Discovering Similar Music for Alpha Wave Music

Discovering Similar Music for Alpha Wave Music Yu-Lung Lo ( ), Chien-Yu Chiu, and Ta-Wei Chang Department of Information Management, Chaoyang University of Technology, 168, Jifeng E. Road, Wufeng District, Taichung 41349, Taiwan, R.O.C. yllo@cyut.edu.tw, ciouyu09@gmail.com, mylyfwy771@gmail.com Abstract. When people close eyes to relax, an alpha wave in the frequency range of 8 13 Hz appears from brain signals. There were many medical reports proofed that some specific music can resonate with the alpha wave and strengthen the wave. Therefore, this alpha wave music can improve more relaxing for people and are very helpful when they need to take a rest. Due to the alpha wave music is classified manually by experts only, it is not popular in the market currently. In this paper, we will investigate the content-based features of the alpha wave music and use them to analyze the similarity between alpha wave music and existing music genres. The purpose of this research is to find the music which is similar to alpha wave music, such that we can recommend to users for relaxing before the automatic classification scheme for alpha wave music being developed. 1 Introduction Listening music can stimulate the brain s functioning and music therapy uses music to help patients to improve or maintain their physical and spiritual health [7]. People usually listen to music to relieve stress when they feel under the pressure. However, which music can help people to relieve the pressure? The Dr. Hans Berger discovered four major types of brain waves exist, including β (Beta) wave, α (Alpha) wave, θ (theta) waves, and δ (delta) wave [3]. There are different frequencies of brain wave detected while humans are in different state of mind. Among them, the frequency of alpha wave between 8 Hz and 13 Hz, as shown in Fig. 1, is measured by EEG (Electroencephalography) when people close their eyes for a short rest. There were many medical reports proofed that some specific music can resonate with the alpha wave and strengthen it [1, 2, 7]. Therefore, the alpha wave music can improve more relaxing for people and is very helpful when they need to take a rest. That s why most people like to listen to music when relaxing. Although it rare, there still a few alpha wave music albums can be found on the market, such as: Masterworks The Journey Home Into The Deep Gaia Cloudscapes. Springer Science+Business Media Singapore 2018 K.J. Kim and N. Joukov (eds.), Mobile and Wireless Technologies 2017, Lecture Notes in Electrical Engineering 425, DOI 10.1007/978-981-10-5281-1_63

572 Y.-L. Lo et al. Fig. 1. Frequency of alpha brain waves detected by EEG [16] The content of digital music provides many features which can be used for music analysis and retrieval. The music features, such as melody, rhythm, and chord, can represent the music styles and characteristics. Therefore, content-based music classification as well as music retrieval is an important research field for music databases. There were approaches for content-based music classification, such as music classification using significant repeating patterns by [8], hierarchical genre classification for large music collections by [4], automatic chord recognition for music classification and retrieval by Cheng et al. [5], content-based multi-feature music classification by Lo et al. [9], and so forth. However, these existing music classification approaches are almost all categorized by styles and genres, such as pop, classical, jazz, folk, etc. Lo et al. [10] also proposed content-based classification of alpha wave music, however, this study emphasized on analyzing the common features of already identified alpha wave music. It can not substantiate the accuracy for further application on classifying of alpha wave music. Accordingly, till now, the alpha wave music is classified manually by expertise only and is very rare on the market. In this paper, we will investigate the content-based features of music and use them to analyze the similarity between alpha wave music and existing music genres. The purpose of this research is to find the music which is similar to alpha wave music, such that we can recommend user to listen such music for relaxing before the scheme of automatic classification of alpha wave music being developed. We hope our effort will help people not only to find more relaxed music but also to aid of music therapy. 2 Related Works In recent years, automatic classification of music data can be discriminated into two categories. One is based on analysis of music content for classification, such as SRP- Based Classification by Lin et al. [8], Hierarchical Genre Classification by Brecheisen et al. [4], content-based multi-feature music classification by Lo et al. [9], and so on. The other category is the application of training by learning machines in which naive Bayesian, linear and neural network are employed to build classifiers for styles, such as Extreme Learning Machine by Loh et al. [11], Multiple-Instance Learning by Mandel [12], Automatic Chord Recognition by Cheng et al. [5], Multi-modal Music Genre Classification Zhen et al. [15], Optimized Feature Vector by Deepa et al. [6], and so forth. In addition, the Multi-Label Music Mood Classification proposed by Myint et al. [13] uses new mood taxonomy model to classify music.

Discovering Similar Music for Alpha Wave Music 573 These music classification approaches are just about all categorized by styles and genres, such as pop, classical, jazz, folk, etc. Lo et al. [10] studied the common features of the already identified alpha wave music. However, their work cannot be verified consistent with the consequence of classifying by expertise. Accordingly, till now, the alpha wave music is classified manually by expertise only. Therefore, to find the music which is similar to alpha wave music may also be a good way to recommend for people. 3 Research Method 3.1 Music Features Our research method based on music features and comparison schemes. A musical composition consists of three basic elements - note, rhythm and harmony. Chords are a part of harmony as well. Moreover, the pitch change is also an important characteristic to compose music. Notes A melody was composed by notes. The fundamental frequency of musical note A above middle C is usually set at 440 Hz [14]. The pitch ratio between any two successive notes of the scale is exactly equal to 12 2 (about 1.05946). The A an octave above that is 880 Hz because Rhythms Chords they are twelve notes apart. We would like to explore the alpha wave music to ascertain whether there are specific notes existing most likely to come about the harmonic resonance in the brain. Rhythm is the pattern of musical movement through time. It is formed by a series of notes differing in duration and stress. For example: the 2/2 time signature means two half-note (crotchet) beats per bar and the beat pattern is strong-weak, the 3/4 time signature means three quarternote beats per bar and the beat pattern is strong-weak-weak, and the 4/4 time signature means four quarter-note beats per bar and the beat pattern is strong-weak-strong-weak. Most of Waltzes music are the 3/4 time signature. Generally, quick tempo can boost the human s spirit and slow tempo can make people to feel relaxing. We would also like to analyze the connection between music rhythm and alpha wave music. A chord in music is any harmonic set of two or more notes that is heard as if sounding simultaneously. The most frequently encountered chords are triads, so called because they consist of three distinct notes, further notes may be sevenths, ninths, and so forth. There are also four types of triads - major, minor, augmented, and diminished. People listening to various chords have distinct feelings such as sorrowful for major chords, suddenly enlightened for diminished seventh chords, and unexpectedly flying overhead for major second chords. The affection of chord may be an interesting direction for studying alpha wave music. Pitch change Pitch change is the variation of two adjacent notes. For example, a melody segment of the Little Bee is So Mi Mi Fa Re Re Do Re Me Fa such that

574 Y.-L. Lo et al. the pitch changes will be 2 0 +1 2 0 1 1 1 1. Since the pitch change is not effected by music key up and down, it is a favorable feature for query by example in music retrieval. Normally, the pitch change of hot music is more significant that may inspirit people. On the contrary, the pitch change of lyrical music is smoother that may allow people to relax. The alpha wave music seems to have the same effect as lyrical music does. Among them, the chord is complicate in variety. Therefore, only notes, rhythms and pitch changes will be investigated in our studies. 3.2 Comparison Schemes Our study used distance functions and machine learning to explore which music is more similar to alpha wave music. 3.2.1 Distance Function In [10], we can first analyses the music content, such as notes and rhythms, as features for individual genre (ex: alpha, classical, and so forth). Let the frequencies of n highest occurrences are x 1, x 2, x n for a feature of a music genre then these values can be the coordinates of the centre as in an n dimensional space. Thus, alpha wave music can be examined by the distance from the centre of each music genre. Suppose a music has been analyzed and the n highest occurrences of a music feature are y 1, y 2,, y n with in decreasing order. The distance function d(y 1, y 2,, y n ) for the music to the centre can be derived as Eq. (1). d(y 1, y 2,, y n )= n (x i y i ) 2 (1) Thus, the most closest genre of an alpha wave music then can be decided when the distances to the centre of all music genres have been examined. i=1 3.2.2 Distance Function with Weight We derived another distance function for our experiment as shown in Eq. (2). It is similar to Eq. (1) except that a weight factor w i is added. Where w i denotes the weight for the ith music feature. wd(y 1, y 2,, y n )= n (w i (x i y i )) 2 (2) i=1 3.2.3 Machine Learning Machine learning is an algorithm that analyzes the rules from the sample data and uses the rules to automatically predict the unknown data. It is also often used in data

Discovering Similar Music for Alpha Wave Music 575 classification such as [11, 12]. Therefore, we hope that through the machine learning technology to analyze the characteristic regularity of the music genre and to carry out the classification of alpha wave music analysis. Our experiment will use support vector machine through LIBSVM [17] and MATLAB [18] to achieve. Support Vector Machine (SVM) It is an approach for statistical classification and regression analysis. In which, the classified data is trained to find a hyperplan to establish a classification model. Then, such model is used to exam the data that has not yet been classified. LIBSVM (A Library for Support Vector Machines) Proposed in [17], it supports diverse classifications for easy used of SVMs (such as C-SVC, nu-svc) and regression analyses (such as epsilon-svr, nu- SVR). We denote that we will exam music data in two ways for SVM experiment. The first one uses the highest frequency of the n characteristics of music in a genre for SVM training to establish a classification model. The second way uses Eq. (2) to evaluate the distance for music to the center of the belonged genre such that the distances can be used for SVM training to establish a classification model. 4 Experiment Analyses 4.1 Experimental Setting We collect classical, folk, pop, jazz, and blue five music genres in our database and each genre has 150 pieces of music. Since there have been not numerous alpha wave music albums classified by experts in the market, we have merely collected 87 scores of alpha wave music for our music database. Therefore, our experimental database contains total 837 of music. We also extracted notes, rhythms, and pitch changes of collected music as the features and analyzed their occurrence frequencies for experiment. Thus, the distance equations and LIBSVM can be applied in our experiment. The experimental results are shown in the following sessions. 4.2 Experimental Results 4.2.1 Analysis of Notes To start experiment, we first analyze the occurrence frequencies of notes of music data in our database and then the centre coordinates of features for each music genre (except alpha wave music) can be established. Having these centre coordinates, the alpha wave

576 Y.-L. Lo et al. music can be examined by Eqs. (1) and (2) one by one to find the nearest centre coordinate which may be the most likely similar music genre. The numbers of the highest occurrences (n) for the centre coordinate of a music genre examined are varied from 2 to 7. We used the center coordinate value corresponding to each music feature as weight (w i ) in Eq. (2), so that it with high frequency has a relatively high weight value to strengthen it in calculating the distance. The experimental results for analysis of notes are shown in Fig. 2. The results show that there are 63% 72% of alpha wave music similar to classical music. (a) by equation (1) (b) by equation (2) Fig. 2. Note analysis by distance equations Furthermore, we used LIBSVM to analyze which genre the alpha wave music is most likely close to. The top 2 to 7 of highest occurrence notes are used for training to establish classification model for each music genre. Then, we can use these classification models to exam alpha wave music. The experimental results are shown in Fig. 3(a). This result demonstrates that there are 77% 90% of alpha wave music similar to classical music. In addition, the distances of each music to the center coordinate of belonged genre is computed by Eq. (2) also used to train for building classification models. The alpha wave music is also investigated in this classification model. This experimental result is shown in Fig. 3(b) and it demonstrates that there are 87% 97% of alpha wave music being similar to classical music. (a) note occurrence for training (b) distance for training Fig. 3. Note analysis by LIBSVM

Discovering Similar Music for Alpha Wave Music 577 4.2.2 Analysis of Rhythms This experiment is the same as analysis of notes except that rhythms is instead of notes. The experimental results are shown in Figs. 4 and 5. The result is worse than analysis of notes. The alpha wave music is not quite close to a certain music genre. (a) by equation (1) (b) by equation (2) Fig. 4. Rhythm analysis by distance equations (a) rhythm occurrence for training (b) distance for training Fig. 5. Rhythm analysis by LIBSVM 4.2.3 Analysis of Pitch Changes This experiment is the same as analysis of notes except that pitch changes is instead of notes. The experimental results are shown in Figs. 6 and 7. There are up to 96% of alpha wave music similar to classical in Fig. 6(b) and up to 100% of alpha wave music similar to blue in Fig. 7(a) and (b). The experimental results of Figs. 6 and 7 are inconsistent which needs more further studies.

578 Y.-L. Lo et al. (a) by equation (1) (b) by equation (2) Fig. 6. Pitch change analysis by distance equations (a) pitch change occurrence for training (b) distance for training Fig. 7. Pitch change analysis by LIBSVM 4.3 Further Analysis for Classical and Blue From the previous experimental results can be found that the rhythm of the alpha wave music is not biased towards a specific genre. However, the notes of alpha wave music is closer to classical music genre, as well as the pitch changes of alpha wave music is closer to the blue music genre. In this section we further analyze which music to recommend in classical and blue, and such music may be able to achieve the effect of alpha wave music. Since the notes of alpha wave music are closer to classical music, we use the top two highest occurrence frequencies of classical notes for the two-dimensional center coordinates. In addition, we also use the occurrences of the same two notes in blue music as a two-dimensional center coordinates. Then, at each center coordinate, draw a circle for classical and blue music genres in which each covers 90% of belonged music in the database, as shown in Fig. 8(a). The main purpose of taking only cover 90% for drawing music circle is to exclude some music with special or exceptional features in their belonged genres. It can avoid the radius of circle being too large and becoming a sparse circle. We also used pitch change instead of notes for classical and blue to draw circles again, as shown in Fig. 8(b).

Discovering Similar Music for Alpha Wave Music 579 (a) by notes (b) by pitch changes Fig. 8. Circles for classical and blue music In Fig. 8(a), we find that the circle of classical music is included in the circle of blue music. That means the blue music which falls in the domain of classical music circle has common features in both music genres. Such blue music with common features may be closer to the alpha wave music and are worth to recommend for users. On the contrary, the circle of blue music is included in the circle of classical music in Fig. 8(b). There is also some blue music with common features similar to alpha wave music and are worth to recommend. We only used two-dimensional space (n = 2) in this study. However, we proposed this approach which can be deduced to the higher dimension analyses (n > 2) in order to obtain the music closer alpha wave music to be recommended. 5 Conclusion When people take a short rest with closed eyes, an alpha wave appears with brain signals. There were many medical reports proofed that some specific music can resonate with the alpha wave and strengthen the wave to improve more relaxing. Although there are many existing schemes for music classification, to categorize alpha wave music has not succeeded yet. Till now, the alpha music is classified manually by expertise only and rarely to be found in the market. In this research, we explored the contents of classical, pop, jazz, folk, blue, and alpha wave music by distance equations and learning machine approaches. We found that the notes of alpha wave music are closest to classical music as well as the pitch changes of alpha wave music are closest to blue music. Our further studies discovered that some music is similar to alpha wave music containing common features of classical and blue music. We would like to recommend such music to people for relaxing.

580 Y.-L. Lo et al. References 1. Basar E (1980) EEG brain dynamics. Elsevier Science, Amsterdam 2. Basar E (1988) Dynamics of sensory and cognitive processing by the brain. Springer, Berlin 3. Berger H (1969) On the electroencephalogram of man (Electroencephalography and clinical neurophysiology supplement No. 28). In: Gloor P (ed) Elsevier Science Ltd. ISBN-10: 0444407391 4. Brecheisen S, Kriegel H-P, Kunath P, Pryakhin A (2006) Hierarchical genre classification for large music collections. In: IEEE 7th international conference on multimedia and expo, pp 1385 1388 5. Cheng HT, Yang YH, Lin YC, Chen HH (2008) Automatic chord recognition for music classification and retrieval. In: IEEE international conference on multimedia and expo, pp 1505 1508 6. Deepa PL, Suresh K (2011) An optimized feature set for music genre classification based on support vector machine. In: Proceedings of IEEE conference on recent advances in intelligent computational systems (RAICS), pp 610 614 7. Goodman KD (2011) Music therapy education and training: from theory to practice. Charles C. Thomas, Springfield. ISBN 0-398-08609-5 8. Lin C-R, Liu N-H, Wu Y-H, Chen ALP (2004) Music classification using significant repeating patterns. Lecture notes in computer science, vol 2973. Springer, Heidelberg, pp 506 518 9. Lo YL, Lin YC (2012) Content-based multi-feature music classification. In: International conference on innovation and management, Republic, Palau 10. Lo YL, Lai Z-Y (2014) Content-based classification of alpha wave music. In: 2014 international conference on business and information (BAI 2014) 11. Loh QJB, Emmanuel S (2006) ELM the classification of music genres. In: 9th international conference on control, automation, robotics and vision, pp 1 6 12. Mandel M, Ellis DPW (2008) Multiple-instance learning for music information retrieval. In: 9th international conference on music information retrieval, pp 577 582 13. Myint EEP, Pwint M (2010) An approach for multi-label music mood classification. In: 2nd international conference on signal processing systems (ICSPS), pp 290 294 14. Rayleigh JWS, Lindsay RB (1945) The theory of sound. Courier Corporation, New York 15. Zhen C, Xu J (2010) Multi-modal music genre classification approach. In: 3rd IEEE international conference on computer science and information technology (ICCSIT), pp 398 402 16. Gamboa H (2005) Α wave. Wikipedia. http://en.wikipedia.org/wiki/αlpha_wave 17. http://www.csie.ntu.edu.tw/~cjlin/libsvm/ 18. http://www.mathworks.com/