained by integrating the request deviation amount and the recognition deviation amount is calculated, and the recommended playlist candidate having the highest recommendation priority is selected as the recommended playlist. To the user. Further, in the playlist generation processing according to the present embodiment, as described above, a narration sound including the name of the playlist to be presented to the user is created and reproduced. Therefore, it is possible to more appropriately express the image of the music to be reproduced and to present a playlist close to the image requested by the user to the user.

If you decide to recommended playlist different playlist L ₂ is a playlist L ₁ to request the user in the example shown in (a) of FIG. 13A, the playlist generating apparatus 1, the name of the playlist L ₂ Create and play back narration audio. For example, the name of the play list L ₁ and L ₂ are each "bright" and "cool", it requested the name of the play list of users (search keyword) is assumed to be a "bright". In this case, the playlist creation device 1 narrates, for example, "The playlist of" bright "could not be created due to the lack of the number of songs. Therefore, the playlist of" cool "close to" bright "is reproduced." After playing the audio, play the songs in the playlist. Therefore, the user can listen to the music in the playlist after grasping that the “cool” music is played, and it is difficult for the user to feel uncomfortable with the image of the played music.

  When presenting another playlist instead of the playlist requested by the user, the songs to be played may be all songs within the range of the presented playlist, or songs within the range of the presented playlist. Out of the songs included in the playlist requested by the user. In that case, for example, a narration sound “Play a playlist“ bright ”and“ cool ”” may be created and played. Also, songs within the range of the presented playlist or songs included in the playlist requested by the user may be reproduced. In this case, for example, a narration sound “Play a“ bright ”or“ cool ”playlist” may be created and played. Further, the range of the playlist is not limited to the circle shown in FIG. 8, but may be any space shape.

  FIG. 14 is a diagram illustrating another example of the range of the playlist. FIG. 14 shows a two-dimensional image feature space (HK plane) similar to FIG. 13A (a).

As shown in FIG. 14, the ranges 311, 312, 313, 314, and 315 of the five playlists L ₁ , L ₂ , L ₃ , L ₄ , and L ₅ in the two-dimensional image feature space are each rectangular. There may be. In this case, the range of each playlist in the playlist DB 124 is specified by, for example, the coordinates of a point at the upper left corner in the image feature space, the range in the H-axis direction, and the range in the K-axis direction. In the example shown in FIG. 14, the range 311 of the playlist _{L 1} is a point in the upper left corner coordinates (h1, k1), are specified in the H-axis direction of the range .DELTA.H1, the K axis range ΔK1 .

  Furthermore, when the image feature space is a space of three or more dimensions, the range of each playlist is not limited to a sphere or a hypersphere, but may be a rectangular parallelepiped or a hypercube.

  Further, when calculating the recognition deviation amount and the required deviation amount, not only the normal distribution of the play list but also a uniform probability density distribution may be used.

FIG. 15 is a diagram illustrating the relationship between the range of the playlist and the probability density distribution. FIG. 15A shows a case where the range of the playlist in the two-dimensional image feature space (HK plane) is rectangular, as in FIG. Also in FIG. 15 (b) shows the probability density distribution p of H-axis direction (H) in the range 315 and around the playlist _{L 5} in (a) of FIG. 15.

As shown in (a) of FIG. 15, the range 311 and 315 of the play list _{L 1} and _{L 5} in the two-dimensional image feature space may each rectangle. In this case, assumed to have the distribution of each play list L ₁ and L ₅ is not limited to the normal distribution of the above uniform probability in the range 315 of the playlist L ₅ as shown in FIG. 15 (b) The density distribution may be p (H). If the assumption to have distributed playlist _{L 5} and a probability density distribution p (H), and h5 ≦ H ≦ h5 + ΔH5 if only p (H) = 1 / SA5 , the case of H <h5 and h5 + ΔH5 <H is Let p (H) = 0. Here, SA5 is the area of a rectangle representing the range 315 of the playlist _{L 5} in the image feature space.

  When the distribution assumed by the playlist is the probability density distribution p (H), the values of the positions of the points 214, 215, and 216 representing the music as shown in the lower graph of FIG. The distribution of music within the range can be represented by a graph in which only 1 / SA5 is set. Therefore, the recognition error amount may be calculated based on the degree of deviation of the point at 1 / SA5 in the graph.

[Second embodiment]
FIG. 16 is a diagram illustrating a functional configuration of a playlist generation device according to the second embodiment.

  As shown in FIG. 16, the playlist generation device 1 includes a request playlist input unit 101, a keyword / attribute extraction unit 102, a music narrowing unit 103, a recommended playlist candidate generation unit 104, a priority calculation unit 105 and a play list determination unit 106. Further, the playlist generation device 1 includes a narration creation unit 107, a playlist reproduction unit 108, and an audio output unit 109. Further, the playlist generation device 1 includes a keyword DB 121, a narrow-down attribute DB 122, a song DB 123, and a playlist DB 124.

  Further, the playlist generation device 1 further includes a skip operation input unit 115 and an adaptive processing unit 116. That is, the playlist generation device 1 of the present embodiment has a configuration in which the skip operation input unit 115 and the adaptive processing unit 116 are added to the playlist generation device 1 of the first embodiment. Therefore, the playlist generation device 1 of the present embodiment also functions as a playback device that reproduces a playlist, like the playlist generation device 1 of the first embodiment.

  The keyword DB 121, the narrow-down attribute DB 122, the song DB 123, and the playlist DB 124 in the playlist generation device 1 of the present embodiment are databases in which the information described in the first embodiment is registered. Further, the required playlist input unit 101 and the keyword / attribute extraction unit 102 in the playlist generation device 1 of the present embodiment have the functions described in the first embodiment. Also, the song narrowing unit 103, the recommended playlist candidate generating unit 104, the priority calculating unit 105, and the playlist determining unit 106 in the playlist generating device 1 of the present embodiment have been described in the first embodiment. Has functions. Further, the narration creation unit 107, the playlist reproduction unit 108, and the audio output unit 109 in the playlist generation device 1 of the present embodiment have the functions described in the first embodiment, respectively. Note that, when an operation on the skip operation input unit 115 is performed during the reproduction of the playlist, the playlist reproduction unit 108 in the playlist generation device 1 of the present embodiment skips the music (in other words, the reproduction of the music being reproduced). The reproduction process of the next song is started (after the reproduction process is completed). In addition, when a song is skipped, the playlist playback unit 108 outputs the playlist being played back and the song ID of the skipped song to the adaptive processing unit 116.

  The skip operation input unit 115 in the playlist generation device 1 of the present embodiment is, for example, a push button switch. When the user operates the skip operation input unit 115, the skip operation input unit 115 outputs a signal instructing the skip operation to the playlist reproduction unit 108. When receiving the signal instructing the skip operation, the playlist reproduction unit 108 skips the music and starts the reproduction processing of the next music. At this time, the playlist playback unit 108 outputs the playlist being played back and the song ID of the skipped song to the adaptive processing unit 116.

  In addition, the adaptive processing unit 116 in the playlist generation device 1 according to the present embodiment performs the assumed distribution and range of the playlist based on the playlist being reproduced acquired from the playlist reproduction unit 108 and the song ID of the skipped song. Re-learn.

  FIG. 17 is a diagram illustrating a configuration example of the adaptive processing unit. As shown in FIG. 17, the adaptive processing unit 116 includes a playlist range correction unit 116a and a playlist DB re-learning unit 116b. In addition, the adaptive processing unit 116 includes a database 116c in which correction data is stored, and a database 116b in which playlist learning data is registered.

  The playlist range correction unit 116a corrects the assumed distribution and range in the currently reproduced playlist based on the currently reproduced playlist and the song ID of the skipped song. The playlist range modification unit 116a acquires the image feature amount of the skipped song from the song DB 123 using the song ID of the skipped song. Further, the playlist range correction unit 116a corrects the assumed distribution and range of the playlist being reproduced so that the image feature amount of the song acquired from the song DB 123 falls outside the range of the playlist. The playlist range modification unit 116a stores, in the database 116c, modification data of the playlist in which the assumed distribution and the range have been modified.

  The playlist DB re-learning unit 116b re-learns the assumed distribution and range of the playlist based on the correction data in the database 116c and the playlist learning data in the database 116d. The playlist learning data is data used when learning the assumed distribution and range of the playlist. The learning of the assumed distribution and range of the playlist is performed using an arbitrary learning method when creating the playlist DB 124. For example, first, the evaluator creates initial data specifying the playlist and the coordinates of the song (image feature amount) in the image feature space based on the name of the playlist and the subjective image of the song. Then, using the initial data as learning data, a conversion function from a word feature to an image feature space and a conversion function from a music feature to an image feature are learned by, for example, kernel regression. The word feature amount is obtained by using a known conversion technique for converting a playlist name (search keyword), that is, a word into a semantic feature amount. As the music feature amount, for example, an acoustic feature amount (for example, a mel frequency cepstrum coefficient (MFCC)) used in voice recognition or the like can be used.

  After re-learning the assumed distribution and range of the playlist, the playlist DB re-learning unit 116b updates the assumed distribution and range of the playlist being reproduced in the playlist DB to the re-learned assumed distribution and range.

  The playlist generation process performed by the playlist generation device 1 of the present embodiment is the same as the process described in the first embodiment. That is, the play list generation device 1 of the present embodiment performs the processing shown in FIGS. 9A, 9B, 11, and 12 as the play list generation processing.

  Further, when the skip operation input unit 115 receives a skip operation by the user during the reproduction of the play list, the play list generation device 1 of the present embodiment performs a process of updating the play list DB in the adaptive processing unit 116.

FIG. 18 is a flowchart illustrating a process performed by the adaptive processing unit.
When the skip operation unit 115 in the playlist generation device 1 of the present embodiment receives a skip operation by the user during the reproduction of the playlist, the skip operation input unit 115 causes the playlist reproduction unit 108 to skip the song being reproduced. Output instructions The playlist playback unit 108 that has received the instruction to skip the song being played ends the playback of the song audio data that is being played, and starts playing the next song audio data. At this time, the playlist playback unit 108 outputs the song ID of the skipped song and the playlist being played back to the adaptive processing unit 116.

  At this time, as shown in FIG. 18, the adaptive processing unit 116 receives the song ID of the skipped song and the playlist being reproduced (step S21), and uses the song ID of the skipped song as key information for the skipped song. An image feature amount is obtained (step S22). The processing of steps S21 and S22 is performed by the playlist range correction unit 116a. In the process of step S22, the playlist range correction unit 116a searches the song DB 123, and acquires the image feature amount of the song associated with the song ID of the skipped song.

  When the image feature amount of the skipped song is acquired, the playlist range modification unit 116a then reduces the range of the playlist being reproduced based on the acquired image feature amount (step S23). In the process of step S23, the playlist range modification unit 116a reduces the range of the playlist so that the image feature amount of the skipped song is outside the range of the playlist.

  After reducing the playlist range, the playlist range modification unit 116a next determines whether to adjust the playlist range (step S24). In the determination of step S24, for example, it is determined whether or not to expand and adjust the range of the playlist based on the reduction amount of the range of the playlist.

  When adjusting the range of the playlist (Step S24; Yes), the playlist range correcting unit 116a expands the range of the playlist in a direction not including the feature amount of the skipped song (Step S25). Thereafter, the playlist range correction unit 116a registers the changed range of the playlist whose range has been changed in the correction data database 116c (step S26). On the other hand, when the range of the playlist is not adjusted (step S24; No), the playlist range correction unit 116a does not perform the process of step S25, and stores the changed range of the changed playlist in the database of the correction data. 116c (step S26). That is, when adjusting the range of the playlist (Step S24; Yes), the playlist range correction unit 116a registers the range of the playlist reduced and enlarged in Step S23 in the correction data database 116c in Step S23. . On the other hand, when the range of the playlist is not adjusted (Step S24; No), the playlist range correction unit 116a registers the range of the playlist reduced in Step S23 in the correction data database 116c.

  After registering the corrected playlist range in the correction data database 116c, the adaptive processing unit 116 re-learns the assumed distribution and range of the playlist based on the correction data and the playlist learning data (step S27). . The process of step S27 is performed by the playlist DB re-learning unit 116b. The playlist DB re-learning unit 116b includes, for example, the playlist range after the modification registered in the modification data in the playlist learning data, and re-learns the range of the playlist by kernel regression for the playlist range. I do. The playlist DB re-learning unit 116b re-learns the assumed distribution of the playlist based on, for example, the range of the playlist after the re-learning and the image feature amount of each song.

  After completing the re-learning of the assumed distribution and range of the playlist in step S27, the playlist DB re-learning unit 116b sets the assumed distribution and range of the corresponding playlist in the playlist DB 124 by assuming the playlist obtained by re-learning. The distribution and the range are updated (step S28).

  As described above, in the playlist generation device 1 according to the present embodiment, when the operation of skipping a song is performed during the reproduction of the playlist, the playlist is generated such that the image feature amount of the skipped song is out of the range of the playlist. Modify the range. Therefore, when the same playlist is reproduced from the next time onward, songs skipped in the past are excluded from the playlist. Therefore, a song that does not match the image of the playlist for the user can be excluded from the playlist, and the image of the playlist to be reproduced can be closer to the image of the user.

  FIG. 19 is a diagram illustrating an example of a method of correcting the range of a playlist. FIGS. 19A and 19B show a range of a playlist in a two-dimensional image feature space (HK plane) and a black circle representing an image feature amount of a song, respectively.

When the playlist generation device 1 performs the playlist generation processing described in the first embodiment, for example, as shown in FIG. 19A, there are five recommended playlist candidates and the image feature space ( It is assumed that the ranges 311, 312, 313, 314, and 315 of the five playlists L ₁ , L ₂ , L ₃ , L ₄ , and L _{5 on} the (HK plane) are each represented by a rectangle. Moreover, the play list L ₂ is determined to be the recommended playlist in the playlist decision unit 106. In this case, the play list reproduction unit 108 sequentially reproduces the music 211,212,213,217,218 five songs contained within 312 of the playlist _{L 2.}

Here, for example, the user during playback of the song 217 is included in the play list L ₂ has performed the skip operation. In this case, adaptive processing unit 116, as shown in (b) of FIG. 19, as the song 217 skipping is outside the range of the play list L2, to reduce the range of the play list L _2. As shown in (a) of FIG. 19, the rightmost in the range 312 of the playlist L _2, that is, when the song 217 the value of H is the largest in the image feature amount outside the range of the play list L _2, songs 217 is moved in the −H direction. Thereby, the image feature amounts of the other four songs 211, 212, and 218 remain in the state they were contained within 312 playlist _{L 2,} to outside only songs 217 playlist _{L 2} Can be. At this time, the movement amount of the side to be moved in the −H direction is, for example, a movement amount at which the distance d between the moved side and the image feature amount of the music piece 217 becomes a predetermined distance.

  Note that if the range of the playlist is reduced every time a song is skipped, the range of the playlist may eventually become 0 and become an invalid playlist. Therefore, when the range of the playlist is reduced, the range of the playlist may be adjusted by expanding the range in another direction by the reduced amount, as in the process of step S25.

FIG. 20 is a diagram illustrating an example of a method of expanding the range of a playlist. FIGS. 20A, 20B, and 20C respectively show a range 312 of the playlist L2 in the _two -dimensional image feature space (HK plane) and a black circle representing the image feature amount of the song. Is shown. Also, FIGS. 20A, 20B, and 20C respectively show the enlargement of the reduced range 312 when the range 312 is reduced by skipping the music 217 as shown in FIG. 4 shows an example of a method.

For example, if obtained by reducing the range 312 to move the right side in the -H direction distance hd in the range 312 of the playlist _{L 2,} as a way to expand the scope 312, as shown in (a) of FIG. 20 , The left side of the range 312 is moved by the distance hd in the −H direction. That is, in the example shown in (a) of FIG. 20, will be the range 312 of the playlist L ₂ is only moved parallel distance hd with -H direction, music without changing the area in the range 312 in the image feature space 217 is outside the range.

Also, if obtained by reducing the range 312 to move the right side in the -H direction distance hd in the range 312 of the playlist _{L 2,} as shown in (b) of FIG. 20, the distance the lower side of the range 312 -K direction The range 312 may be enlarged by moving the area by he. Although not shown, instead of moving the lower side of the range 312 in the −K direction, the range 312 may be enlarged by moving the upper side in the + K direction. Further, although not shown, the range 312 may be enlarged by moving the lower side of the range 312 in the −K direction and moving the upper side in the + K direction.

The movement, if obtained by reducing the range 312 to move the right side in the -H direction distance hd in the range 312 of the playlist _{L 2,} as shown in (c) of FIG. 20, the left side of the range 312 -H direction Alternatively, the range 312 may be enlarged by moving the lower side in the −K direction. Although not shown, the left side, the upper side, and the lower side of the range 312 may be moved in the direction in which the range 312 expands.

FIG. 21 is a diagram illustrating one of the effects obtained by expanding the range of the playlist. In (a) and (b) in FIG. 21 respectively show a black circle representing an image feature amount in the range 312 and songs in a playlist L ₂ in the two-dimensional image feature space (H-K plane). FIG. 21A shows a reduced range 312 when the range 312 is reduced by skipping the music piece 217 as shown in FIG.

One reason to skip user songs 217 contained in the range 312 of the playlist L _2, the image of the song 217 and the like that does not match the playlist L ₂ image for the user. In this case, the cause of the difference between the image image playlists L ₂ songs 217 (conflict), for example, range 312 of the playlist L ₂ is too wide and, like the center of the range 312 is displaced Is mentioned. If the center of the range 312 is shifted, for example, an image of the song 219 located in the vicinity of the range 312 outside a is and the range 312 it is also conceivable to match the image playlist L _2. By adjusting the range 312 after reduction (enlargement modifications), as shown in (b) of FIG. 21, the playlist songs 219 close to the image of the playlist L ₂ existing outside the range of the play list L ₂ L ₂ Range 312. Therefore, it is possible to provide a song that matches the of the play list L ₂ during playlist generation image to more users.

Further, if the skip operation by the user during reproduction of music 219 was to be included in the playlist L ₂ is performed by expanding modify the range 312, range 312 of the playlist L ₂ is included songs 219 It is re-learned (re-corrected) to a range that does not exist. Therefore, it is possible to a proper range more suited to the range 312 of the playlist L ₂ to playlist L ₂ images.

  As described above, in the playlist generation processing according to the present embodiment, when the user performs an operation of skipping a song during the playlist playback, the range of the playlist being played based on the song feature amount of the skipped song is Is corrected, and the assumed distribution and range are re-learned. Therefore, the assumed distribution and range of each playlist can be adaptively approximated to the image held by the user, and the discomfort given to the user due to the discrepancy between the image of the playlist and the image of the reproduced music is further reduced. Becomes possible.

[Third Embodiment]
FIG. 22 is a diagram illustrating a configuration of a playlist playback system according to the third embodiment.

  As shown in FIG. 22, the playlist playback system 4 according to the present embodiment includes a music playback device 5 and a server 6. The music generation device 5 and the server 6 are communicably connected via a communication network 7 such as the Internet.

  The music playback device 5 is a device that acquires music data (song audio data) from the server 6 and plays back the music data. The music reproducing device 5 includes an audio input unit (microphone) 501, an audio output unit 502, a signal processing unit 503, and a communication unit 504.

  The sound input unit 501 collects sound propagating around the music reproducing device 5 and inputs the sound as a sound signal. The audio output unit 502 is, for example, a speaker, and outputs an audio signal including music (song audio) as audio to a space outside the music playback device 5. Note that the audio output unit 502 is not limited to a speaker, and may be an output terminal for connecting an audio device such as an earphone or an amplifier.

  The signal processing unit 503 performs a conversion process of the audio signal input from the audio input unit 501, a reproduction process of the music audio data acquired from the server, and the like. The communication unit 504 performs data communication with a communication unit (not shown) of the server 6.

  FIG. 23 is a diagram illustrating a functional configuration of a server in the playlist playback system according to the third embodiment.

  As shown in FIG. 23, the server 6 in the playlist playback system 4 of the present embodiment includes a communication unit 601, a voice recognition unit 602, a keyword / attribute extraction unit 603, and a music narrowing unit 604. In addition, the server 6 includes a recommended playlist candidate generation unit 605, a priority calculation unit 606, a playlist determination unit 607, a narration creation unit 608, and a playlist reproduction unit 609. Further, the server 6 includes a keyword DB 621, a narrowing down attribute DB 622, a music DB 623, and a playlist DB 624.

  The communication unit 601 performs data communication with the communication unit 504 of the music playback device 5. The communication unit 601 receives the audio signal collected by the audio input unit 501 of the music playback device 5. In addition, the communication unit 601 transmits the audio data reproduced by the play list reproduction unit 609 to the music reproduction device 5.

  The voice recognition unit 602 analyzes a voice signal from the music playback device 5 input via the communication unit 601 and converts significant voice included in the voice signal into text.

  The keyword / attribute extraction unit 603 extracts a search keyword from the keyword DB 621 based on the text obtained by the speech recognition unit 602, and extracts a song narrowing attribute from the narrowing attribute DB 622. The keyword DB 621 is a database in which a plurality of correspondences between texts and search keywords are registered, similarly to the keyword DB 121 described in the first embodiment. The search keyword is a keyword used for searching for a playlist. The narrow-down attribute DB 622 is a database in which a plurality of correspondences between texts and narrow-down attributes of songs are registered, similarly to the narrow-down attribute DB 122 described in the first embodiment. The song narrowing down attribute is attribute information used to narrow down songs to be registered in the playlist.

  The song narrowing unit 604 narrows down the candidate songs to be registered in the playlist to the songs having attributes matching the narrowing attribute from all the songs registered in the song DB 623 based on the extracted narrowing attributes. . The music DB 623 is a database in which a plurality of correspondences between music audio data, music attributes, and image feature amounts of music are registered, similarly to the music DB 123 described in the first embodiment. The song attribute is attribute information used to narrow down song audio data. The music image feature amount is information obtained by digitizing the music image.

  The recommended playlist candidate creation unit 605 extracts a playlist including the songs narrowed down by the song narrowing unit 604 from the playlist DB 624 and creates a recommended playlist candidate. The recommended playlist candidates are playlist candidates that are recommended (presented) to the user (operator) of the music playback device 5. The recommended playlist candidate includes a playlist name, a song ID, and an assumed distribution. Like the playlist DB 124 described in the first embodiment, the playlist DB 624 is a database in which playlist names, assumed distributions, and playlist ranges are registered in association with each playlist. The assumed distribution of the playlist is the average (center) and variance of the playlist in the image feature space when expressing the distribution of the song based on the image feature amount of the song. The range of the playlist is a range in which the music to be registered in the playlist in the image feature space exists.

  The priority calculating unit 606 calculates the priority (recommended priority) of each recommended playlist candidate when the recommended playlist candidate creating unit 605 creates a plurality of recommended playlist candidates. The priority calculation unit 606 calculates the priority of each recommended playlist candidate based on the assumed distribution and range of the recommended playlist candidate and the search keyword.

  The playlist determination unit 607 determines a playlist to be adopted from among a plurality of recommended playlist candidates based on the recommendation priority calculated by the priority calculation unit 606.

  The narration creation unit 608 creates narration audio data including the name of the playlist determined by the playlist determination unit 607.

  The playlist playback unit 609 plays back the narration voice created by the narration creation unit 608, and reads and plays back song audio data included in the playlist determined by the playlist determination unit 607 from the song DB 623.

  As described above, the server 6 in the playlist playback system 4 according to the present embodiment includes the request playlist input unit 101 and the audio output unit 109 in the playlist playback device 1 according to the first embodiment as the communication unit 601 and the audio recognition unit. 102. That is, the server 6 in the playlist reproduction system 4 of the present embodiment functions as a playlist generation device. When receiving the audio signal requesting the reproduction of the playlist from the music reproducing device 5, the server 6 performs a playlist generation process as shown in FIGS. 9A, 9B, 11, and 12. At this time, the voice data and the tune voice data of the narration voice reproduced by the playlist reproduction unit 609 are transmitted to the music reproduction device 5 via the communication unit 601. The music reproducing device 5 performs predetermined processing on the voice data of the narration voice and the music voice data acquired from the server 6 in the signal processing unit 503, and then outputs the voice and the music voice from the voice output unit 502.

  In the playlist reproduction system 4 of the present embodiment, the server 6 performs the playlist generation processing as shown in FIGS. 9A, 9B, 11, and 12. Therefore, in the playlist playback system 4, an information terminal that can communicate and can reproduce an audio signal, such as a car navigation system, can be used as the music playback device 5. When the song DB 623 is provided in the server 6 as in the playlist playback system according to the present embodiment, more song audio data is stored in the song DB 623 than in the playlist generation device 1 according to the first embodiment. It is possible to do.

  Note that the playlist reproduction system 4 is not limited to a configuration in which a playlist generation request is input by voice as described in the present embodiment, but may be a configuration in which a playlist generation request is input as text. Further, the playlist playback system 4 may have a configuration that allows input by voice and input by text.

[Fourth embodiment]
FIG. 24 is a diagram illustrating a configuration of a playlist playback system according to the fourth embodiment.

  As shown in FIG. 24, the playlist playback system 4 according to the present embodiment includes a music playback device 5 and a server 6. The music generation device 5 and the server 6 are communicably connected via a communication network 7 such as the Internet.

  The music reproducing device 5 is a device for reproducing music data (song audio data). The music reproducing device 5 includes a voice input unit (microphone) 501, a voice output unit 502, a communication unit 504, and a voice recognition unit 505. Further, the music playback device 5 includes a data input unit 506, a music / sound data registration unit 507, a narration sound creation unit 508, and a playlist playback unit 509. Further, the music reproducing device 5 includes a database 521 for storing music IDs and music sound data.

  The sound input unit 501 collects sound propagating around the music reproducing device 5 and inputs the sound as a sound signal. The audio output unit 502 is, for example, a speaker, and outputs an audio signal including music (song audio) as audio to a space outside the music playback device 5. Note that the audio output unit 502 is not limited to a speaker, and may be an output terminal for connecting an audio device such as an earphone or an amplifier.

The communication unit 504 performs data communication with a communication unit (not shown) of the server 6.
The voice recognition unit 505 analyzes the voice signal input by the voice input unit 501, and converts significant voice included in the voice signal into text.

  The data input unit 506 receives an input of music audio data created from audio data such as a music CD.

  The song audio data registration unit 507 registers song audio data input via the data input unit 506 and song audio data obtained via the communication network 7 in the database 521. The music sound data registration unit 507 performs a process of acquiring a music ID of the music sound data input via the data input unit 506 from a music DB (not shown) of the server 6. The song sound data registration unit 507 performs a process of registering the song ID of the song sound data registered in the database 521 in a song ID DB (not shown) of the server 6.

  The narration speech creation unit 508 converts the text of the narration sentence created by the server 6 into speech data.

  The playlist playback unit 509 plays back the narration voice created by the narration voice creation unit 508 and plays back the playlist created by the server 6. The server 6 creates a playlist for the music and sound data stored in the database 521 of the music playback device 5. Therefore, the play list reproducing unit 509 reads out and reproduces the tune audio data from the database 521 according to the play list.

  FIG. 25 is a diagram illustrating a functional configuration of a server in the playlist playback system according to the fourth embodiment.

  As shown in FIG. 25, the server 6 in the playlist reproduction system 4 of the present embodiment includes a communication unit 601, a keyword / attribute extraction unit 603, and a music narrowing unit 604. Further, the server 6 includes a recommended playlist candidate generation unit 605, a priority calculation unit 606, a playlist determination unit 607, and a narration text creation unit 608a. In addition, the server 6 includes a keyword DB 621, a refinement attribute DB 622, a song DB 623, a playlist DB 624, and a song ID DB 623.

  The communication unit 601 performs data communication with the communication unit 504 of the music playback device 5. The communication unit 601 receives the text (speech recognition result) transmitted by the music reproducing device 5, the song ID of the song audio data, and the like. In addition, the communication unit 601 transmits the playlist determined by the playlist determination unit 607, the text of the narration text created by the narration text creation unit 608a, and the like to the music playback device 5.

  The keyword / attribute extraction unit 603 extracts a search keyword from the keyword DB 621 based on the text (speech recognition result) acquired from the music reproducing device 5 and extracts a song narrowing attribute from the narrowing attribute DB 622. The keyword DB 621 is a database in which a plurality of correspondences between texts and search keywords are registered, similarly to the keyword DB 121 described in the first embodiment. The search keyword is a keyword used for searching for a playlist. The narrow-down attribute DB 622 is a database in which a plurality of correspondences between texts and narrow-down attributes of songs are registered, similarly to the narrow-down attribute DB 122 described in the first embodiment. The song narrowing down attribute is attribute information used to narrow down songs to be registered in the playlist.

  The song narrowing unit 604 selects a song that is a candidate for registration in the playlist based on the extracted narrowing attribute, and registers a song ID in the song ID DB 625 among all songs registered in the song DB 623. The songs that have the attribute that matches the refinement attribute are narrowed down from the songs that exist. The music DB 623 is a database in which a plurality of correspondences between music audio data, music attributes, and image feature amounts of music are registered, similarly to the music DB 123 described in the first embodiment. The song attribute is attribute information used to narrow down song audio data. The music image feature amount is information obtained by digitizing the music image. The music ID DB 625 is a database in which music IDs of music sound data stored in the database 521 of the music playback device 5 are registered.

  The recommended playlist candidate creation unit 605 extracts a playlist including the songs narrowed down by the song narrowing unit 604 from the playlist DB 624 and creates a recommended playlist candidate. The recommended playlist candidates are playlist candidates that are recommended (presented) to the user (operator) of the music playback device 5. The recommended playlist candidate includes a playlist name, a song ID, and an assumed distribution. Like the playlist DB 124 described in the first embodiment, the playlist DB 624 is a database in which playlist names, assumed distributions, and playlist ranges are registered in association with each playlist. The assumed distribution of the playlist is the average (center) and variance of the playlist in the image feature space when expressing the distribution of the song based on the image feature amount of the song. The range of the playlist is a range in which the music to be registered in the playlist in the image feature space exists.

  The priority calculating unit 606 calculates the priority (recommended priority) of each recommended playlist candidate when the recommended playlist candidate creating unit 605 creates a plurality of recommended playlist candidates. The priority calculation unit 606 calculates the priority of each recommended playlist candidate based on the assumed distribution and range of the recommended playlist candidate and the search keyword.

  The playlist determination unit 607 determines a playlist to be adopted from among a plurality of recommended playlist candidates based on the recommendation priority calculated by the priority calculation unit 606.

  The narration sentence creation unit 608 creates a narration sentence text including the name of the playlist determined by the playlist determination unit 607.

  In the playlist playback system 4 of the present embodiment, the requested playlist input unit 101, the narration sound creation unit 107b, the playlist playback unit 108, and the audio output unit 109 of the playlist playback device 1 according to the first embodiment play music. The playback device 5 is provided. That is, the server 6 in the playlist reproduction system 4 of the present embodiment functions as a playlist generation device. In the playlist generation process performed by the playlist reproduction system 4, among the processes of steps S1 to S9 shown in FIGS. 9A and 9B, a part of steps S1, S2, and S8, and the process of S9 are performed by the music reproducing apparatus. 5 does. That is, in the playlist generation process performed by the playlist reproduction system 4, the server 6 performs the processes of steps S3 to S7 illustrated in FIG. 9A.

  Note that, as described above, the server 6 in the playlist playback system 4 of the present embodiment generates a playlist only for the music and sound data registered in the database 521 of the music playback device 5. Therefore, the music playback device 5 performs a process of registering the song ID of the song audio data registered in the database 521 in the song ID DB 625 of the server 6. This registration process is performed in the following procedure, for example.

  For example, when music sound data is input from the data input unit 506, the music reproducing device 5 performs a process of acquiring a music ID for the music sound data input by the music sound data registration unit 507. The song audio data registration unit 507 searches the song DB 623 of the server 6 using the input song audio data, and acquires the song ID of the song audio data. The song audio data registration unit 507 registers the acquired song ID in the song ID DB 625 of the server 6 and also registers the song audio data and the song ID in the database 521 of the music playback device 5.

  On the other hand, among the processes performed by the server 6, the song narrowing unit 604 narrows down songs based on the narrowing attribute (step S4), and only the songs with the song IDs registered in the song ID DB 625 become the narrowing attributes. Narrow down to songs with matching song attributes.

  As described above, in the playlist playback system 4 according to the present embodiment, most of the playlist generation processing for the music sound data registered in the music playback device 5 is performed by the server 6 different from the music playback device 5. Do. Therefore, the processing load on the music playback device 5 can be reduced.

  Further, in the playlist playback system 4 according to the present embodiment, since only the music sound data registered in the music playback device 5 is to be registered in the playlist, the music playback device 5 is transmitted from the server 6 via the communication network 7. The transfer of the song audio data is not performed. Therefore, even when the communication quality is unstable, such as when wireless communication is performed while moving, the music sound data can be stably reproduced.

  Note that the playlist reproduction system 4 is not limited to a configuration in which a playlist generation request is input by voice as described in the present embodiment, but may be a configuration in which a playlist generation request is input as text. Further, the playlist playback system 4 may have a configuration that allows input by voice and input by text.

[Fifth Embodiment]
FIG. 26 is a diagram illustrating a configuration of a playlist playback system according to the fifth embodiment.

  As shown in FIG. 26, the playlist playback system 4 according to the present embodiment includes a music playback device 5, a server 6, and a plurality of portable electronic devices 8 (8A, 8B). The music generation device 5 and the server 6 are communicably connected via a communication network 7 such as the Internet. Further, the music reproducing device 5 and the portable electronic device 8 are communicably connected by short-range wireless communication such as Bluetooth (registered trademark). The server 6 in the playlist playback system 4 of the present embodiment may have the same configuration as the server 6 of the fourth embodiment (see FIG. 25).

  The music reproducing device 5 is a device for reproducing music data (song audio data). The music reproducing device 5 includes a voice input unit (microphone) 501, a voice output unit 502, a first communication unit 504, a second communication unit 510, and a voice recognition unit 505. Further, the music playback device 5 includes a data input unit 506, a music / sound data registration unit 507, a narration sound creation unit 508, and a playlist playback unit 509. Further, the music reproducing device 5 includes a database 521 for storing music IDs and music sound data.

  The sound input unit 501 collects sound propagating around the music reproducing device 5 and inputs the sound as a sound signal. The audio output unit 502 is, for example, a speaker, and outputs an audio signal including music (song audio) as audio to a space outside the music playback device 5. Note that the audio output unit 502 is not limited to a speaker, and may be an output terminal for connecting an audio device such as an earphone or an amplifier.

  The first communication unit 504 performs data communication with a communication unit (not shown) of the server 6. The second communication unit 510 performs data communication with the communication unit 801 included in the portable electronic device 8.

  The voice recognition unit 505 analyzes the voice signal input by the voice input unit 501, and converts significant voice included in the voice signal into text.

  The data input unit 506 receives an input of music audio data created from audio data such as a music CD.

  The song audio data registration unit 507 registers song audio data input via the data input unit 506 and song audio data obtained via the communication network 7 in the database 521. The music sound data registration unit 507 performs a process of acquiring a music ID of the music sound data input via the data input unit 506 from a music DB (not shown) of the server 6. The song sound data registration unit 507 performs a process of registering the song ID of the song sound data registered in the database 521 in a song ID DB (not shown) of the server 6.

  The narration speech creation unit 508 converts the text of the narration sentence created by the server 6 into speech data.

  The playlist playback unit 509 plays back the narration voice created by the narration voice creation unit 508 and plays back the playlist created by the server 6. The server 6 creates a playlist for the music and sound data stored in the database 521 of the music playback device 5. Therefore, the play list reproducing unit 509 reads out and reproduces the tune audio data from the database 521 according to the play list.

  The portable electronic device 8 (8A, 8B) is an electronic device capable of wireless communication or wired communication such as a smartphone or a tablet computer. The portable electronic device 8 includes a communication unit 801, a database 821 in which music data is registered, and an input operation unit (not shown). The communication unit 801 performs data communication with the second communication unit 510 of the music playback device 5. The communication unit 801 of the portable electronic device 8 and the second communication unit 510 of the music playback device 5 perform data communication by wireless communication according to a predetermined wireless communication standard.

  In the playlist playback system 4 according to the present embodiment, the requested playlist input unit 101, the narration sound creation unit 107b, the playlist playback unit 108, and the audio output unit 109 of the playlist playback device 1 according to the first embodiment play music. The playback device 5 is provided. Therefore, in the playlist generation processing performed by the playlist reproduction system 4, among the processing of steps S1 to S9 shown in FIGS. 9A and 9B, a part of steps S1, S2, and S8, and the processing of S9 are performed by the music reproducing apparatus. 5 does. That is, in the playlist generation process performed by the playlist reproduction system 4, the server 6 performs the processes of steps S3 to S7 illustrated in FIG. 9A. Therefore, also in the playlist reproduction system 4 of the present embodiment, the server 6 functions as a playlist generation device.

  Note that, as described above, the server 6 in the playlist playback system 4 of the present embodiment generates a playlist only for the music and sound data registered in the database 521 of the music playback device 5. Therefore, the music playback device 5 performs a process of registering the song ID of the song audio data registered in the database 521 in the song ID DB 625 of the server 6. This registration process is performed in the following procedure, for example.

  For example, when music sound data is input from the data input unit 506, the music reproducing device 5 performs a process of acquiring a music ID for the music sound data input by the music sound data registration unit 507. The song audio data registration unit 507 searches the song DB 623 of the server 6 using the input song audio data, and acquires the song ID of the song audio data. The song audio data registration unit 507 registers the acquired song ID in the song ID DB 625 of the server 6 and also registers the song audio data and the song ID in the database 521 of the music playback device 5.

  Further, in the playlist playback system 4 of the present embodiment, it is possible to transfer the music and sound data registered in the database 821 of the portable electronic device 8 to the music playback device 5. Also in the case where the music sound data of the portable electronic device 8 is transferred to the music reproduction device 5, the music reproduction device 5 acquires the music ID by the same method, registers the music ID in the music ID DB 625, and stores the music sound data and the music ID. Is registered in the database 521 of the music playback device 5.

  On the other hand, among the processes performed by the server 6, the song narrowing unit 604 narrows down songs based on the narrowing attribute (step S4), and only the songs with the song IDs registered in the song ID DB 625 become the narrowing attributes. Narrow down to songs with matching song attributes.

  As described above, in the playlist playback system 4 according to the present embodiment, most of the playlist generation processing for the music sound data registered in the music playback device 5 is performed by the server 6 different from the music playback device 5. Do. Therefore, the processing load on the music playback device 5 can be reduced.

  Further, in the playlist playback system 4 according to the present embodiment, since only the music sound data registered in the music playback device 5 is to be registered in the playlist, the music playback device 5 is transmitted from the server 6 via the communication network 7. The transfer of the song audio data is not performed. Therefore, even when the communication quality is unstable, such as when wireless communication is performed while moving, the music sound data can be stably reproduced.

  Further, in the playlist playback system 4 according to the present embodiment, even when the song audio data is not registered in the database 521 of the music playback device 5, the song audio data is acquired from the portable electronic device 8 and played. can do. Therefore, the work of registering the music sound data generated from the music CD or the like in the database 521 of the music reproducing device 5 in advance can be omitted.

  Furthermore, in the playlist playback system 4 according to the present embodiment, as shown in FIG. 26, song audio data can be collected from a plurality of portable electronic devices 8 and registered in the database 521 of the music playback device 5. . Therefore, when the play list is reproduced, each user can listen to music that is not registered in the portable electronic device 8 owned by the user.

  The playlist generation device 1 and the music reproduction device 5 are configured to input a playlist generation request and a reproduction request by voice input using the request playlist input unit 101 (microphone 501). Such a configuration is particularly preferably applied to an in-vehicle device. When the driver of the vehicle uses the playlist generation device 1 or the music playback device 5 while driving, it is difficult to select a playlist while looking at the device body or a display device connected to the device. Therefore, it is preferable to select a playlist by voice input as described in the above embodiment. However, also in the playlist reproduction system 4 of the present embodiment, the playlist generation request and the reproduction request are not limited to voice input, and may be input as text by operating an input device such as a touch panel or a switch. .

  Further, the playlist generation device 1, the music playback device 5, and the server 6 can be realized using a computer and a program executed by the computer. Hereinafter, the playlist generation device 1 and the like realized using a computer and a program will be described with reference to FIG.

  FIG. 27 is a diagram illustrating a hardware configuration of a computer. As shown in FIG. 27, a computer 10 that can be operated as the playlist generation device 1 includes a Central Processing Unit (CPU) 1001, a main storage device 1002, an auxiliary storage device 1003, an input device 1004, a display Device 1005. In addition, the computer 10 further includes an interface device 1006, a storage medium driving device 1007, and a communication device 1008. These elements 1001 to 1008 in the computer 10 are interconnected by a bus 1010 so that data can be transferred between the elements.

  The CPU 1001 is an arithmetic processing device that controls various operations of the computer 10 by executing various programs including an operating system.

  The main storage device 1002 includes a read only memory (ROM) and a random access memory (RAM) (not shown). For example, a predetermined basic control program read by the CPU 1001 when the computer 10 is started is recorded in the ROM in advance. The RAM is used as a work storage area when necessary when the CPU 1001 executes various programs. The RAM of the main storage device 1002 can be used for temporarily storing extracted search keywords, narrowing attributes, recommended playlist candidates, and the like.

  The auxiliary storage device 1003 is a storage device such as a hard disk drive (HDD) or a solid state drive (SSD) having a larger capacity than the main storage device 1002. The auxiliary storage device 1003 can store various programs executed by the CPU 1001, various data, and the like. In the auxiliary storage device 1003, for example, as a program to be stored, for example, there is a reproduction program of music and sound data including the above-described playlist generation processing. The data to be stored in the auxiliary storage device 1003 includes, for example, a keyword DB 121, a narrow-down attribute DB 122, a music DB 123, and a playlist DB 124.

  The input device 1004 is, for example, a push button switch, a keyboard device, a mouse device, or the like. When a user (operator) performs a predetermined operation, information corresponding to the operation is input to the CPU 1001 or the like. Further, as the input device 1004, for example, a touch panel device that is disposed so as to overlap the display surface of the display device 1005 can be used.

  The display device 1005 is a display device such as a liquid crystal display. The display device 1005 displays various texts, images, and the like according to display data transmitted from the CPU 1001 or the like.

  The interface device 1006 is an input / output device that connects the computer 10 to another electronic device and enables data transmission and reception between the computer 10 and the other electronic device. The interface device 1006 includes, for example, a terminal to which a cable having a connector of the Universal Serial Bus (USB) standard can be connected. Examples of electronic devices that can be connected to the computer 10 by the interface device 1006 include a microphone 11 and a speaker 12.

  The storage medium driving device 1007 reads out programs and data recorded on a portable storage medium (not shown) and writes data and the like stored in the auxiliary storage device 1003 on the portable storage medium. As a portable storage medium, for example, a flash memory provided with a USB standard connector can be used. Further, as a portable storage medium, an optical disk such as a Compact Disk (CD), a Digital Versatile Disc (DVD), and a Blu-ray Disc (Blu-ray is a registered trademark) can be used.

  The communication device 1008 is a device that communicably connects the computer 10 and a communication network such as the Internet or a Local Area Network (LAN), and controls communication with another communication terminal (computer) via the communication network.

  When operating the computer 10 as the playlist generation device 1, the CPU 1001 reads a program including a playlist generation process as shown in FIGS. 9A and 9B from the auxiliary storage device 1003 or the like and executes the program. At this time, the CPU 1001 searches each database such as the keyword DB 121 stored in the auxiliary storage device 1003 or the like, generates a recommended playlist candidate, and calculates the priority of each candidate. Then, a playlist to be recommended to the user (operator) is determined based on the calculated priority. Further, the CPU 1001 generates and reproduces a narration sound including the name of the determined playlist, and also reproduces song audio data of a song included in the determined playlist.

  When the computer 10 is operated as the music reproducing device 5, the CPU 1001 reads a program including the process of registering the music and sound data and the process of reproducing the playlist described in the fourth embodiment from the auxiliary storage device 1003 and the like. Execute. At this time, the computer 10 obtains music sound data from a music CD or the like using the storage medium driving device 1007, for example. Thereafter, in the computer 10, the CPU 1001 acquires the song ID from the server 6 via the communication device 1008, and stores the song audio data and the song ID in the auxiliary storage device 1003. Further, the CPU 1001 performs a voice recognition process on the voice signal input from the microphone 11 and converts significant voice included in the voice signal into text. Further, the CPU 1001 performs a process of creating and reproducing narration audio including the name of the playlist to be reproduced, a process of reading music audio data included in the playlist from the auxiliary storage device 1003, and reproducing the same.

  Note that the computer 10 does not need to include all the components shown in FIG. 27, and some of the components can be omitted depending on the application and conditions. For example, in the case of the computer 10 which is assumed to be mounted on a vehicle and used, the storage medium driving device 1007 may be omitted.

  The computer 10 is not limited to a general-purpose computer that realizes a plurality of functions by executing various programs. For example, the computer 10 is a dedicated information processing device specialized in processing for operating the playlist generation device 1. May be.

  As described above, in the present specification, the playlist generation device 1 and the playlist reproduction system 4 that generate a playlist based on the feature amount representing the image of the music sound data (music data) have been described. However, the data to be registered in the playlist is not limited to the music sound data, but may be any data that can be classified by the feature amount representing the image. For example, the data registered in the playlist may be a document for reading and reproducing a novel or the like. Further, the data registered in the playlist may be image data, moving image data, or the like. When registering image data in a playlist, by playing back the image data in a slide show based on the playlist, it is possible to resolve the discrepancy between the image of the playlist and the image of the image to be played.

Regarding the embodiment including each of the examples described above, the following supplementary notes are further disclosed.
(Appendix 1)
Computer
The music is narrowed down based on the attribute information indicating the attribute of the music to be registered in the playlist included in the playlist generation request from the user,
Generating a requested playlist based on the feature amount representing the image of the narrowed down song and the range of the feature amount corresponding to the image of the playlist included in the generation request;
When there is a bias in the feature amount of the song registered in the request playlist within the range of the feature amount of the generated request playlist, based on the feature amount of the song registered in the request playlist. Changing the name of the playlist presented to the user,
A playlist generation method, characterized by performing a process.
(Appendix 2)
When there is a bias in the feature amount of the song registered in the request playlist within the range of the feature amount of the request playlist, the computer determines the feature amount of the song registered in the request playlist. Executing a process of changing a name of the playlist to be presented to the user based on the changed image and creating a recommended playlist recommended to the user based on the changed image.
3. The playlist generation method according to claim 1, wherein
(Appendix 3)
In the process of creating the recommended playlist, the computer includes, within the narrowed-down songs, a range of the feature amount corresponding to the image of the playlist after the change, and the feature amount of the requested playlist. Creating the recommended playlist in which only songs having feature amounts within the range are registered;
3. The playlist generation method according to claim 2, wherein
(Appendix 4)
In the process of creating the recommended playlist, the computer has registered all songs having feature amounts within the feature amount range corresponding to the image of the playlist after the change among the narrowed songs. Creating the recommended playlist,
3. The playlist generation method according to claim 2, wherein
(Appendix 5)
The computer further executes a process of generating a playlist candidate recommended to the user based on the narrowed down feature amount of the song,
In the process of changing the name of the playlist to be presented to the user, the computer includes a distribution of a feature amount of the song registered as the recommended playlist candidate and the distribution of the feature amount registered in the request playlist. Determining a playlist to be recommended to the user based on a distribution of a feature amount of a song;
3. The playlist generation method according to claim 1, wherein
(Appendix 6)
A process of determining a playlist to be recommended to the user based on the distribution of the feature amounts of the songs registered in the recommended playlist candidates and the distribution of the feature amounts of the songs registered in the request playlist; Then, the computer is
Using the distribution of the feature amounts of the songs included in the recommended playlist candidates and the distribution of the feature amounts of the songs assumed to be the recommended playlist candidates, the recognition misregistration amount is calculated. Calculating the required deviation amount using the distribution of the feature amount of the music assumed as the candidate of the playlist to be recommended and the distribution of the feature amount of the music assumed in the requested playlist,
Calculating a recommendation priority for the recommended playlist candidate based on the recognition shift amount and the request shift amount;
Determine the candidate with the highest recommendation priority as a playlist to recommend to the user,
A playlist generation method according to claim 5, wherein
(Appendix 7)
The computer calculates the recognition deviation amount and the required deviation amount with the distribution of the feature amount of the music as a normal distribution,
6. The playlist generation method according to claim 6, wherein
(Appendix 8)
The computer calculates the recognition shift amount and the required shift amount as a uniform distribution of the feature amount of the music,
6. The playlist generation method according to claim 6, wherein
(Appendix 9)
Said computer,
Play songs included in the playlist presented to the user,
When the operation information for skipping the reproduction is input during the reproduction of the song, the playlist is set such that the feature amount of the skipped song is out of the range of the feature amount of the song to be registered in the playlist. Correcting the range of the feature amount, and further executing a process,
3. The playlist generation method according to claim 1, wherein
(Appendix 10)
In the process of correcting the range of the feature amount of the playlist, the computer reduces the range of the feature amount of the playlist,
A playlist generation method according to claim 9, wherein the playlist is generated.
(Appendix 11)
In the process of correcting the range of the feature amount of the playlist, the computer includes:
After reducing the range of the feature amount of the playlist,
Expanding the range of the feature amount of the playlist in a manner that does not include the feature amount of the skipped song;
A playlist generation method according to claim 9, wherein the playlist is generated.
(Appendix 12)
Said computer,
Generating audio data including the name of the playlist to be presented to the user, reproducing the audio data and presenting the name of the playlist to the user, further executing a process;
3. The playlist generation method according to claim 1, wherein
(Appendix 13)
Extraction for extracting a keyword representing an image of the playlist and attribute information representing an attribute of a song to be registered in the playlist from a database prepared in advance based on a playlist generation request from a user. Department and
A song narrowing-down unit that narrows down the songs to be registered in the playlist based on the attribute information;
A recommended playlist candidate generation unit that generates a playlist candidate to be recommended to the user based on the feature amount representing the image of the tuned song;
A priority calculation unit that generates a requested playlist corresponding to the keyword, and calculates a priority for the recommended playlist candidate based on the requested playlist and a candidate playlist recommended to the user;
A playlist determination unit that determines a playlist to be presented to the user based on the calculated priority,
An information generation unit that generates information that presents the determined name of the playlist to the user.
A playlist generation device, characterized in that:
(Appendix 14)
The playlist determining unit determines only songs having feature amounts that fall within the range of the feature amount corresponding to the determined image of the playlist and fall within the range of the feature amount corresponding to the image of the requested playlist. Create the registered playlist,
14. The playlist generation device according to supplementary note 13, wherein
(Appendix 15)
The priority calculation unit,
Recognition misregistration that calculates a recognition misregistration amount using the distribution of the feature amount of the song registered in the recommended playlist candidate and the distribution of the feature amount of the song assumed as the recommended playlist candidate. An amount calculator,
Request deviation amount calculation using the distribution of the feature amount of the music assumed as the candidate of the recommended playlist and the distribution of the characteristic amount of the music assumed in the required playlist. Department and
A shift amount integration unit that integrates the recognition shift amount and the request shift amount and calculates a recommendation priority for the recommended playlist candidate.
14. The playlist generation device according to supplementary note 13, wherein
(Appendix 16)
The playlist generation device further includes an identification information database for registering identification information of the music stored in the external device,
The song narrowing unit narrows only songs that satisfy the attribute information from the songs registered in the identification information database,
14. The playlist generation device according to supplementary note 13, wherein
(Appendix 17)
The playlist generation device may be configured such that, when an operation of skipping a song being played by the user is performed during playback of the playlist presented to the user, the skipped song is the feature of the playlist. An adaptive processing unit that corrects the playlist so as to be out of the range of the amount.
14. The playlist generation device according to supplementary note 13, wherein
(Appendix 18)
The music is narrowed down based on the attribute information indicating the attribute of the music to be registered in the playlist included in the playlist generation request from the user,
Generating a requested playlist based on the feature amount representing the image of the narrowed down song and the range of the feature amount corresponding to the image of the playlist included in the generation request;
When there is a bias in the feature amount of the song registered in the request playlist within the range of the feature amount of the generated request playlist, based on the feature amount of the song registered in the request playlist. Changing the name of the playlist presented to the user,
A program that causes a computer to execute processing.
(Appendix 19)
When there is a bias in the feature amount of the song registered in the request playlist within the range of the feature amount of the request playlist, the computer sends the feature amount of the song registered in the request playlist to the computer. Changing the name of the playlist to be presented to the user based on the changed image, creating a recommended playlist recommended to the user based on the changed image, and executing a process.
18. The program according to supplementary note 18, wherein:
(Appendix 20)
The process of creating the recommended playlist includes, within the narrowed-down songs, the range of the feature amount corresponding to the image of the playlist after the change, and the range of the feature amount of the requested playlist. Including a process of creating the recommended playlist in which only songs having the following feature amounts are registered:
20. The program according to claim 19, wherein:
(Appendix 21)
The process of creating the recommended playlist includes, among the narrowed-down songs, the recommended playlist in which all songs having feature amounts within a range of the feature amount corresponding to the image of the playlist after the change are registered. Including the process of creating
20. The program according to claim 19, wherein:
(Appendix 22)
The computer further executes a process of generating a playlist candidate recommended to the user based on the narrowed down feature amount of the music,
The process of changing the name of the playlist to be presented to the user includes, in the computer, a distribution of the feature amounts of the songs registered in the recommended playlist candidates and the Determining a playlist to be recommended to the user based on the difference from the distribution of the feature amount of the song, and changing the name of the playlist to the name of the recommended playlist;
18. The program according to supplementary note 18, wherein:
(Appendix 23)
A playlist recommended to the user based on a difference between the distribution of the feature amounts of the songs registered in the recommended playlist candidates and the distribution of the feature amounts of the songs registered in the requested playlist. The process to determine is
Using the distribution of the feature amounts of the songs included in the recommended playlist candidates and the distribution of the feature amounts of the songs assumed to be the recommended playlist candidates, the recognition misregistration amount is calculated. Calculating the required deviation amount using the distribution of the feature amount of the music assumed as the candidate of the playlist to be recommended and the distribution of the feature amount of the music assumed in the requested playlist,
Calculating a recommendation priority for the recommended playlist candidate based on the recognition shift amount and the request shift amount;
Determining the candidate with the highest recommendation priority as a playlist recommended to the user, including:
23. The program according to supplementary note 22, wherein:
(Appendix 24)
On the computer,
Play songs included in the playlist presented to the user,
When the operation information for skipping the reproduction is input during the reproduction of the song, the playlist is set such that the feature amount of the skipped song is out of the range of the feature amount of the song to be registered in the playlist. Correcting the range of the feature amount, causing the process to be further executed,
18. The program according to supplementary note 18, wherein:
(Appendix 25)
The process of modifying the range of the feature amount of the playlist includes a process of reducing the range of the feature amount of the playlist.
25. The program according to supplementary note 24, wherein:
(Supplementary Note 26)
The process of modifying the range of the feature amount of the playlist includes:
After reducing the range of the feature amount of the playlist, expanding the range of the feature amount of the playlist in a manner that does not include the feature amount of the skipped song.
25. The program according to supplementary note 24, wherein:
(Appendix 27)
On the computer,
Generating audio data including the name of the playlist to be presented to the user, reproducing the audio data and presenting the name of the playlist to the user, and further executing a process.
18. The program according to supplementary note 18, wherein:
(Appendix 28)
Computer
Data narrowing down is performed based on attribute information indicating attributes of data to be registered in the playlist included in the playlist generation request from the user,
Generating a requested playlist based on a feature amount representing an image of the narrowed-down data, and a range of the feature amount corresponding to the image of the playlist included in the generation request;
When the feature amount of the data registered in the requested playlist is biased within the range of the feature amount of the generated requested playlist, based on the feature amount of the data registered in the requested playlist. Generating a playlist name to be presented to the user;
A play list generation method, characterized by performing a process.

Reference Signs List 1 playlist generation apparatus 4 playlist reproduction system 5 music reproduction apparatus 6 server 7 communication network 8, 8A, 8B portable electronic device 10 computer 11 microphone 12 speaker 101 request playlist input section 102, 603 keyword / attribute extraction section 103, 604 music narrowing down unit 104, 605 recommended playlist candidate generation unit 105, 606 priority calculation unit 105a recognition shift amount calculation unit 105b request shift amount calculation unit 105c playlist search unit 105d shift amount integration unit 106, 607 playlist determination unit 107, 608 Narration creator 107a, 608a Narration text creator 107b, 508 Narration audio creator 108, 509, 609 Playlist playback unit 109, 502 Audio output unit 115 Skip operation input unit 116 Adaptation processing unit 1 6a playlist range modification unit 116b playlist DB re-learning section 116c (of correction data) (data for the playlist learning) database 116d database 121,621 keyword DB
122,622 Refinement attribute DB
123,623 Song DB
124,624 Playlist DB
501 Voice input unit 503 Signal processing unit 504, 510, 601, 801 Communication unit 506 Data input unit 507 Music voice data registration unit 602 Voice recognition unit 521 Database (of music ID and music voice data) 821 Database (of music voice data) 1001 CPU
1002 main storage device 1003 auxiliary storage device 1004 input device 1005 display device 1006 interface device 1007 storage medium drive device 1008 communication device

Claims (12)

Computer
The music is narrowed down based on the attribute information indicating the attribute of the music to be registered in the playlist included in the playlist generation request from the user,
A feature amount representing an image of the narrowed down song and a range in which the feature amount of the song to be registered in the playlist, which is a range associated with each playlist on the database, are defined. Based on the range of the playlist that is, a playlist in which the feature amount representing the image is registered as a song included in the range is generated as a recommended playlist candidate,
The distribution of the feature amount of the song registered in the generated recommended playlist candidate is biased with respect to the range assumed for a requested playlist that is a playlist requested to be generated by the generation request. The distribution of the songs included in the recommended playlist candidate, the range assumed for the recommended playlist candidate, and the range assumed for the requested playlist. Determining a recommended playlist that is a playlist to be presented to the user from the recommended playlist candidates;
A playlist generation method, characterized by performing a process. In the process of determining the recommended playlist, the computer,
A difference between the distribution of the music included in the recommended playlist candidate and the distribution assumed for the recommended playlist candidate is calculated as a recognition error amount, and the difference is assumed for the recommended playlist candidate. the difference between the distributions assumed for the request playlist and the distribution is calculated as the required shift amount,
A weighted sum of the recognition shift amount and the request shift amount for each of the recommended playlist candidates is calculated as a recommendation priority for each of the recommended playlist candidates,
The recommended playlist candidate having the highest recommendation priority is determined as the recommended playlist.
The playlist generation method according to claim 1, wherein: The computer calculates the recognition shift amount and the required shift amount with the distribution of the feature amount of the song as a normal distribution,
3. The playlist generation method according to claim 2, wherein: The computer calculates the recognition shift amount and the required shift amount as a uniform distribution of the feature amount of the music,
3. The playlist generation method according to claim 2, wherein: Said computer,
Play songs included in the recommended playlist,
When operation information for skipping playback is input during the playback of the song, the feature amount of the skipped song is outside the range of the feature amount of the song to be registered in the recommended playlist. Modifying the range assumed for the recommended playlist, further executing a process,
The playlist generation method according to claim 1, wherein: In the process of modifying the range of the recommended playlist, the computer reduces the range of the recommended playlist,
The playlist generation method according to claim 5, wherein: In the process of correcting the range for the recommended playlist, the computer may include:
After reducing the range for the recommended playlist,
Expanding the range of the recommended playlist in a manner that does not include the feature amount of the skipped song;
The playlist generation method according to claim 5, wherein: Said computer,
Generating audio data including the name of the recommended playlist, reproducing the audio data and presenting the name of the recommended playlist to the user, further executing a process;
The playlist generation method according to claim 1, wherein: A song narrowing unit that narrows down songs based on attributes of songs to be registered in the playlist included in the playlist generation request from the user;
A feature amount representing an image of the narrowed down song and a range in which the feature amount of the song to be registered in the playlist, which is a range associated with each playlist on the database, are defined. A recommended playlist candidate generating unit that generates a playlist in which the feature amount representing the image is registered as a recommended playlist candidate based on the range of the playlist that is included,
The distribution of the feature amount of the song registered in the generated recommended playlist candidate is biased with respect to the range assumed for a requested playlist that is a playlist requested to be generated by the generation request. The distribution of the songs included in the recommended playlist candidate, the range assumed for the recommended playlist candidate, and the range assumed for the requested playlist. A recommended playlist determining unit that determines a recommended playlist that is a playlist to be presented to the user from the recommended playlist candidates.
A playlist generation device, characterized in that: The recommended playlist determination unit,
A difference between the distribution of the song included in the recommended playlist candidate and the distribution assumed for the recommended playlist candidate, a recognition shift amount calculation unit that calculates the recognition shift amount,
A difference between the distribution assumed for the recommended playlist candidate and the distribution assumed for the requested playlist, a request shift amount calculation unit that calculates the request shift amount,
A shift amount integration unit that calculates a weighted sum of the recognition shift amount and the request shift amount for each of the recommended playlist candidates, as a recommendation priority for each of the recommended playlist candidates,
The recommended playlist candidate having the highest recommendation priority is determined as the recommended playlist.
The playlist generation device according to claim 9, wherein The music is narrowed down based on the attribute information indicating the attribute of the music to be registered in the playlist included in the playlist generation request from the user,
A feature amount representing an image of the narrowed down song and a range in which the feature amount of the song to be registered in the playlist, which is a range associated with each playlist on the database, are defined. Based on the range of the playlist that is, a playlist in which the feature amount representing the image is registered as a song included in the range is generated as a recommended playlist candidate,
The distribution of the feature amount of the song registered in the generated recommended playlist candidate is biased with respect to the range assumed for a requested playlist that is a playlist requested to be generated by the generation request. The distribution of the songs included in the recommended playlist candidate, the range assumed for the recommended playlist candidate, and the range assumed for the requested playlist. Determining a recommended playlist that is a playlist to be presented to the user from the recommended playlist candidates;
A program that causes a computer to execute processing. Computer
The data is narrowed down based on the attribute information indicating the attribute of the data to be registered in the list included in the list generation request from the user,
It defines a feature amount representing an image of the narrowed-down data and a range that is associated with each list on the database and in which the feature amount of data to be registered in the list is distributed. Based on the range of the list, a list in which the feature amount representing the image is registered with data included in the range is generated as a recommendation list candidate,
When the distribution of the feature amount of the data registered in the generated recommendation list candidate is biased with respect to the range assumed for a request list that is a list requested to be generated by the generation request. On the basis of the distribution of the data included in the recommendation list candidate, the range assumed for the recommendation list candidate, and the range assumed for the request list, Determining a recommendation list that is a list to be presented to the user;
A list generation method, characterized by performing a process.

Images