IPSJ-MUS

Size: px

Start display at page:

Download "IPSJ-MUS"

しょうすけえいさか
6 years ago
Views:

1 Vol.29-MUS-81 No.2 29/7/ ground-truth RWC Method for Calculating the Subjective-based Music Similarity Measure Yusuke Hiraga, 1 Yasunori Ohishi 2 and Kazuya Takeda 1 In this paper, we propose a method for representing a subjective music similarity by several acoustic similarities. For establishing ground-truth similarity, we are collecting subjective similarity data for pairs of 22 tracks of RWC music database. A preliminary experiment is conducted for subjectivesimilarity data collected from 16 subjects. The subjective similarity is associated with the acoustic similarities using linear regression model. The result indicates that weighted sum of several acoustic similarity measures well predict the subjective similarity. 1 Graduate School of Information Science, Nagoya University 2 NTT NTT Communication Scince Laboratories, NTT Corporation 1. 1)2) 3)4) Mel-Frequency Cepstrum Coefficient: MFCC 5)6)7) Hoashi 3) Vignoli Pauws 4) 5 Lampropoulos 8) ground-truth ISMIR MIREX IMIRSEL MIREX Audio Music Similarity and Retrieval Symbolic Melodic Similarity Audio Mood Classification Evalutron 6 9) 1) query 1 candidates candidate query 3 NOT Similar, Somewhat Similar, VERY Similar c 29 Information Processing Society of Japan

Vol.29-MUS-81 No.2 29/7/29 1 IMIRSEL Audio Music Similarity and Retrieval 7 1 RWC 11)12) 22 ground-truth 2. 2.1 RWC 11) 1 12) 12 22 1) 15 2.

2 Vol.29-MUS-81 No.2 29/7/29 1 IMIRSEL Audio Music Similarity and Retrieval 7 1 RWC 11)12) 22 ground-truth RWC 11) 1 12) ) ground-truth web 1 web Flash FlashPlayer PC 2 RWC / 15 MP3, 192kbps, CBR, joint-stereo 2 c 29 Information Processing Society of Japan

3 Vol.29-MUS-81 No.2 29/7/ 数アペ曲楽似ていると評価された回数 2 度頻るれさクッェチ C1 C2 C3 C4 C5 C6 C7 評価基準 1 8 数アペ曲楽似ていると評価された回数 3 度頻るれさクッェチ評価基準 C1 C2 C3 C4 C5 C6 C 被験者番号 2 3 C1: C2: C3: C4: C5: C6: C7: 4 5 C1: C2: C3: C4: C5: C6: C7: % y y = [y 1, y 2,..., y N ] T (1) i y i = i (2) y i i 5 3 i y i y i = 3/5 = STFT STFT 3 3 STFT 44.1kHz 5ms 25ms t 楽曲音響信号特徴抽出 6 LBG アルゴリズム音響特徴ベクトルインテンシティ特徴ベクトル ( フレーム毎に算出 ) コードブックベクトル量子化頻度分布 ( 音響ベクトル ) 3 c 29 Information Processing Society of Japan

4 13) 13) MFCC 4 MFCC 1 L k x(t k) k= L x(t) = L (3) k= L k2 L = 2 STFT 16kHz I(t) D j (t) f s 2 I(t) = X(t, f) (4) f= D j(t) = 1 I(t) H j f=l j X(t, f) (5) f s X(t, f) t f j L j H j [, f ), 2 n [ f 2 n, f 2 n 1 ),..., [ f 2 2, f 2 1 ] n n = 7 16 ( 1 ) (6) f c(t) = fs /2 f= X(t, f) f f /2 X(t, f) (7) f= ( 2 ) fs /2 B(t) = (f f f= c) 2 X 2 (t, f) fs/2 f= X2 (t, f) ( 3 ) f r 95% f r f= f s /2 X(t, f) =.95 X(t, f) (9) f= ( 4 ) F = f s/2 ( ) 2 X(t, f) X(t 1, f) (1) f= 4 j X j,1, X j,2,..., X j,n X j,1, X j,2,..., X j,n 3 ( 1 ) P j = log 1 αn ( 2 ) V j = log 1 αn ( 3 ) αn k=1 αn k=1 Vol.29-MUS-81 No.2 29/7/29 (8) X j,k (11) X j,n k+1 (12) C j = P j V j (13) α α =.2 4 c 29 Information Processing Society of Japan

5 7 3 7 = 21 (4 + 21) 2 = 5 MFCC MFCC 16 MFCC 32 HPCP 14) 12 A G # MIRtoolbox1.1 15) ) LBG y 3.2 m i d mi ŷ ŷ = Φ(Φ T Φ) 1 Φ T y (14) 1 d d 1N Φ = (15) 1 d M1... d MN M, N M = 4 M = N = leave-one-out Root Mean Square Error: RMSE y ŷ RMSE 1 RMSE = (y i ŷ i ) N 2 (16) i y RMSE 4.2 RMSE 4 RMSE 5 7 r r = Vol.29-MUS-81 No.2 29/7/29 5 c 29 Information Processing Society of Japan

6 1) Rauber A., Pampalk, E., and Merkl, D., Using Psycho-Acoustic Models and Self-Organizing Maps to Create a Hierarchical Structure of Music by Sound Similarity, Proc. ISMIR, pp.71 8 (22). 2) Goto, M. and Goto, T., Musicream: New Music Playback Interface for Streaming, Sticking, and Recalling Musical Pieces, Proc. ISMIR (25). 3) Hoashi, K., Matsumoto, K. and Inoue, N., Personalization of User Profiles for 4 RMSE MFCC y i y i < y i < y i MFCC MFCC Vol.29-MUS-81 No.2 29/7/29 Content-based Music Retrieval on Relevance Feedback, Proceedings of ACM Multimedia 23, pp (23). 4) Vignoli, F. and Pauws, S., A Music Retrieval System Based on User-Driven Similarity and its Evaluation, Proc. ISMIR (25). 5) McKinney, M.F. and Breebaart, J., Features for Audio and Music Classification, Proc. ISMIR (23). 6) Pampalk, E., Flexer, A. and Widmer, G., Improvements of Audio-Based Music Similarity and Genre Classification, Proc. ISMIR (25). 7) Lidy,T. and Rauber, A., Evaluation of Feature Extractors and Psycho-Acoustic Transformations for Music Genre Classification, Proc. ISMIR, pp (25). 8) Lampropoulos, A.S., Sotiropoulos, D.N. and Tsihrintzis, G.A., Individualization of Music Similarity Perception via Feature Subset Selection, IEEE, International Conference on Systems, Man and Cybernetics 24 (24). 9) Gruzd, A.A., Downie, J.S., Jones, M.C. and Lee, J.H., Evalutron 6: Collecting Music Relevance Judgments, Proceedings of the 7th ACM/IEEE-CS joint conference on Digital libraries (27). 1) Jones, M.C., Downie, J.S., and Ehmann, A.F., Human similarity judgements: Implications for the design of formal evaluations, Proc. ISMIR27 (27). 11) Goto, M., Hashiguchi, H., Nishimura, T. and Oka, R., RWC Music Database: Popular, Classical, and Jazz Music Databases, ISMIR, pp (22). 12) Goto, M., Hashiguchi, H., Nishimura, T. and Oka, R., RWC Music Database: Music Genre Database and Musical Instrument Sound Database, ISMIR, pp (23). 13) Lu, L., Liu, D. and Zhang, H.-J., Automatic Mood Detection and Tracking of Music Audio Signals, IEEE Transactions on audio, speech, and language processing, vol. 14, No. 1 (26). 14) Gómez, E., Tonal description of music audio signals, Ph.D. thesis, UPF, Barcelona, Spain (26). 15) Lartillot, O. and Toiviainen, P., MIR in Matlab (II): A toolbox for musical feature extraction from audio, Proc. ISMIR (27). 16) Takahashi, R., Ohishi, Y., Kitaoka, N. and Takeda, K., Building and Combining Document and Music Spaces for Music Query-By-Webpage System, INTER- SPEECH 28, pp (28). 6 c 29 Information Processing Society of Japan

IPSJ SIG Technical Report Vol.2011-MUS-91 No /7/ , 3 1 Design and Implementation on a System for Learning Songs by Presenting Musical St

IPSJ SIG Technical Report Vol.2011-MUS-91 No /7/ , 3 1 Design and Implementation on a System for Learning Songs by Presenting Musical St 1 2 1, 3 1 Design and Implementation on a System for Learning Songs by Presenting Musical Structures based on Phrase Similarity Yuma Ito, 1 Yoshinari Takegawa, 2 Tsutomu Terada 1, 3 and Masahiko Tsukamoto