TWI486949B - Music emotion classification method - Google Patents

Description

Musical emotional classification

The invention relates to a music classification method, in particular to a music emotion classification method.

In the era of information explosion, the acquisition of music is quite easy. With the growing popularity of video games, many video games will later release game soundtracks for the music they use, or even hold concerts. In order to conform to the content of the game scene, music usually has a story, so if the music can be classified by emotion and the player can select music that suits his or her mood, the video game will be more popular with the player.

Although there are currently developers who have developed so-called software systems or devices that can classify music emotions, the classification methods of such software systems or devices are based on pre-set rules of music to classify music emotions, because everyone feels about music. They are not the same, and the classification results are often not accepted by users. Therefore, this classification method cannot be applied to everyone, and the applicability is poor. In addition, the analysis accuracy of music containing speech is also quite poor.

Accordingly, it is an object of the present invention to provide an emotional classification method for emotionally classifying music through tonality, intervals, rhythms, and timbres.

Another object of the present invention is to provide an emotion classification method that allows a user to classify emotions of music according to his or her own feelings.

Thus, the emotional classification method of the music of the present invention is applicable to the software And/or hardware-based construction in an electronic device for emotionally classifying music, comprising the steps of: (A) respectively setting a corresponding emotional adjective parameter for the plurality of music samples stored in the electronic device; B) performing music characteristics such as tonality, rhythm, interval and timbre for each music sample of step (A), and outputting a plurality of music feature parameters for each music sample; (C) multi-class support vector machine Statistically analyzing the relationship between the emotional adjective parameters of the music samples and the corresponding music feature parameters, and constructing a multi-category emotion prediction model; and (D) performing each new music stored in the electronic device for subsequent storage Music feature analysis correspondingly outputs a plurality of music feature parameters, and analyzes the music feature parameters of the new music by the multi-category emotion prediction model, and classifies the music by emotion, and outputs a pair of emotional adjectives that should be new music parameter.

The effect of the invention: by using the support vector machine to analyze the musical characteristic parameters such as tonality, interval, rhythm and timbre of the music, and the statistical analysis of the emotional adjective parameters set by the music, a construct can be constructed A multi-category emotion prediction model in which music is classified into emotions and has high classification accuracy.

The above and other technical contents, features and advantages of the present invention will be apparent from the following detailed description of the preferred embodiments.

As shown in FIG. 1, a preferred embodiment of the emotion classification method of the music of the present invention is applicable to an electronic device through software and/or hardware, and can be analyzed through tonality, interval, rhythm and tone. Come to the electronics The music stored by the device is emotionally classified, and the music may be video game music, song music, or instrumental music. The sentiment classification method includes the following steps:

Step (1) setting emotional adjective parameters for the music sample. First, a plurality of emotional adjective parameters representing their emotional affiliation are respectively set for a plurality of music samples whose emotion classification has been determined.

In this embodiment, the music samples are classified into five categories according to their emotion classification, and each of the emotion categories corresponds to an emotional adjective and an emotional adjective parameter, and the emotional adjectives are “quiet” and “cold” respectively. ", "singular", "humorous", and "enthusiasm". However, the type of emotional adjectives used is not limited to the above types.

Step (2) Performing music feature analysis on all music samples to obtain music feature parameters of the music samples. For each music sample, four musical features such as tonality, interval, rhythm and timbre are analyzed, and the music characteristic parameters of each music sample are output correspondingly.

Among them, the analysis of tonal music characteristics is to analyze the tonal types contained in each music sample, and the proportion of various tonality (%), such as C major, D major, E major and B major. Tune, and the proportion of tonality in C minor, D minor and B minor, and statistically analyze the average proportion (%) of various tonality of all music samples corresponding to each emotional adjective parameter. The interval music feature is to analyze the complete consonant interval, incomplete consonant interval, and the proportion of non-coordinated intervals used in each music sample, and statistically analyze the complete consonant interval of all music samples corresponding to each emotional adjective parameter. Incomplete consonant interval, and the average share of non-coordinated intervals proportion(%). The analysis of the rhythm music feature is to analyze the beat per minute (bpm) of each music sample, and statistically analyze the average rhythm speed of all the music samples corresponding to each emotional adjective parameter. The analysis of the timbre music features is to analyze the sound shape of each music sample, that is, to analyze its spectral distribution.

After analyzing the music features of each music sample, the music features are data encoded and normalized to produce desired music feature parameters.

In this embodiment, the tonal music features, the interval music features, the rhythm music features, and the timbre music features of the music samples are respectively analyzed by the MATLAB program software, but when implemented, due to the numerous analysis methods of the aforementioned music features, Not limited to the software used in this case. In addition, since data coding and normalization are common technical means for existing statistical analysis, and there are many ways, they will not be described in detail.

Step (3) establish a multi-category emotion prediction model with Support Vector Machine (SVM). The multi-category emotion prediction model is established by the support vector machine for the emotional adjective parameters set in each music sample in step (1) and the music characteristic parameters obtained in the step (2) of the music samples.

In this embodiment, statistical analysis is performed according to the music feature parameters of each music sample and the emotional adjective parameters corresponding to each music sample, thereby forming a multi-category classification problem, and then decomposing the multi-category classification problem into a series of ones. One-against-one (OAO) support vector machine model, cross-validated by Gaussian kernel function, and Based on the judgment results of all support vector machine models, a multi-category emotion prediction model is established. At this time, the multi-category emotion prediction model can be used to emotionally classify other music stored separately for the electronic device.

Step (4) emotionally classifying other music stored in the electronic device by the multi-category emotion prediction model. After the multi-category emotion prediction model is established according to the foregoing music sample, the multi-category emotion prediction model can perform the emotion classification processing of the music. When classifying, the music characteristics of the new music waiting for analysis are first analyzed to obtain musical characteristic parameters such as tonality, interval, rhythm and timbre of each new music to be classified, and then the multi-category emotion prediction is performed. The model analyzes the music feature parameters of each new music to be classified, and finds the emotional adjective parameters corresponding to each new music to be classified, and then analyzes the emotional category to which each new music belongs, and completes After the sentiment analysis, the electronic device is driven to feedback an animated image corresponding to the emotional adjective parameter of the analyzed new music, thereby reminding the electronic device user.

Step (5) classifying and storing the analyzed new music according to the parameters of the emotional adjectives, and establishing a music emotion classification database.

Through the above steps, the present invention analyzes the musical characteristic parameters such as tonality, interval, rhythm and timbre of the music samples, and performs statistical analysis of the support vector machine with the emotional adjective parameters to construct a multi-category emotion prediction model, which is more The accuracy of the emotional classification of the category sentiment prediction model is very high.

It must be noted that the preliminary emotional classification of the aforementioned music samples may be based on the user who uses the electronic device constructed with the method. Feel the classification, so the music emotion classification model of the multi-category emotion prediction model established in step (3) will be close to personal preference. Therefore, the emotional classification method of the music of the present invention can conveniently construct the user of the electronic device of the method. According to its own feelings, it establishes a unique multi-category emotion prediction model, which makes the classification result of music more in line with individual needs.

Of course, during the implementation, the emotional classification of the music samples can also be generated through the analysis of the questionnaire. At this time, the game classification results of the multi-category emotion prediction model constructed by the method will be closer to the generality of the general public. Cognitive, but applicable to game software developers.

Therefore, the emotion classification method of the music of the present invention can conveniently establish an exclusive multi-category emotion prediction model that meets various needs according to the emotional classification difference of the music originally established by the user, and can improve the traditional music emotion classification system only according to the original The set standard analysis mode has the disadvantage of classifying music emotions. And by adding the part of the timbre analysis, it is also possible to perform a more detailed analysis of the speech part of the music, which helps to improve the emotional classification accuracy of the music containing the voice.

In summary, by using the support vector machine to analyze the musical characteristics such as tonality, interval, rhythm and timbre of music, and the statistical analysis of the emotional adjective parameters set by the music, it is possible to construct an accurate A multi-category sentiment prediction model that classifies music emotions, and can improve the sentiment analysis accuracy of music with speech content by timbre analysis, which will help video game players and video game developers to quickly acquire emotional classification of music. Music developers can design game products that meet the needs of players for specific game emotional content or situations.

In addition, the present invention can also be used by the user to establish an exclusive multi-category emotion prediction model according to the emotional classification of the music samples made by the user according to his own listening experience, so that the emotional classification result of the music can be closer to itself in the future. Preferences, but can improve the traditional music sentiment classification system can only be based on its original set of analysis mode for music emotional classification of the shortcomings, quite convenient and practical. Therefore, the object of the present invention can be achieved.

The above is only the preferred embodiment of the present invention, and the scope of the invention is not limited thereto, that is, the simple equivalent changes and modifications made by the scope of the invention and the description of the invention are All remain within the scope of the invention patent.

BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 is a flow chart showing the steps of a preferred embodiment of the emotional classification method for music of the present invention.

Claims (3)

A method for classifying emotions of music, which is applied to an electronic device through software and/or hardware, and used for emotional classification of music, comprising the following steps: (A) storing a plurality of music for the electronic device The sample respectively sets a corresponding emotional adjective parameter; (B) performs music feature analysis such as tonality, rhythm, interval and timbre for each music sample of step (A), and outputs a plurality of music features corresponding to each music sample. (C) statistically analyzing the relationship between the emotional adjective parameters of the music samples and the corresponding musical feature parameters by a multi-class support vector machine, and constructing a multi-category emotion prediction model; and (D) Each new music stored in the electronic device is subjected to music feature analysis to output a plurality of music feature parameters, and the music feature parameters of the new music are analyzed by the multi-category emotion prediction model, and the new music is emotionally classified. And output a pair of emotional adjective parameters that should be new music. According to the emotional classification method of the music described in claim 1, the method further includes a step (E): analyzing the emotional adjective parameter corresponding to the new music according to the step (D), storing the new music classification, establishing A music emotion classification database. According to the emotion classification method of the music described in claim 1, the step (D) also displays an image via the electronic device for reminding when analyzing the emotional adjective parameters of each new music.