KR101987605B1 - Method and apparatus of music emotion recognition - Google Patents

Abstract

Disclosed are a method and an apparatus for recognizing sensibility of music. According to one embodiment of the present invention, the method for recognizing sensibility of music comprises the following steps: receiving data related to lyrics of the music; generating a vocabulary dictionary based on the data; extracting a feature vector based on a weight corresponding to words included in the vocabulary dictionary and data; and determining sensibility of the music using the feature vector as an input of an artificial neural network.

Description

[0001] METHOD AND APPARATUS OF MUSIC EMOTION RECOGNITION [0002]

The following embodiments relate to a method and apparatus for recognizing music emotion.

Recognition of music emotion is attracting attention in the field of retrieving music information. Music retrieval using emotion is one of the main criteria that users use. The actual music database grows and grows daily, so much work is required to maintain the updates.

However, manually annotating music with emotion tags is somewhat subjective, costly, and time consuming. This problem can be solved through an automatic recognition system.

Early stage music sensibility recognition systems were mostly based on audio content analysis. A bi-modal music sensibility system with improved accuracy by combining audio and lyrics has been developed.

The importance of audio or lyrics depends on the style of the music. For example, in dance music, audio is relevant, and in poetic music, lyrics are the key. Various psychological studies have confirmed the importance of the words to convey semantic information. However, in spite of the importance of housework, conventional research on the perception of housework - based music emotion has limitations.

Embodiments can provide a technique for recognizing music emotion.

According to an embodiment of the present invention, there is provided a music emotion recognition method comprising the steps of: receiving data related to lyrics of music; generating a lexicon dictionary based on the data; and generating a lexicon dictionary and a weight corresponding to words included in the data Extracting a feature vector based on the feature vector, and determining the emotion of the music using the feature vector as an input to the artificial neural network.

Wherein the generating step comprises the steps of: filtering the data based on the type of language; filtering the filtered data based on the part of speech based on the kind of language; And removing the meaningless words from the dictionary to generate the dictionary dictionary.

The step of removing the meaningless word from the filtered data based on the part of speech of the word to generate the lexical dictionary may include removing at least one of numbers, adjectives, alphabets, and related pronouns from the filtered data based on the part of speech And generating the vocabulary dictionary.

Wherein the extracting step comprises the steps of: generating a set of words based on the data; generating a generation vector based on the lexical dictionary and the set of words; And extracting the feature vector by converting the feature vector.

The step of generating the set of words may include dividing words included in the data, and recovering a prototype of the word to generate a set of words.

Wherein the step of extracting the feature vector by converting a component of the occurrence vector based on the weight comprises: calculating a first weight based on the number of words contained in the set of words; Calculating a second weight based on the first weight and the second weight, and extracting the feature vector by transforming the components of the generation vector based on the product of the first weight and the second weight.

The step of calculating the first weight may include calculating the first weight using a TF-IDF (Term Frequency - Inverse Document Frequency).

The step of calculating the second weight may comprise calculating the second weight based on a sigmoid function.

The determining may comprise calculating a probability value corresponding to a plurality of emotion groups using the artificial neural network, and determining emotion of the music based on the probability value.

The artificial neural network is a DBN (Deep Belief Network), and the DBN can be learned using transfer learning.

A music emotion recognition apparatus according to an embodiment of the present invention includes a receiver for receiving data related to lyrics of music, and a lexical dictionary generating unit for generating a lexical dictionary based on the data, based on the lexical dictionary and a weight corresponding to a word included in the data And a processor for extracting a feature vector and using the feature vector as an input to an artificial neural network to determine emotion of the music.

Wherein the processor is configured to filter the data based on the type of language, filter the filtered data based on the part of the word based on the type of the language, and extract a meaningless word from the filtered data based on the part of the word To generate the dictionary dictionary.

The processor may generate the lexical dictionary by removing at least one of a number, an alias, an alphabet, and a relative pronoun from the filtered data based on the part-of-speech.

Wherein the processor is configured to generate a set of words based on the data, generate a generation vector based on the lexical dictionary and the set of words, and convert the components of the generation vector based on the weights, Can be extracted.

The processor may divide words contained in the data and recover a prototype of the word to generate a set of words.

Wherein the processor is further configured to calculate a first weight based on the number of words contained in the set of words and to calculate a second weight based on a nonlinear function according to a predetermined constant and to calculate the first weight and the second weight The feature vector can be extracted by converting the component of the generation vector based on the product of the feature vectors.

The processor may calculate the first weight using a TF-IDF (Term Frequency - Inverse Document Frequency).

The processor may calculate the second weight based on a sigmoid function.

The processor may calculate a probability value corresponding to a plurality of emotion groups using the artificial neural network, and determine emotion of the music based on the probability value.

The artificial neural network is a DBN (Deep Belief Network), and the DBN can be learned using transfer learning.

1 is a schematic block diagram of a music emotion recognition apparatus according to an embodiment.
Fig. 2 shows the overall operation of the music emotion recognition apparatus shown in Fig.
Fig. 3 shows an operation in which the music emotion recognition apparatus shown in Fig. 1 generates a lexical dictionary.
4A shows the distribution of the feature vectors by the TF-IDF weight.
FIG. 4B shows distribution of feature vectors according to weights according to the music emotion recognition apparatus shown in FIG. 1. FIG.
Fig. 5 shows a result of comparison between the recognition accuracy of the conventional art and the music emotion recognition apparatus shown in Fig.
6 is a flowchart of the operation of the music emotion recognition apparatus shown in Fig.

In the following, embodiments will be described in detail with reference to the accompanying drawings. However, various modifications may be made in the embodiments, and the scope of the patent application is not limited or limited by these embodiments. It is to be understood that all changes, equivalents, and alternatives to the embodiments are included in the scope of the right.

The terms used in the examples are used for descriptive purposes only and are not to be construed as limiting. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises" or "having" and the like refer to the presence of stated features, integers, steps, operations, elements, components, or combinations thereof, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

The terms first, second, or the like may be used to describe various elements, but the elements should not be limited by terms. The terms may be named for the purpose of distinguishing one element from another, for example, without departing from the scope of the right according to the concept of the embodiment, the first element being referred to as the second element, The second component may also be referred to as a first component.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this embodiment belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the contextual meaning of the related art and are to be interpreted as either ideal or overly formal in the sense of the present application Do not.

In the following description of the present invention with reference to the accompanying drawings, the same components are denoted by the same reference numerals regardless of the reference numerals, and redundant explanations thereof will be omitted. In the following description of the embodiments, a detailed description of related arts will be omitted if it is determined that the gist of the embodiments may be unnecessarily blurred.

1 is a schematic block diagram of a music emotion recognition apparatus according to an embodiment.

Referring to FIG. 1, the music emotion recognition apparatus 10 can recognize emotion of music. The music sensibility recognition device 10 can recognize the sensitivity of music based on the lyrics of the music. The music sensibility recognition apparatus 10 can recognize the sensitivity of music by analyzing the lyrics of the music using the artificial neural network.

The emotion of music may include the emotion of the person listening to the music. Emotions can mean feelings or feelings that happen to a phenomenon or work. The sensibility of music can mean emotions included in the Russel emotion group. For example, the sensibility of music can include happiness, tense, sad, and relax.

The music sensibility recognition apparatus 10 may be implemented as a printed circuit board (PCB) such as a motherboard, an integrated circuit (IC), or a system on chip (SoC). For example, the music sensibility recognition apparatus 10 may be implemented as an application processor.

The music sensibility recognition apparatus 10 may be implemented in a personal computer (PC), a data server, or a portable device.

Portable devices include laptop computers, mobile phones, smart phones, tablet PCs, mobile internet devices (MIDs), personal digital assistants (PDAs), enterprise digital assistants (EDAs) A digital still camera, a digital video camera, a portable multimedia player (PMP), a personal navigation device or a portable navigation device (PND), a handheld game console, an e-book e-book, or a smart device. A smart device can be implemented as a smart watch, a smart band, or a smart ring.

The music sensibility recognition apparatus 10 uses a new feature extraction technique that utilizes a semantic that gives a higher weight to the words related to the emotion word, unlike the conventional emotion recognition system using the BoW (Bag of Word) It is possible to provide a recognition rate.

The music emotion recognition apparatus 10 includes a receiver 100 and a processor 200. [ The music sensibility recognition apparatus 10 may further include a memory 300. [

The receiver 100 may receive data related to the lyrics of the music. The receiver 100 can receive data from the memory 300 or receive data related to music lyrics from the outside. The receiver 100 may output data related to the lyrics of the music to the processor 200.

The data related to the lyrics of the music may include a target music to be recognized, a lyrics of a target music to be recognized, lyrics of music for learning an artificial neural network, and a lexical data set for generating a vocabulary dictionary.

The processor 200 may generate a vocabulary dictionary based on data related to the lyrics of the received music. The processor 200 may generate a vocabulary dictionary by filtering data related to the lyrics of the music.

For example, the processor 200 may filter data related to lyrics of music based on the type of language. The processor 200 may filter the filtered data based on the part of the word based on the type of language.

The processor 200 may remove words that are not meaningful from the filtered data based on the part of speech. Specifically, the processor 200 may remove at least one of numbers, exclamations, alphabets, and relative pronouns from the filtered data based on the parts of speech. The order and type of filtering may be changed as needed. The filtering operation will be described in detail with reference to Fig.

The processor 200 can extract a feature vector based on the generated lexical dictionary and a weight corresponding to a word included in the data. The processor 200 may generate a set of words based on data related to the lyrics of the music.

Specifically, the processor 200 may divide words included in the data. The processor 200 may recover the original form of the segmented word to generate a set of words.

Processor 200 may generate a generation vector based on a lexical dictionary and a set of words. The processor 200 may extract the feature vector by transforming the components of the occurrence vector based on the weights.

Specifically, the processor 200 may calculate the first weight based on the number of words contained in the set of words. The processor 200 may calculate the first weight using a TF-IDF (Term Frequency - Inverse Document Frequency).

The processor 200 may calculate the second weight based on a non-linear function according to a predetermined constant. For example, the processor 200 may calculate a second weight based on a sigmoid function.

The processor 200 may extract the feature vector by transforming the components of the generation vector based on the product of the first weight and the second weight.

The processor 200 may use the feature vector as an input to the artificial neural network to determine the emotion of the music. The processor 200 may calculate a probability value corresponding to a plurality of emotion groups using the artificial neural network. At this time, the processor 200 can determine the sensitivity of the music based on the probability value.

The processor 200 can learn an artificial neural network. The artificial neural network may include RNN (Recurrent Neural Network), CNN (Convolutional Neural Network), and DBN (Deep Belief Network). For example, an artificial neural network may include DNB.

The processor 200 can learn the artificial neural network using the transition learning. For example, the processor 200 may learn the DBN using transition learning.

The memory 300 may store a learning parameter of the artificial neural network, a probability value according to the emotion model, data on the received music, lyric information, and the like.

The memory 300 may be implemented as a volatile memory device or a non-volatile memory device.

The volatile memory device may be implemented in a dynamic random access memory (DRAM), a static random access memory (SRAM), a thyristor RAM (T-RAM), a zero capacitor RAM (Z-RAM), or a twin transistor RAM (TTRAM).

Non-volatile memory devices include electrically erasable programmable read-only memory (EEPROM), flash memory, magnetic RAM (MRAM), spin transfer torque (MRT) , FeRAM (Ferroelectric RAM), PRAM (Phase change RAM), Resistive RAM (RRAM), Nanotube RRAM, Polymer RAM (PoRAM), Nano Floating Gate Memory (NFGM), a holographic memory, a molecular electronic memory device, or an insulator resistance change memory.

Fig. 2 shows the overall operation of the music emotion recognition apparatus shown in Fig. 1, and Fig. 3 shows an operation in which the music emotion recognition apparatus shown in Fig. 1 generates a lexical dictionary.

Referring to FIG. 2 and FIG. 3, the music sensibility recognition apparatus 10 recognizes the emotion of a person hidden in the lyrics of the music, and classifies and recommends the music based on emotion. The music sensibility recognition apparatus 10 can construct a dictionary of words necessary for music lyrics analysis.

The music sensibility recognition apparatus 10 can extract a feature vector of 1082 dimensions for the music lyrics inputted through the vector quantization and the weighted technique for the emotional vocabulary based on the lexical dictionary.

The music sensibility recognition apparatus 10 can generate a probability value for the emotional group of the Russell of the individual music through the DBN (Deep Belief Network) developed using the transition learning.

The music sensibility recognition apparatus 10 can generate a dictionary of words based on data related to the lyrics of the received music.

The music sensibility recognition apparatus 10 constructs a vocabulary dictionary based on the vocabulary most frequently used in the MSD (Million Song Dataset), which is the most used data set in the field of music emotion recognition (MER) .

The music sensibility recognition apparatus 10 can generate a vocabulary dictionary by filtering data related to music lyrics.

The music sensibility recognition apparatus 10 can generate a vocabulary dictionary by filtering data related to music lyrics. The music sensibility recognition apparatus 10 can remove words corresponding to noises and noises irrespective of sensitivity, and can select only vocabulary to be used for the feature vector.

Specifically, the music sensibility recognition apparatus 10 can filter data based on the type of language. For example, the music sensibility recognition apparatus 10 can remove words written in a language other than English by using an available language detection API (Application Programming Interface).

The music sensibility recognition apparatus 10 can filter data based on the part of speech. For example, the musical-feeling recognition apparatus 10 uses part-of-speech to leave only parts of speech (for example, adjectives, nouns, and verbs) that can affect the emotional word of music Can be filtered.

The music sensibility recognition apparatus 10 can remove words that are meaningless. For example, the music emotion recognition apparatus 10 may filter languages such as numbers and simple exclamations (e.g., yeah).

Thus, the music sensibility recognition apparatus 10 can generate a lexical dictionary including a plurality of words. For example, the music sensibility recognition apparatus 10 can generate a lexical dictionary composed of 1082 English words.

The music sensibility recognition apparatus 10 can extract a feature vector based on the generated lexical dictionary and a weight corresponding to a word included in the data. The music sensibility recognition apparatus 10 can improve the existing BoW expression and express the lyric using a new BoW which gives a larger weight to the emotional vocabulary.

The music sensibility recognition apparatus 10 can extract a feature vector through three processes. First, the music sensibility recognition apparatus 10 can perform preprocessing on the lyrics of the received music.

Specifically, the music sensibility recognition apparatus 10 can divide words included in the data. The music sensibility recognition apparatus 10 can recover a prototype of a divided word to generate a set of words. For example, the music sensibility recognition apparatus 10 can perform tokenization in which the lyrics of each music are divided into a set of words, and stemming in which each original is restored to the divided words.

Second, the music sensibility recognition apparatus 10 can perform vector quantization. The music sensibility recognition apparatus 10 can represent the occurrence music of 1082 dimensions by quantizing individual music lyrics based on the generated lexical dictionary.

Thirdly, the music sensibility recognition apparatus 10 can calculate the weight based on the emotional vocabulary. The music sensibility recognition apparatus 10 can calculate the first weight based on the number of words included in the set of words. The music sensibility recognition apparatus 10 can calculate the first weight using the TF-IDF (Term Frequency - Inverse Document Frequency). For example, the music sensibility recognition apparatus 10 may calculate the first weight? _{A, i} using Equation (1).

Here, N is the frequency with which the corresponding word appears in the lyrics of all the music in the database, and N _i may be the frequency with which the word appears in the lyrics of the i-th music.

The music sensibility recognition apparatus 10 can calculate the second weight based on a nonlinear function according to a predetermined constant. The music sensibility recognition apparatus 10 can calculate the second weight based on the sigmoid function. For example, the music sensibility recognition apparatus 10 can calculate the second weight? _{S, i} using Equation (2). Hereinafter, the second weight may be defined as a sentiment score.

Here, α means a slope decision constant, and S _i can mean a constant determined by an emotional word dictionary. For example, S _i may be a value provided in an emoticon that is valued from -3 to +3 for words related to emotion.

The emotional dictionary used by the music sensibility recognition apparatus 10 may include SentiStrength and SentiWordNet used in data mining. Also, the slope decision constant < RTI ID = 0.0 > a < / RTI >

The music sensibility recognition apparatus 10 can extract the feature vector by converting the components of the generation vector based on the product of the first weight and the second weight. For example, the music sensibility recognition apparatus 10 can calculate? _I , which is a weight value of each element of the generation vector, using Equation (3).

The music sensibility recognition apparatus 10 can recognize the emotion of the music using the artificial neural network. In addition, the music sensibility recognition apparatus 10 can learn an artificial neural network.

For example, the music emotion recognition apparatus 10 can perform emotion recognition using a DBN. The emotional class used by the musical emotion recognition apparatus 10 is a representative emotion of each quadrant in the emotional model of Russell composed of arousal and valance axes as follows: happy, relaxed, Sadness, and tension.

In general, defining the structure of a deep neural network and optimizing parameters can require vast amounts of data and high-performance computing and learning time. The music sensibility recognition apparatus 10 can use transfer learning to reduce such a cost.

 The structure of the DBN used by the music sensibility recognition apparatus 10 may include an input layer, two hidden layers, and an output layer. In this case, the number of input nodes may be 1082, and the number of nodes of the first and second hidden layers may be 1000 and 500, respectively.

Finally, the output layer can be four because it corresponds to Russell quadrant emotion. Finally, the music sensibility recognition apparatus 10 can learn the parameter values of the DBN through fine-tuning using the collected music lyrics data of 3000 songs.

FIG. 4A shows the distribution of the feature vectors according to the TF-IDF weights, and FIG. 4B shows the distribution of the feature vectors according to the weights according to the music emotion recognition apparatus shown in FIG.

Referring to FIGS. 4A and 4B, the music emotion recognition apparatus 10 can improve the discrimination power of the individual music lyrics by considering emotional word based weights using emotion scores. Referring to FIG. 4B, individual musicians can be represented in a more recognizable form through the above-described feature vector extraction method.

4A and 4B, the feature vectors generated by the music emotion recognition apparatus 10 can represent different feature vector distributions according to emotional groups, unlike the case where only the TF-IDF is used. Accordingly, the music sensibility recognition apparatus 10 can further improve the performance of the music classification result.

Fig. 5 shows a result of comparison between the recognition accuracy of the conventional art and the music emotion recognition apparatus shown in Fig.

Referring to FIG. 5, it can be seen that the music sensibility recognition apparatus 10 has a higher recognition rate than the conventional technique. The music sensibility recognition device 1 shows the performance when the SentiWordNet emoticon is used and the music sensibility recognition device 2 can show the performance when the SentiStrenth emotion word dictionary is used.

Finally, it can be confirmed that the highest recognition performance is obtained when SentiStrenth is used.

6 is a flowchart of the operation of the music emotion recognition apparatus shown in Fig.

Referring to FIG. 6, the receiver 100 may receive data related to the lyrics of the music (610). The receiver 100 can receive data from the memory 300 or receive data related to music lyrics from the outside. The receiver 100 may output data related to the lyrics of the music to the processor 200.

The processor 200 may generate a vocabulary dictionary based on data related to the lyrics of the received music (630). The processor 200 may generate a vocabulary dictionary by filtering data related to the lyrics of the music.

For example, the processor 200 may filter data related to lyrics of music based on the type of language. The processor 200 may filter the filtered data based on the part of the word based on the type of language.

The processor 200 may remove words that are not meaningful from the filtered data based on the part of speech. Specifically, the processor 200 may remove at least one of numbers, exclamations, alphabets, and relative pronouns from the filtered data based on the parts of speech.

The processor 200 may extract the feature vector based on the generated lexical dictionary and a weight corresponding to the word included in the data (650). The processor 200 may generate a set of words based on data related to the lyrics of the music.

Specifically, the processor 200 may divide words included in the data. The processor 200 may recover the original form of the segmented word to generate a set of words.

Processor 200 may generate a generation vector based on a lexical dictionary and a set of words. The processor 200 may extract the feature vector by transforming the components of the occurrence vector based on the weights.

Specifically, the processor 200 may calculate the first weight based on the number of words contained in the set of words. The processor 200 may calculate the first weight using a TF-IDF (Term Frequency - Inverse Document Frequency).

The processor 200 may calculate the second weight based on a non-linear function according to a predetermined constant. For example, the processor 200 may calculate a second weight based on a sigmoid function.

The processor 200 may extract the feature vector by transforming the components of the generation vector based on the product of the first weight and the second weight.

The processor 200 may use the feature vector as an input to the artificial neural network to determine the emotion of the music (670). The processor 200 may calculate a probability value corresponding to a plurality of emotion groups using the artificial neural network. At this time, the processor 200 can determine the sensitivity of the music based on the probability value.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer readable medium may include program instructions, data files, data structures, and the like, alone or in combination. Program instructions to be recorded on the medium may be those specially designed and constructed for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

Although the embodiments have been described with reference to the drawings, various technical modifications and variations may be applied to those skilled in the art. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI > or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (20)

A step of the music sensibility recognition device receiving data related to the lyrics of the music;
Generating a lexicon dictionary based on the data;
Extracting a feature vector based on a weight corresponding to a word dictionary included in the lexical dictionary and the data; And
Determining the emotion of the music using the feature vector as an input to the artificial neural network
Lt; / RTI >
Wherein the extracting comprises:
Generating a set of words based on the data;
Generating a generation vector based on the lexical dictionary and the set of words; And
And extracting the feature vector by transforming the components of the occurrence vector based on the weights
Music Sensibility Recognition Method.
The method according to claim 1,
Wherein the generating comprises:
Filtering the data based on a language type;
Filtering the filtered data based on the part of speech based on the type of language; And
Removing the meaningless word from the filtered data based on the part of speech to generate the dictionary of words
And a music emotion recognition method.
3. The method of claim 2,
Wherein the step of generating the lexical dictionary by removing a meaningless word from the filtered data based on the part of speech of the word comprises:
Removing at least one of a number, an annotation, an alphabet, and a relative pronoun from the filtered data based on the part-of-speech to generate the lexicon dictionary
And a music emotion recognition method.
delete The method according to claim 1,
Wherein the step of generating the set of words comprises:
Dividing words included in the data; And
Recovering the original form of the word to generate a set of words
And a music emotion recognition method.
The method according to claim 1,
Wherein the step of extracting the feature vector by transforming the components of the generation vector based on the weights comprises:
Calculating a first weight based on the number of words included in the set of words;
Calculating a second weight based on a non-linear function according to a predetermined constant; And
Extracting the feature vector by transforming a component of the occurrence vector based on a product of the first weight and the second weight;
And a music emotion recognition method.
The method according to claim 6,
Wherein the step of calculating the first weight comprises:
Calculating the first weight using a TF-IDF (Term Frequency - Inverse Document Frequency)
And a music emotion recognition method.
The method according to claim 6,
Wherein the calculating the second weight comprises:
Calculating the second weight based on a sigmoid function,
And a music emotion recognition method.
The method according to claim 1,
Wherein the determining comprises:
Calculating a probability value corresponding to a plurality of emotion groups using the artificial neural network; And
Determining the emotion of the music based on the probability value
And a music emotion recognition method.
The method according to claim 1,
The artificial neural network is a DBN (Deep Belief Network)
The DBN is learned using transfer learning
Music Sensibility Recognition Method.
A receiver for receiving data related to lyrics of music; And
Generating a lexicon dictionary based on the data, extracting a feature vector based on the lexicon dictionary and a weight corresponding to a word included in the data, and using the feature vector as an input of an artificial neural network, Determining processor
Lt; / RTI >
The processor comprising:
Generating a set of words based on the data, generating a generation vector based on the lexical dictionary and the set of words, and extracting the feature vector by converting a component of the generation vector based on the weight
Music sensibility recognition device.
12. The method of claim 11,
The processor comprising:
Filtering the data based on the type of the language, filtering the filtered data based on the part of the word based on the type of the language, removing the meaningless word from the filtered data based on the part of speech of the word, To create a dictionary
Music sensibility recognition device.
13. The method of claim 12,
The processor comprising:
Removing at least one of numbers, exclamations, alphabets, and relative pronouns from the filtered data based on the part-of-speech to generate the lexical dictionary
Music sensibility recognition device.
delete 12. The method of claim 11,
The processor comprising:
Dividing words included in the data, restoring the original form of the word to generate a set of words
Music sensibility recognition device.
12. The method of claim 11,
The processor comprising:
Calculating a first weight based on a number of words included in the set of words, calculating a second weight based on a nonlinear function according to a predetermined constant, and calculating a second weight based on a product of the first weight and the second weight And extracts the feature vector by converting the component of the occurrence vector
Music sensibility recognition device.
17. The method of claim 16,
The processor comprising:
The first weight is calculated using a TF-IDF (Term Frequency - Inverse Document Frequency)
Music sensibility recognition device.
17. The method of claim 16,
The processor comprising:
Calculating the second weight based on a sigmoid function
Music sensibility recognition device.
12. The method of claim 11,
The processor comprising:
Calculating a probability value corresponding to a plurality of emotion groups using the artificial neural network, and determining emotion of the music based on the probability value
Music sensibility recognition device.
12. The method of claim 11,
The artificial neural network is a DBN (Deep Belief Network)
The DBN is learned using transfer learning
Music sensibility recognition device.

Images