1,a) 1,b) J-POP 100 The Algorithm to Extract Characteristic Chord Progression Extended the Sequential Pattern Mining Shinohara Toru 1,a) Numao Masayuki 1,b) Abstract: Chord is an important element of music because it supports melody and harmony of whole music. However, it is difficult to compose the chord progression using the music theory efficiently for the beginner of music composer. In addition, it is conceivable that professional composers often regard specific chord progression as important, such as cadenza pattern, and they imitate the chord progression used by the other composers. In this paper, we suggest the pattern mining method specified for music information processing by extending the sequential pattern mining, one of the frequent pattern mining so that it can scan the data in opposite order and it can process pseudo items using independent threshold. We report that chord progressions representing character of composers and cadenza patterns are extracted after we apply the method from 100 musics including J-pop, rock and European music. Keywords: Sequential Pattern Mining, Music Information Retrieval, Computer-aided Composition 1. DeskTop Music DTM 1 Graduate School of Informatics and Engineering, The Uniersity of Electro-Communication a) toru.shinohara@uec.ac.jp b) numao@cs.uec.ac.jp c 2014 Information Processing Society of Japan 1
2 [1] [2] Jiménez MusicXML motifs [3] Ogihara N-gram [4] Pérez N-gram [5] N-gram [6] [7] 2. 2.1 [8] 1995 Agrawal Σ = {i 1, i 2,..., i m } i j S S = {< a 1, a 2,..., a r > a i Σ} T = {S 1, S 2,..., S n } α β α β α β T α α support support(α) = {α α S} (1) α β α =< a 1,..., a l >, β =< a 1,..., a l, a > α β α β confidence confidence(α β) = support(β) support(α) (2) support, minimum support ζ support(α) ζ α α 2.2 2001 Pei PrefixSpan[9] prefix projection S =< a 1, a 2,, a l > a a 1 a, a 2, a,, a j 1 a, a j = a j(1 < j < l) < a 1,, a j > S a prefix prefix(s, a) < a j+1,, a m > S a postfix postfix(s, a) T a projected database T a T a = {S (S T ) (S = postfix(s, a)) (S ϵ)}(3) PrefixSpan ( 1 ) T 1 support c 2014 Information Processing Society of Japan 2
( 2 ) support 2 ( 3 ) support support PrefixSpan [10][11] 3. 3.1 i j c j S M T D D P m m P support support(p ) P ζ M P t t C, Am I VII support support support 3.2 P =< c 1,..., c m > c 1 c m c m c 1 c j = I < V, I > < IV, I > confidence α =< a l,..., a 1 >, β =< a, a l,..., a 1 > α β confidence confidence confidence confidence confidence 3.3 1 Σ M M N = {pseudo 1,..., pseudo N } D M =< c 1,..., c m > M =< pseudo 1,..., pseudo N, c 1,..., c m > M =< c 1,..., c m, pseudo N,..., pseudo 1 > D i j pseudo k M pseudo k (c 1 c m ) 0 pseudo k 1 1 c j support support pseudo k support ζ c 2014 Information Processing Society of Japan 3
support ζ pseudok ζ pseudok ζ pseudo k ζ ζ pseudok pseudo k comp k 3.4 < V, I, IV, I, IV, I, IV, I, IV > (4) p ϵ, q ϵ P 0 + 1 P =< q pp+ > (5) P =< V, I, IV, I, IV, I, IV, I, IV > p =< I, IV >, q =< V > < V, I, IV, I > a < a, a > 3.5 support ζ support ζ pseudok 2 0 4. 4.1 4.1.1 confidence confidence < i 1,..., i m > i 5 χ 2 support ζ = 2, ζ compk = 1 P 1 =< I > P 2 =< comp k, I > c j support I confidence support confidence 1.00 χ 2 YAMAHA MIDI XF J-POP 100 XF 4.1.2 < I > 1 s = support(< comp j >) 1 confidence 1 < I > confidence confidence 1.00 2 s = support(< comp j >) c 2014 Information Processing Society of Japan 4
1 < I > confidence comp j s χ 2 3 177.62 1.68 10 10 < I, V/VII > 0.21154 0.61111 3 178.40 1.33 10 10 < I, V/VII > 0.31818 1.00000 Utada Hikaru 3 262.49 6.59 10 23 < I, VI m > 0.52941 0.50000 3 303.47 1.47 10 29 < I, III/ VI > 0.39394 0.61905 2 212.74 2.38 10 15 < I, V > 0.31081 0.47917 2 222.21 9.79 10 17 < I, VI m > 0.61538 0.60377 2 467.75 1.64 10 58 < I, V 7sus4 /II > 0.18519 0.55556 2 502.91 3.59 10 66 < I, I M9 /III > 0.44444 0.88889 - - - - < I, V/VII > 0.14491 0.71141 2 confidence 1.00 comp j P s χ 2 < I 7, IV M7 > 3 6.60 4.72 10 1 < V 6, IV > 3 10.18 1.71 10 2 Utada Hikaru < VII aug, VII 6 > 3 0.00 1.00 < VI m7, VI m7 /V > 3 96.70 3.35 10 6 < I add9 /III, IV M9 > 2 9.91 2.71 10 1 < III/ VI, VI m > 2 16.36 9.55 10 4 < II 7, II m7 > 2 35.94 3.31 10 4 < I M9 /III, IV > 2 0.00 1.00 χ 2 = 0 D D s = 2 s = 3 4.2 4.2.1 4.1 P 1 =< I > P 2 =< I, comp k > c j support 4.1 < V, I > 4.2.2 3 s = support(< comp j >) confidence 3 confidence < V, I > confidence 5. c 2014 Information Processing Society of Japan 5
3 < I > confidence comp j s χ 2 3 115.39 6.80 10 7 < V, I > 0.50000 0.28846 3 218.85 1.13 10 22 < IV, I > 0.48148 0.29545 Utada Hikaru 3 429.03 3.32 10 61 < II m7, I > 1.00000 0.52941 3 62.76 1.25 10 1 < V, I > 0.50000 0.63636 2 150.36 9.83 10 12 < IV, I > 0.84375 0.36486 2 50.53 4.92 10 1 < V, I > 0.60938 0.75000 2 346.42 1.57 10 45 < V 7sus4, I > 0.80769 0.38889 2 231.29 8.61 10 25 < II m7 /V, I > 0.50000 0.44444 - - - - < V, I > 0.36371 0.31100 I I M7 I M9 I (1995). [9] Pei, J., Han, J., Mortazavi-Asl, B., Pinto, H., Chen, Q., Dayal, U. and Hsu, M.: PrefixSpan: Mining Sequential Patterns Efficiently by Prefix-Projected Pattern Growth, ICDE 2001, April, pp. 215 24 (2001). [10] Kitakami, H., Kanbara, T., Mori, Y., Kuroki, S. and Yamazaki, Y.: Modified PrefixSpan method for motif discovery in sequence databases, PRICAI 2002: Trends in Artificial Intelligence, Springer, pp. 482 491 (2002). [11] Antunes, C. and Oliveira, A. L.: Generalization of pattern-growth methods for sequential pattern mining with gap constraints, Machine Learning and Data Mining in Pattern Recognition, Springer, pp. 239 251 (2003). [1] Cambouropoulos, E., Crawford, T. and Iliopoulos, C.: Pattern processing in melodic sequences: Challenges, caveats and prospects, Computers and the Humanities, Vol. 35, No. 1, pp. 9 21 (2001). [2] DEIM Forum 2012 F6-4 (2012). [3] Jiménez, A., Molina-Solana, M., Berzal, F. and Fajardo, W.: Mining transposed motifs in music, Journal of Intelligent Information Systems, Vol. 36, No. 1, pp. 99 115 (2011). [4] Ogihara, M. and Li, T.: N-gram chord profiles for composer style representation, Proceedings of the International Conference on Music Information Retrieval (IS- MIR) 2008-Session 5d-MIR Methods (2008). [5] Pérez-Sancho, C., Rizo, D. and Inesta, J. M.: Genre classification using chords and stochastic language models, Connection science, Vol. 21, No. 2-3, pp. 145 159 (2009). [6] Scholz, R., Vincent, E. and Bimbot, F.: Robust modeling of musical chord sequences using probabilistic N-grams, Acoustics, Speech and Signal Processing, 2009. ICASSP 2009. IEEE International Conference on, IEEE, pp. 53 56 (2009). [7] 75 pp. 2 285 2 286 (2013). [8] Agrawal, R. and Srikant, R.: Mining Sequential Patterns, Data Engineering, 1995. Proceedings of the Eleventh International Conference on, IEEE, pp. 3 14 c 2014 Information Processing Society of Japan 6