本論文主要探討摺積混合下盲音源分離的演算法，希望能解決實際環境下，語音訊號處理中所描述的雞尾酒派對問題。本論文利用相關性來量測獨立性，由於相關性在統計學上是屬於二階的統計特性，表現此統計特性的方式為一個對稱的方陣，稱之為共變異矩陣。實際運算時先將訊號轉至頻域，接著計算訊號的交頻譜來表現語音訊號的二階統計特性。利用聯合對角化演算法對每個離散頻率計算解混合矩陣，使分離出來的訊號能夠盡可能的不相關。為了能有較佳的分離效果，我們利用語音訊號在時頻域上有稀疏性的假設，估計出某時頻點應該由哪位說話人獨占，且利用共變異矩陣的特徵值來近似原說話人與干擾訊號在此點上的能量比。接著，利用此比值建立一組遮罩，將以不相關的訊號通過這組遮罩來更加壓抑干擾訊號。為了避免遮罩產生分離訊號之頻譜的不連續性，我們將此遮罩轉換至倒頻域，在低倒頻率的部分用較小的平滑係數處理，藉此保持分離訊號的諧振，在高倒頻率的部分用較大的係數平滑之，讓分離出來的訊號之頻譜不會過於不連續。實驗時嘗試將兩個麥克風在實際環境中錄到的雙人混合語音分離開，將只有不相關的訊號和通過遮罩後的加強訊號作主觀評量，發現透過遮罩的確可以使干擾訊號更加的被壓抑。

[1] A.Ziehe, P.Laskov, K.R.Mueller, and G.Nolte, \A fast algorithm for joint diagonalization
with non-orthogonal transformations and its application to blind
source separation," Journal of Machine Learning Research 5, pp. 777{800, 2004.
[2] R. Vollgraf and K. Obermayer, \Quadratic optimization for simultaneous matrix
diagonalization," IEEE Trans. Signal Process., vol. 54, pp. 3270{3278.
[3] D. Pham, \Joint approximate diagonalization of positive denite hermitian matrices,"
SIAM J. on Matrix Anal. and Appl, vol. 22, pp. 1136{1152, 2000.
[4] A. Jourjine, S. Rickard, and O. Yzlmaz, \Blind separation of disjoint orthogonal
signals: Demixing n sources from 2 mixtures," Proc. IEEE ICASSP, pp. 2985{
2988, 2000.
[5] O.Yilmaz, A.Jourjine, and S.Rickard, \Blind separation of speech mixtures via
time-frequency masking," Proc. IEEE TSP, vol. 52, 2004.
[6] K. Pearson, \On lines and planes of closest t to systems of points in space,"
Philosophical Magazine 2, vol. 6, p. 559572, 1901.
[7] P. Comon, \Independent component analysis, a new concept?," Signal Process-
ing, vol. 36, pp. 287{314, 1994.
[8] A. J. Bell and T. J. Sejnowski, \An information-maximization approach to blind
separation and blind deconvolution," Neural Computation, vol. 7, pp. 1129{
1159, 1995.
[9] S.Amari, A.Cichocki, and H.H.Yang, \A new learning algorithm for blind signal
separation," Advance in Neural Infomation Processing Systems, vol. 8, pp. 757{
763, 1996.
[10] A. Hyvrinen, \Fast and robust xed-point algorithms for independent component
analysis," IEEE Trans. on Neural Networks, vol. 10(3), pp. 626{634, 1999.
53
[11] A.Blin, S.Araki, and S.Makino, \Underdetermined blind separation of convolutive
mixtures of speech using time-frequency mask and mixing matrix estimation,"
IEICE Trans. Fundamentals, vol. E88-A, pp. 1693{1700, 2005.
[12] S.Winter, H.Sawada, S.Araki, and S.Makino, \Overcomplete bss for convolutive
mixtures based on hierarchical clustering," Proc.ICA2004, pp. 652{660, 2004.
[13] J. Cardoso and A. Souloumiac, \Blind beamforming for non-gaussian signals,"
IEE Proceedings-F, vol. 140, no. 3, pp. 362{370, 1993.
[14] P. Comon, \Tensor diagonalization, a useful tool in signal processing,," Proc.
10th IFAC Symp. Syst. Iden, vol. 1, pp. 77{82, 1994.
[15] S. Choi, S. ichi Amari, A. Cichocki, and R. wen Liu, \Natural gradient learning
with a nonholonomic constraint for blind deconvolution of multiple channels,"
Proceedings of the International Workshop on Independent Component
and Blind Souece Separation(ICA'99), pp. 371{376, 1999.
[16] K. Hao Shen Huper, \Newton-like methods for parallel independent component
analysis," Proceedings of the 2006 16th IEEE Signal Processing Society
Workshop on Machine Learning for Signal Processing, pp. 283{288, 2006.
[17] R. Mukai, H. Sawada, S. Araki, and S. Makino, \Real-time blind source separation
and doa estimation using small 3-d microphone array," Proc. WAS-
PAA2005, pp. 9{12, 2005.
[18] D. T. Pham, \Blind separation of convolutive audio mixtures using nonstationarity,"
Proceedings of ICA 2003 conference, pp. 975{980, 2003.
[19] K. Matsnoka, \Minimal distortion pirinciple for blind source separation," Pro-
ceedings of the 41st SICE Annual Conference,, vol. 4, pp. 2138{2143, 2002.
[20] N.Madhu, C.Breithaupt, and R.Martin, \Temporal smoothing of spectral masks
in the cepstral domain for speech separation," ICASSP2008, pp. 45{48, 2008.