IPSJ SIG Technical Report 1, Instrument Separation in Reverberant Environments Using Crystal Microphone Arrays Nobutaka ITO, 1, 2 Yu KITANO, 1

Size: px

Start display at page:

Download "IPSJ SIG Technical Report 1, Instrument Separation in Reverberant Environments Using Crystal Microphone Arrays Nobutaka ITO, 1, 2 Yu KITANO, 1"

いつやつくとの
5 years ago
Views:

1 1, Instrument Separation in Reverberant Environments Using Crystal Microphone Arrays Nobutaka ITO, 1, 2 Yu KITANO, 1 Nobutaka ONO 1 and Shigeki SAGAYAMA 1 This paper deals with instrument separation from a music signal recorded with a microphone array in a reverberant environment. To this end, we apply our signal processing method for suppressing diffuse noise, which utilizes symmetrical microphone arrangements called crystal microphone arrays. We show some results of experiments where the proposed method is applied to instrument separation in a reverberant environment Graduate School of Information Science and Technology, The University of Tokyo 2 METISS group, IRISA-INRIA, France 1) / 2) 1 3) Wiener 4) Wiener 3),5),6) 7) c 2009 Information Processing Society of Japan

2 T H Hermite τ Z ω R α(τ, ω) β(τ, ω) φ αβ (τ, ω), E[α(τ, ω)β (τ, ω)] (1) (τ, ω) Φ (τ, ω), E[ (τ, ω) H (τ, ω)] (2) E[ ] 2.2 s(τ, ω) x m(τ, ω) v m(τ, ω) m s(τ, ω) x m (τ, ω) d m (ω) d m(ω) = e jωδ m (3) δ m m x m (τ, ω) = s(τ, ω)d m (ω) + v m (τ, ω) (4) x(τ, ω) = s(τ, ω)d(ω) + v(τ, ω) (5) x(τ, ω), [x 1 (τ, ω) x 2 (τ, ω)... x M (τ, ω)] T (6) d(ω), [d 1 (ω) d 2 (ω)... d M (ω)] T (7) v(τ, ω), [v 1(τ, ω) v 2(τ, ω)... v M (τ, ω)] T (8) M s(τ, ω) v(τ, ω) 2.3 Wiener 4),8) ŝ(τ, ω), w H (τ, ω)x(τ, ω) (9) E [ ŝ(τ, ω) s(τ, ω) 2] (10) ŝ o (τ, ω), E[s(τ, ω)x H (τ, ω)]e[x(τ, ω)x H (τ, ω)] 1 x(τ, ω) (11) = φ ss(τ, ω)d H (ω)φ 1 xx(τ, ω)x(τ, ω) (12) (1) (2) φ ss (τ, ω) Φ xx(τ, ω) φ ss (τ, ω), E [ s(τ, ω) 2] (13) Φ xx (τ, ω), E[x(τ, ω)x H (τ, ω)] (14) (12) ŝ o(τ, ω) y(τ, ω), dh (ω)φ 1 xx(τ, ω)x(τ, ω) (15) d H (ω)φ 1 y(τ, ω) φ yy(τ, ω) = dh (ω)φ 1 xx(τ, ω)φ xx (τ, ω)φ 1 [d H (ω)φ 1 (16) ] 2 1 = (17) d H (ω)φ 1 (12) (17) ŝ o (τ, ω) 4),8) ŝ o(τ, ω) = φ ss(τ, ω) dh (ω)φ 1 xx(τ, ω)x(τ, ω) φ yy (τ, ω) d }{{} H (ω)φ 1 }{{}, p(τ, ω) = y(τ, ω) ŝ o (τ, ω) y(τ, ω) Wiener p(τ, ω) 2.4 φ ss(τ, ω) Wiener p(τ, ω) φ ss (τ, ω) x(τ, ω) 7) (18) 2 c 2009 Information Processing Society of Japan

3 s(τ, ω) v(τ, ω) Φ xx (τ, ω) = φ ss (τ, ω)d(ω)d H (ω) + Φ vv (τ, ω) (19) Φ xx(τ, ω) φ ss(τ, ω) d(ω)d H (ω) d(ω) Φ vv (τ, ω) Φ vv(τ, ω) Φ xx(τ, ω) φ ss(τ, ω) Φ vv (τ, ω) Φ vv (τ, ω) (a) v m (τ, ω) φ vm v m (τ, ω) φ v1 v 1 (τ, ω) = φ v2 v 2 (τ, ω) = = φ vm v M (τ, ω) (20) (b) v m (τ, ω) φ vm v n (τ, ω) r mn m n r mn = r kl φ vm v n (τ, ω) = φ vk v l (τ, ω) (21) (1) (2) φ vmvn (τ, ω) Φ vv (τ, ω) (m, n) (a) Φ vv (τ, ω) (b) Φ vv (τ, ω) Φ vv(τ, ω) 1 Φ vv (τ, ω) 9),10) 1 1 (a) φ v1 v 1 (τ, ω) = φ v2 v 2 (τ, ω) = φ v3 v 3 (τ, ω) = φ v4 v 4 (τ, ω) =: α(τ, ω) (22) r 12 = r 21 = r 34 = r 43 (23) r 13 = r 31 = r 24 = r 42 (24) r 14 = r 41 = r 23 = r 32 (25) (b) φ v1 v 2 (τ, ω) = φ v2 v 1 (τ, ω) = φ v3 v 4 (τ, ω) = φ v4 v 3 (τ, ω) =: β(τ, ω) (26) φ v1 v 3 (τ, ω) = φ v3 v 1 (τ, ω) = φ v2 v 4 (τ, ω) = φ v4 v 2 (τ, ω) =: γ(τ, ω) (27) φ v1 v 4 (τ, ω) = φ v4 v 1 (τ, ω) = φ v2 v 3 (τ, ω) = φ v3 v 2 (τ, ω) =: δ(τ, ω) (28) 3 c 2009 Information Processing Society of Japan

4 (22) (26) (27) (28) Φ vv(τ, ω) α(τ, ω) β(τ, ω) γ(τ, ω) δ(τ, ω) β(τ, ω) α(τ, ω) δ(τ, ω) γ(τ, ω) Φ vv (τ, ω) = γ(τ, ω) δ(τ, ω) α(τ, ω) β(τ, ω) δ(τ, ω) γ(τ, ω) β(τ, ω) α(τ, ω) α(τ, ω) β(τ, ω) γ(τ, ω) δ(τ, ω) ) Φ vv (τ, ω) U (19) (29) (30) U H Φ xx (τ, ω)u = φ ss (τ, ω)u H d(ω)d H (ω)u + U H Φ vv (τ, ω)u (31) U H Φ vv(τ, ω)u U H Φ xx(τ, ω)u u H mφ xx (τ, ω)u n = φ ss (τ, ω)u H md(ω)d H (ω)u n (m n) (32) u m U m u H mφ xx (τ, ω)u n u H md(ω)d H (ω)u n (32) φ ss(τ, ω) φ ss (τ, ω) [ ] u H m d(ω)d H u H (ω)u n m Φ xx(τ, ω)u n ˆφ ss (τ, ω), m n u H m d(ω)d H (ω)u n 2 m n 2.5 Wiener (33) φ ss (τ, ω) Wiener (33) ˆp(τ, ω), ˆφ ss(τ, ω) (34) ˆφ yy (τ, ω) ˆφ yy (τ, ω) Zelinski 3) φ yy (τ, ω) ˆφ yy(τ, ω), 1 M φ xm x M m (τ, ω) (35) m=1 y(τ, ω) φ yy(τ, ω) (35) (35) p(τ, ω) 0 p(τ, ω) 1 ˆp(τ, ω) 3. ˆp(τ, ω) 0, if ˆp(τ, ω) < 0 ˆp(τ, ω) 1, if ˆp(τ, ω) > ) RT ms SiSEC 12) 6 s 16 khz (15) (34) Fourier Hamming (15) Φ xx x(τ, ω)x H (τ, ω) (34) (33) Φ xx (τ, ω) (35) φ xm x m (τ, ω) 32 x(τ, ω)x H (τ, ω) x m(τ, ω)x n(τ, ω) (15) (33) d(ω) SN 4) LSD 13) 1 2 SN (36) 4 c 2009 Information Processing Society of Japan

2 @mic 1 LSD(dB) 1 2 3 (@mic 1) 10.1 9.2 7.8 4.8 4.8 6.2 4.6 6.1 6.5 2 1 @mic 1 SN (db) 1 2 3 (@mic 1) 4.4 8.9 1.2 4.1 1.1 4.6 10.8 3.

5 1 LSD(dB) (@mic 1) SN (db) (@mic 1) (a) (b) LSD LSD SN 3 3@mic 1 3 (c) (d) 3 3 (c) 3 (a) 3 1 (c) (d) (d) 5 c 2009 Information Processing Society of Japan

6 db SN 1) M. Brandstein and D. Ward, Microphone Arrays: Signal Processing Techniques and Applications. Berlin: Springer-Verlag, ) 3) R. Zelinski, A microphone array with adaptive post-filtering for noise reduction in reverberant rooms, in Proc. ICASSP 88, New York, Apr. 1988, pp ) K. U. Simmer, J. Bitzer, and C. Marro, Post-filtering techniques, in Microphone Arrays: Signal Processing Techniques and Applications, M. Brandstein and D. Ward, Eds. Berlin: Springer-Verlag, 2001, ch. 3, pp ) I. A. McCowan and H. Bourlard, Microphone array post-filter based on noise field coherence, IEEE Trans. Speech Audio Process., vol. 11, no. 6, pp , Nov ) S. Lefkimmiatis and P. Maragos, A generalized estimation approach for linear and nonlinear microphone array post-filters, Speech Commun., vol.49, no. 7 8, pp , July Aug ) N. Ito, N. Ono, and S. Sagayama, A blind noise decorrelation approach with crystal arrays on designing post-filters for diffuse noise suppression, in Proc. ICASSP 2008, Las Vegas, Apr. 2008, pp ) H. L. Van Trees, Optimum Array Processing. New York: John Wiley & Sons, ) H. Shimizu, N. Ono, K. Matsumoto, and S. Sagayama, Isotropic noise suppression in the power spectrum domain by symmetric microphone arrays, in Proc. WASPAA, New Paltz, NY, Oct. 2007, pp ) N. Ono, N. Ito, and S. Sagayama, Five classes of crystal arrays for blind decorrelation of diffuse noise, in Proc. SAM, Darmstadt, Germany, Jul. 2008, pp ) J. A. Allen, D. A. Berkeley, Image method for efficiently simulating small-room acoustics, JASA, vol. 65, no. 4, pp , Apr ) 13) I. Cohen, Multichannel Post-Filtering in Nonstationary Noise Environments, IEEE Trans. Signal Process., vol. 52, no. 5, pp , May CREST 6 c 2009 Information Processing Society of Japan

H(ω) = ( G H (ω)g(ω) ) 1 G H (ω) (6) 2 H 11 (ω) H 1N (ω) H(ω)= (2) H M1 (ω) H MN (ω) [ X(ω)= X 1 (ω) X 2 (ω) X N (ω) ] T (3)

H(ω) = ( G H (ω)g(ω) ) 1 G H (ω) (6) 2 H 11 (ω) H 1N (ω) H(ω)= (2) H M1 (ω) H MN (ω) [ X(ω)= X 1 (ω) X 2 (ω) X N (ω) ] T (3) 72 12 2016 pp. 777 782 777 * 43.60.Pt; 43.38.Md; 43.60.Sx 1. 1 2 [1 8] Flexible acoustic interface based on 3D sound reproduction. Yosuke Tatekura (Shizuoka University, Hamamatsu, 432 8561) 2. 2.1 3 M