CN108334617A - The method of semantic-based music retrieval - Google Patents
The method of semantic-based music retrieval Download PDFInfo
- Publication number
- CN108334617A CN108334617A CN201810123009.XA CN201810123009A CN108334617A CN 108334617 A CN108334617 A CN 108334617A CN 201810123009 A CN201810123009 A CN 201810123009A CN 108334617 A CN108334617 A CN 108334617A
- Authority
- CN
- China
- Prior art keywords
- music
- semantic
- user
- vector
- semantic vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 20
- 230000003993 interaction Effects 0.000 claims description 16
- 238000013480 data collection Methods 0.000 claims description 9
- 238000013527 convolutional neural network Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 3
- 239000012141 concentrate Substances 0.000 claims 1
- 230000002452 interceptive effect Effects 0.000 abstract description 17
- 230000000694 effects Effects 0.000 abstract description 5
- 238000002372 labelling Methods 0.000 description 7
- 238000013461 design Methods 0.000 description 5
- 238000011160 research Methods 0.000 description 5
- 230000008451 emotion Effects 0.000 description 3
- 230000002996 emotional effect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 239000013589 supplement Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000036651 mood Effects 0.000 description 1
- 238000000465 moulding Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001502 supplementing effect Effects 0.000 description 1
- 230000000699 topical effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/68—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/686—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, title or artist information, time, location or usage information, user ratings
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/60—Information retrieval; Database structures therefor; File system structures therefor of audio data
- G06F16/68—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/683—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Library & Information Science (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Artificial Intelligence (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The method of semantic-based music retrieval belongs to cloud computing big data field, is for solving the problems, such as：In Chinese songs data set, still lacking effective automation semantic tagger and interactive retrieval, technical essential is：Audio content is mapped in the semantic feature space based on label from original audio low-level image feature using semantic tagger model.Effect is：The system can more natural, accurate acquisition user search be intended to, so that user is obtained preferably retrieval experience, user facilitated to find and find that " thinking " wants song.
Description
Technical field
The present invention relates to a kind of searching systems, more particularly to the music retrieval system based on semantic data.
Background technology
The semanteme realized in music, such as mood, style user's subjective feeling are that people is encouraged to go to listen to and pursue
One of an important factor for music.And as music high-level semantics feature, the identification and semantic-based retrieval of music semanteme, always
It is the difficult point and hot spot of music information retrieval (Music Information Retrieval, MIR) area research person research.Sound
" wide gap " between happy bottom physical features and high-level semantics features once perplexs and restricts content-based music retrieval and grind
The development studied carefully.Go deep into the development of technology with research, researcher establishes between musical features and semanteme by various modes
Contact, gradually reduces wide gap.Many research work such as carry out emotional semantic classification based on audio frequency characteristics, based on music lyrics auxiliary
Emotional semantic classification, audio frequency characteristics are combined with lyrics text, the emotional semantic classification etc. of audio frequency characteristics lyrics text and socialization mark so that
Music emotion classification has been achieved for preferable effect at present.Music semantic tagger system is embodied as having supervision by some systems
Multi-tag labeling system (Supervised Multi-class Labeling, SML) so that based on retrieving for music semanteme
To realize, people can retrieve the music for meeting certain semantic label.But semantic cognition often varies with each individual, and has very strong
Subjectivity and personalization features.In addition, similar musical database focuses mostly in western music, rarely has completion to Chinese songs
Automate the musical database of semantic tagger.
Invention content
In order to solve audio content being mapped to the problems in the semantic feature space based on label, the present invention proposes as follows
Technical solution：
A kind of method of semantic-based music retrieval, includes the following steps：Music retrieval；Music Semantic interaction and music
Recommend.
Further, the method for the music retrieval is as follows：Music data collection is marked, per song is by table in the data set
The feature vector of a dimension is shown as according to collection；According to convolutional neural networks structure, using between feature vector in known mark data set
Relationship train network, obtain network parameter；Using trained network model, using per song in data set to be retrieved as
Input obtains network output；It is exported according to network, calculates the semantic vector of per song in data set to be retrieved, obtain semanteme
Vector set；It is inputted example music as convolutional neural networks, obtains network output valve, the semantic vector of sample calculation music；
Calculate the COS distance between semantic vector collection and example music and example music；Enable in semantic space with example music most phase
As music list be COS distance minimum preceding n songs set；Output is most like with example music in semantic space
Music list.
The method of the music Semantic interaction is as follows：User's currently playing music is input to trained convolutional network,
Obtain network output；Calculate the semantic vector of user's currently playing music：For the per song that mark music data is concentrated, calculate
Semantic vector, and obtain semantic vector collection；Using k-means algorithms to semantic vector clustering, is concentrated and sung according to music data
Number of tracks amount determines classification number, chooses the central set of each classification into semantic vector core set；It calculates and semantic vector core set
In each vector COS distance, obtain semantic most like music semantic vector；All songs row in classification where returning
Table is as recommendation music album.
Further, the method that the music is recommended is as follows：If the music record set that user is played contains content, newly
User semantic configuration file is built, UID, user semantic vector are inserted into, user semantic vector length is semantic label number；If with
The music record set that family is played contains content, and therein one first song semantics vector is taken to sum with user semantic vector step-by-step,
And it is assigned to the semantic vector of user；Threshold value T is set, and the c obtained value that will add up from the semantic vector of user sorts, and comes
Top-T labels are set to 1, remaining label position is 0；According to UID, corresponding user's language is updated using the semantic vector of user
Adopted configuration file；To the label in music semantic vector, the number of user's broadcasting is had collected, and label is arranged according to number is listened to
Sequence sets the label to rank in the top to the point of interest of user, is stored in user semantic configuration file, utilizes configuration text
Part recommends semantic similar music album for user.
Advantageous effect：Audio content can be mapped to the semantic feature based on label by using semantic tagger model
In space, it is intended to which the extraction across acoustic feature is difficult, finds contacting between original signal and semanteme.Meanwhile to music number
Automatic marking also is completed using trained model according to the music crawled in library, makes retrieval example with musical database in language
The comparison that similarity is carried out in adopted space obtains semantic relevant music, good user interface is designed, by semantic information
It gives full expression to user, while user being facilitated to make corresponding modification to semantic meaning representation mode, individual is expressed to some convenient for user
The subjective feeling of music.
Description of the drawings
Fig. 1 is music retrieval page figure；
Fig. 2 is Semantic interaction page figure；
Fig. 3 is that music recommends page figure；
Fig. 4 is semantic label and its classification figure；
The semantic-based music retrieval interaction models frame diagrams of Fig. 5.
Specific implementation mode
The present invention provides a kind of music retrieval method based on semantic data collection, the effect of the method is, it can be with
By using semantic tagger model, audio content is mapped in the semantic feature space based on label, it is intended to special across acoustics
The extraction of sign is difficult, finds contacting between original signal and semanteme.Meanwhile to the music crawled in musical database
Automatic marking is completed using trained model, retrieval example and musical database is made to carry out the ratio of similarity in semantic space
Compared with the semantic relevant music of acquisition designs good user interface, semantic information is given full expression to user, while side
Just user makes corresponding modification to semantic meaning representation mode, and the personal subjective feeling to some music is expressed convenient for user.It preserves and uses
Family is conducive to extend Chinese music data collection, and then optimize to have and supervise labeling algorithm efficiency, devise to the markup information of song
User semantic configuration file, for music recommendation and query expansion.The system being capable of more natural, accurate acquisition user search meaning
Figure makes user obtain preferably retrieval experience, the song for facilitating user to find and find that " thinking " is wanted.
A kind of semantic-based music retrieval method includes mainly what music retrieval, music Semantic interaction and music were recommended
Step, system interface design and realization result are as follows：
Music retrieval：
On the basis of automating dimensioning algorithm according to the music semanteme of convolutional neural networks, further design a set of based on language
The music retrieval algorithm of justice, as shown in algorithm 1：
Algorithm 1:Music retrieval algorithm based on exemplary semantic
Input:Music data collection is marked, per song is expressed as the feature vector of a dimension according to collection in the data set；
Output:The music list most like with example music in semantic space；
Algorithm description：
According to designed convolutional neural networks structure, trained using the relationship between feature vector in known mark data set
Network obtains network parameter；
Using trained network model network output is obtained using per song in data set to be retrieved as input；
It is exported according to network, calculates the semantic vector of per song in data set to be retrieved, obtain semantic vector collection；
It is inputted example music as convolutional neural networks, obtains network output valve, the semantic vector of sample calculation music；
Calculate the COS distance between semantic vector collection and example music and example music；
It is the collection of the preceding n songs of COS distance minimum to enable music list most like with example music in semantic space
It closes；
Output music list most like with example music in semantic space.
Music Semantic interaction
In order to make full use of customer interaction information, the proposed algorithm based on interactive information is designed, the algorithm, Ke Yigen are passed through
The music information being currently played according to user calculates its semantic vector.Meanwhile collecting the language that user inputs on interactive interface
Adopted label forms semantic label collection.To the music that labeled data is concentrated, semantic vector concentration is mapped that, to the vector set
In music, it is clustered using k-means algorithms.According to the musical works quantity in data set, it is 20 first to be polymerized to size
The semantic vector of each class center is put together composition semantic vector core set, to reduce similarity by the classification of song
The number compared improves system feedback speed.Finally, by semantic vector core set, the shortest semantic vector pair of COS distance
The music list answered recommends album as music.As shown in algorithm 2
Algorithm 2:Music recommendation algorithm based on interactive information
Input:User's currently playing music, marks music data collection, and user interactively enters tally set；
Output:Semantic similar music album；
Algorithm description:
1. user's currently playing music is input to trained convolutional network, network output is obtained；
2. calculating the semantic vector of user's currently playing music；
3. for the per song that mark music data is concentrated, semantic vector is calculated, and obtain semantic vector collection；
4. using k-means algorithms to semantic vector clustering, number of songs is concentrated to determine classification according to music data
Number, chooses the central set of each classification into semantic vector core set；
5. calculate the COS distance with each vector in semantic vector core set, obtain semantic most like music semanteme to
Amount；
6. all list of songs in classification where returning are as recommendation music album.
Music is recommended
There is inseparable relationship between user and music, label, if using music as tie, it is established that user and semanteme
Relationship between label, so that it may with for finding that user's listens to custom and interest in music, therefore, the present invention devises user's language
The generating algorithm of adopted configuration file, as shown in algorithm 3.
Algorithm 3:User semantic configuration file generating algorithm
Input:The music record set that user's registration UID, user are played；
Output:User semantic configuration file
Algorithm description:
If the music record set that user is played contains content, user semantic configuration file is created, is inserted into UID, user
Semantic vector Su≡<0,0 ..., 0>, user semantic vector length is semantic label number c；
2. if the music record set that user is played contains content, take therein one first song semantics vector and user's language
Adopted vector step-by-step summation, and it is assigned to the semantic vector of user；
3. threshold value T is arranged, the c obtained value that will add up from the semantic vector of user sorts, and comes Top-T labels
It is set to 1, remaining label position is 0.
4. according to UID, corresponding user semantic configuration file is updated using the semantic vector of user
By the algorithm, to the label in music semantic vector, the number of user's broadcasting is had collected, and label is according to listening to
Number sorts, and sets the label to rank in the top to the point of interest of user, is stored in user semantic configuration file.It utilizes
Configuration file, you can recommend semantic similar music album for user.
Retrieve the foundation of data set
Used CAL500 data sets, 500 first different artistical works contained in the data set, per first song at least by
Three people independently mark.It includes 174 labels to mark in set of words, if used by 80% or more people (or at least two people) per first song
Some word marked, then was 1 in label-vector intermediate value, otherwise value takes 0.Due to retrieving lacking for relevant technology with Chinese songs
It is weary, this patent in order to make system be suitable for Chinese songs use environment, to Chinese music and its label information crawl and
Supplement, is as follows
(1) Chinese songs data crawl
Program is designed to the information crawler of Baidu music based on python third party libraries, crawls linkhttp:// music.baidu.com/tag, it is as follows to crawl process：1. fromhttp://music.baidu.com/tagThe page obtains all
The corresponding url of label and classification.2. traverses the song url obtained under each label url, such as the url of traversal label one by one
=" http://music.baidu.com/tag/%E6%96%B0%E6%AD%8C ", under song.(subsequent url is compiled
It is " new song " after code decoding, it is all songs under new song to be equivalent to traversal label, "http://music.baidu.com/ tag/HYPERLINK"http://music.baidu.com/tag/ newly sings " HYPERLINK " http:// Music.baidu.com/tag/ newly sings " HYPERLINK " http:It is " new to sing that //music.baidu.com/tag/ is newly sung")。③.
All song url are traversed, song information is obtained, file, all_ are written with fixed character string format<Tag name>.txt it is all
The information of song, each all information of song correspond to a line.4. end of runs have crawled the song information under all labels.Journey
Following file can be generated after sort run:1. what is deposited under .file files is each theme：Such as scene, style.2. themes
The tag name for being under file：It is each label pair to have after 80s, DJ dance musics and url_list.txt under such as topical subject
The url answered.3. that tag name files are deposited is the lyrics (.lrc files) and all_<Tag name>.txt it is the letter of all songs
Breath.
(2) label selects
In order to merge Chinese and English data set, Chinese and English mark label is selected, is first divided into label：
Scape, school, musical instrument, four major class of emotion.From 174 labels of CAL500 data sets, semantic repetition values are rejected, and crawled
The label semanteme correspondence of Chinese seeks common ground, and finally selectes 33 bilingual semantic labels, labeling and correspondence are such as
Shown in Fig. 4.
(3) semantic vector collection generates
Selection according to semantic label and mark situation, calculate semantic vector and data can be obtained according to label for labelling situation
The corresponding semantic vector collection of Chinese and English song is concentrated, obtained semantic vector collection is divided into training set and test set is used for convolutional Neural
The training and test of network.Database purchase label classification, label and song mark situation has also been devised in this patent, and can be with
When generative semantics vector set needed according to algorithm.
Strategy is crawled according to label selection and Chinese songs, crawl and arranges Chinese songs 1000 is first.With CAL500 data
After the English song of concentration merges, length is removed or song that label is not inconsistent, Chinese and English song totally 1483 is included in data set.
According to music recommendation algorithm, above-mentioned music is clustered, is formed and recommends album 70.
Music retrieval：As shown in Fig. 1, in search interface, user can be sought by mobile slider bar in music libraries
Look for qualified music.If " pure and fresh value " is larger, then the music style searched out can be biased to pure and fresh, nature.Certainly, system
It can be obtained according to user to the situation of movement comprehensive search of " pure and fresh value ", " sentiment value ", " sunlight value ", " ardent value " slider bar
The music that user wants.
Music Semantic interaction interface：As most important interactive interface, is devised in system and meet music semantic label table
The interactive interface reached is as shown in Figure 2.Wherein, music scenario is expressed as broadcast window background color, and selects and usage scenario language
The color that justice is consistent is corresponding to it, such as romantic wedding scene is characterized with purple.Music style uses one group of performance not unisonance
Personage's picture of happy style indicates that, when music belongs to different-style, the character moulding in broadcast window can occur to change accordingly
Become.Musical instrument label in music uses the musical instrument icon representation in broadcast window small circle.Meanwhile the feelings that the song is shown
Sense corresponding can show as selected state in emotion button area, and also use different colours and show corresponding feelings
Feel information, such as blue indicates quiet, red and indicates passion.The interface using comprising for the use of three, first when user chooses
When a piece of music plays, system obtains the semantic vector of the song from server end, parsing song all by which label for labelling mistake,
The semantic form of expression on interface is corresponded to according to label, is presented to user.Secondly, when user is discontented with the semanteme that system provides
Or when supplementing, semantic meaning representation control that can be in arbitrary selection interface, preserve oneself chooses server end.Server
End receives user to the markup information of song, is stored in the interim table of database, and judges to be marked by different user when a first song
When number is more than certain value, it is written in song markup information table.Finally, user changes on interactive interface or selected semanteme is controlled
After part, when exiting broadcast interface, system receives user semantic control value, changes into semantic vector, according to music interaction proposed algorithm,
Include recommending on interface in music by recommendation results.
Music based on interactive information is recommended：After system receives the semantic information that user is submitted by interactive interface, it understand oneself
Dynamic generative semantics vector is as example is recommended, according to music interaction proposed algorithm, selection and exemplary semantic most phase from data set
As music album be shown to user, recommend album interface as shown in Figure 3 in addition, system has been designed and Implemented user's registration, stepped on
Recording function, login user will be generated for user according to user semantic configuration file generating algorithm and preserve user semantic configuration text
In part to database.When user logs on, can be examined in data set according to the semantic vector in user semantic configuration file
Album most like Suo Yuyi recommends in homepage.
The present invention realizes a semantic-based interactive music search method, and innovation is with convolutional Neural
The mapping from music content to music semantic vector space is realized based on network algorithm, is devised with the friendship for enriching semanteme
Mutual interface is used for the expression of music semanteme.It devises based on interactive music retrieval and proposed algorithm, and generates user semantic
Configuration file.It can be converted to semantic vector according to relevant operation of the user on interactive interface, be retrieved in data set semantic
Similar music, or recommend for user to meet the relevant musical album of user semantic.It crawls and has arranged Chinese and English data set, pass through
Interactive system realizes the update to labeled data collection and supplement, to strengthen the scalability of data set.The prototype system
Realization, to semantic-based music retrieval algorithm research and develop music searching engine, have stronger researching value and reality
With value.
In another embodiment, a kind of music retrieval exchange method based on semantic data collection, including：
S1, by the way that from original audio low-level image feature, such as common MFCC coefficients set out, using semantic tagger model, by sound
In frequency content map to the semantic feature space based on label, it is intended to which the extraction across acoustic feature is difficult, finds original signal
Contacting between semanteme.Meanwhile the music crawled in musical database is also completed using trained model automatic
Mark carries out the comparison of similarity by retrieving example and musical database in semantic space, obtains semantic relevant music,
Good user interface is designed, semantic information is given full expression to user, while user being facilitated to do semantic meaning representation mode
Corresponding modification expresses the personal subjective feeling to some music convenient for user.Markup information of the user to song is preserved, favorably
In the Chinese music data collection of extension, and then optimize to have and supervise labeling algorithm efficiency, devises user semantic configuration file, be used for sound
Happy recommendation and query expansion.The system can more natural, accurate acquisition user search be intended to, so that user is obtained preferably retrieval
Experience, the song for facilitating user to find and find that " thinking " is wanted.
Music retrieval interaction designs of the S2 based on exemplary semantic：In the present invention, program can be according to designed convolution
Neural network structure obtains semantic feature vector, while the data concentrated using labeled data are available and improve interactive information,
Interactive information obtains interaction results by interaction retrieval and proposed algorithm.
Music recommendation algorithms of the S3 based on interactive information：The present invention devises the proposed algorithm based on interactive information, passes through
The algorithm, the music information that can be currently played according to user, calculates its semantic vector.Meanwhile by user on interaction circle
The semantic label inputted on face collects, and forms semantic label collection.To the music that labeled data is concentrated, semanteme is mapped that
In vector set, to the music in the vector set, it is clustered using k-means algorithms.According to the musical works in data set, gather
At different classifications, each classification has 20 songs.It is semantic that the semantic vector of each class center is put together into composition
Vectorial core set improves system feedback speed to reduce the number of similarity-rough set.Finally, by semantic vector core set,
The corresponding music list of the shortest semantic vector of COS distance recommends album as music, is expeditiously provided for client with reaching
The effect of required music.
S4 user semantic configuration files generate：There are inseparable relationship, this patent design between user and music, label
The generating algorithm of user semantic configuration file；Using music as tie, it is established that the relationship between user and semantic label is used for
It was found that user's listens to custom and interest in music, to bring better experience to user.
The preferable specific implementation mode of the above, only the invention, but the protection domain of the invention is not
It is confined to this, any one skilled in the art is in the technical scope that the invention discloses, according to the present invention
The technical solution of creation and its inventive concept are subject to equivalent substitution or change, should all cover the invention protection domain it
It is interior.
Claims (4)
1. a kind of method of semantic-based music retrieval, which is characterized in that include the following steps：Music retrieval；
Music Semantic interaction and
Music is recommended.
2. the method for semantic-based music retrieval as described in claim 1, which is characterized in that the method for the music retrieval
It is as follows：
Music data collection is marked, per song is expressed as the feature vector of a dimension according to collection in the data set；
According to convolutional neural networks structure, network is trained using the relationship between feature vector in known mark data set, obtains net
Network parameter；
Using trained network model network output is obtained using per song in data set to be retrieved as input；
It is exported according to network, calculates the semantic vector of per song in data set to be retrieved, obtain semantic vector collection；
It is inputted example music as convolutional neural networks, obtains network output valve, the semantic vector of sample calculation music；
Calculate the COS distance between semantic vector collection and example music and example music；
It is the set of the preceding n songs of COS distance minimum to enable music list most like with example music in semantic space；
Output music list most like with example music in semantic space.
3. the method for semantic-based music retrieval as described in claim 1, which is characterized in that the music Semantic interaction
Method is as follows：
User's currently playing music is input to trained convolutional network, obtains network output；
Calculate the semantic vector of user's currently playing music：For the per song that mark music data is concentrated, semantic vector is calculated,
And obtain semantic vector collection；
Using k-means algorithms to semantic vector clustering, concentrates number of songs to determine classification number according to music data, choose
The central set of each classification is at semantic vector core set；
The COS distance with each vector in semantic vector core set is calculated, semantic most like music semantic vector is obtained；
All list of songs in classification where returning are as recommendation music album.
4. the method for semantic-based music retrieval as described in claim 1, which is characterized in that the method that the music is recommended
It is as follows：
If the music record set that user is played contains content, user semantic configuration file is created, is inserted into UID, user semantic
Vectorial Su≡<0,0 ..., 0>, user semantic vector length is semantic label number c；
If the music record set that user is played contains content, therein one first song semantics vector and user semantic vector are taken
Step-by-step is summed, and is assigned to the semantic vector of user；
Threshold value T is set, the c obtained value that will add up from the semantic vector of user sorts, and comes Top-T labels and is set to 1,
Remaining label position is 0；
According to UID, corresponding user semantic configuration file is updated using the semantic vector of user；
To the label in music semantic vector, the number of user's broadcasting is had collected, and label sorts according to number is listened to, and will come
Former labels are set as the point of interest of user, are stored in user semantic configuration file, are user using configuration file
Recommend semantic similar music album.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810123009.XA CN108334617A (en) | 2018-02-07 | 2018-02-07 | The method of semantic-based music retrieval |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810123009.XA CN108334617A (en) | 2018-02-07 | 2018-02-07 | The method of semantic-based music retrieval |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108334617A true CN108334617A (en) | 2018-07-27 |
Family
ID=62927117
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810123009.XA Pending CN108334617A (en) | 2018-02-07 | 2018-02-07 | The method of semantic-based music retrieval |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108334617A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109299321A (en) * | 2018-08-31 | 2019-02-01 | 出门问问信息科技有限公司 | A kind of song recommended method and device |
| CN109726310A (en) * | 2018-11-15 | 2019-05-07 | 量子云未来（北京）信息科技有限公司 | A kind of determination method, apparatus and storage medium for recommending music track |
| CN109857900A (en) * | 2019-02-14 | 2019-06-07 | 腾讯音乐娱乐科技（深圳）有限公司 | A kind of similar songs recommended method and relevant device |
| CN110209869A (en) * | 2018-08-13 | 2019-09-06 | 腾讯科技（深圳）有限公司 | A kind of audio file recommended method, device and storage medium |
| CN110399522A (en) * | 2019-07-03 | 2019-11-01 | 中国传媒大学 | A kind of music singing search method and device based on LSTM and layering and matching |
| WO2020082404A1 (en) * | 2018-10-25 | 2020-04-30 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for recalling points of interest using a tagging model |
| GB2583455A (en) * | 2019-04-03 | 2020-11-04 | Mashtraxx Ltd | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| GB2584598A (en) * | 2019-04-03 | 2020-12-16 | Mashtraxx Ltd | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| CN112307254A (en) * | 2020-11-24 | 2021-02-02 | 腾讯科技（深圳）有限公司 | Vector determination method of music label and related device |
| US11068782B2 (en) | 2019-04-03 | 2021-07-20 | Mashtraxx Limited | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| CN113569086A (en) * | 2021-08-05 | 2021-10-29 | 深圳墨世科技有限公司 | Quadrature library aggregation method and device, terminal equipment and readable storage medium |
| CN113643720A (en) * | 2021-08-06 | 2021-11-12 | 腾讯音乐娱乐科技（深圳）有限公司 | Song feature extraction model training method, song identification method and related equipment |
| US11544565B2 (en) | 2020-10-02 | 2023-01-03 | Emotional Perception AI Limited | Processing system for generating a playlist from candidate files and method for generating a playlist |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102890713A (en) * | 2012-09-20 | 2013-01-23 | 浙江大学 | Music recommending method based on current geographical position and physical environment of user |
| CN103324691A (en) * | 2013-06-03 | 2013-09-25 | 河海大学 | Voice frequency searching method based on M-tree |
| CN104035917A (en) * | 2014-06-10 | 2014-09-10 | 复旦大学 | Knowledge graph management method and system based on semantic space mapping |
| US20160196580A1 (en) * | 2015-01-02 | 2016-07-07 | Sk Planet Co., Ltd. | System for content recommendation service, content recommendation device and method of operating the same |
-
2018
- 2018-02-07 CN CN201810123009.XA patent/CN108334617A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102890713A (en) * | 2012-09-20 | 2013-01-23 | 浙江大学 | Music recommending method based on current geographical position and physical environment of user |
| CN103324691A (en) * | 2013-06-03 | 2013-09-25 | 河海大学 | Voice frequency searching method based on M-tree |
| CN104035917A (en) * | 2014-06-10 | 2014-09-10 | 复旦大学 | Knowledge graph management method and system based on semantic space mapping |
| US20160196580A1 (en) * | 2015-01-02 | 2016-07-07 | Sk Planet Co., Ltd. | System for content recommendation service, content recommendation device and method of operating the same |
Non-Patent Citations (1)
| Title |
|---|
| 秦静等: "基于示例语义的音乐检索模型", 《山东大学学报（理学版）》 * |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110209869A (en) * | 2018-08-13 | 2019-09-06 | 腾讯科技（深圳）有限公司 | A kind of audio file recommended method, device and storage medium |
| CN110209869B (en) * | 2018-08-13 | 2023-03-21 | 腾讯科技（深圳）有限公司 | Audio file recommendation method and device and storage medium |
| CN109299321A (en) * | 2018-08-31 | 2019-02-01 | 出门问问信息科技有限公司 | A kind of song recommended method and device |
| US11093531B2 (en) | 2018-10-25 | 2021-08-17 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for recalling points of interest using a tagging model |
| WO2020082404A1 (en) * | 2018-10-25 | 2020-04-30 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for recalling points of interest using a tagging model |
| CN111191107A (en) * | 2018-10-25 | 2020-05-22 | 北京嘀嘀无限科技发展有限公司 | System and method for recalling points of interest using annotation model |
| CN111191107B (en) * | 2018-10-25 | 2023-06-30 | 北京嘀嘀无限科技发展有限公司 | System and method for recalling points of interest using annotation model |
| CN109726310A (en) * | 2018-11-15 | 2019-05-07 | 量子云未来（北京）信息科技有限公司 | A kind of determination method, apparatus and storage medium for recommending music track |
| CN109857900A (en) * | 2019-02-14 | 2019-06-07 | 腾讯音乐娱乐科技（深圳）有限公司 | A kind of similar songs recommended method and relevant device |
| CN109857900B (en) * | 2019-02-14 | 2024-02-09 | 腾讯音乐娱乐科技（深圳）有限公司 | Similar song recommendation method and related equipment |
| GB2584598A (en) * | 2019-04-03 | 2020-12-16 | Mashtraxx Ltd | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| US11080601B2 (en) | 2019-04-03 | 2021-08-03 | Mashtraxx Limited | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| US11068782B2 (en) | 2019-04-03 | 2021-07-20 | Mashtraxx Limited | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| GB2584598B (en) * | 2019-04-03 | 2024-02-14 | Emotional Perception Ai Ltd | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| US11494652B2 (en) | 2019-04-03 | 2022-11-08 | Emotional Perception AI Limited | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| US11645532B2 (en) | 2019-04-03 | 2023-05-09 | Emotional Perception AI Limited | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| GB2583455A (en) * | 2019-04-03 | 2020-11-04 | Mashtraxx Ltd | Method of training a neural network to reflect emotional perception and related system and method for categorizing and finding associated content |
| CN110399522A (en) * | 2019-07-03 | 2019-11-01 | 中国传媒大学 | A kind of music singing search method and device based on LSTM and layering and matching |
| US11977845B2 (en) | 2020-10-02 | 2024-05-07 | Emotional Perception AI Limited | System and method for evaluating semantic closeness of data files |
| US11544565B2 (en) | 2020-10-02 | 2023-01-03 | Emotional Perception AI Limited | Processing system for generating a playlist from candidate files and method for generating a playlist |
| CN112307254A (en) * | 2020-11-24 | 2021-02-02 | 腾讯科技（深圳）有限公司 | Vector determination method of music label and related device |
| CN112307254B (en) * | 2020-11-24 | 2024-04-02 | 腾讯科技（深圳）有限公司 | Vector determination method and related device for music tag |
| CN113569086B (en) * | 2021-08-05 | 2024-01-26 | 深圳墨世科技有限公司 | Method, device, terminal equipment and readable storage medium for aggregating curved libraries |
| CN113569086A (en) * | 2021-08-05 | 2021-10-29 | 深圳墨世科技有限公司 | Quadrature library aggregation method and device, terminal equipment and readable storage medium |
| CN113643720A (en) * | 2021-08-06 | 2021-11-12 | 腾讯音乐娱乐科技（深圳）有限公司 | Song feature extraction model training method, song identification method and related equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108334617A (en) | The method of semantic-based music retrieval | |
| Schedl et al. | Music recommender systems | |
| Kaminskas et al. | Contextual music information retrieval and recommendation: State of the art and challenges | |
| Kaminskas et al. | Location-aware music recommendation using auto-tagging and hybrid matching | |
| Meyers | A mood-based music classification and exploration system | |
| Bertin-Mahieux et al. | Autotagger: A model for predicting social tags from acoustic features on large music databases | |
| CN101981563B (en) | The method and apparatus for the related content shown is selected in conjunction with media | |
| CN108363769A (en) | The method for building up of semantic-based music retrieval data set | |
| Dhaenens et al. | ‘Press play for pride’: The cultural logics of LGBTQ-themed playlists on Spotify | |
| KR20120101233A (en) | Method for providing sentiment information and method and system for providing contents recommendation using sentiment information | |
| Sarin et al. | SentiSpotMusic: a music recommendation system based on sentiment analysis | |
| Lillie | MusicBox: Navigating the space of your music | |
| Hyung et al. | Utilizing context-relevant keywords extracted from a large collection of user-generated documents for music discovery | |
| Kaminskas et al. | Emotion-based matching of music to places | |
| Kemp | Towards a holistic interpretation of musical genre classification | |
| Baumann et al. | Using natural language input and audio analysis for a human-oriented MIR system | |
| Malleswari et al. | Music recommendation system using hybrid approach | |
| Alakija | Living with difference: Ontological security and identification of second-generation members of the Nigerian diaspora in Peckham,‘Little Lagos’, London | |
| Liu | Everybody’s Song Making: Do-it-yourself with and against Artificial Intelligence | |
| Çano et al. | Crowdsourcing emotions in music domain | |
| He et al. | Digital empirical research of influencing factors of musical emotion classification based on pleasure-arousal musical emotion fuzzy model | |
| Hu | A model-based music recommendation system for individual users and implicit user groups | |
| Turnbull | Design and development of a semantic music discovery engine | |
| Strle et al. | Towards user-aware music information retrieval: emotional and color perception of music | |
| Ji | [Retracted] Classification Method of Emotional Characteristics of Vocal Singing under Multicultural Values |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180727 |