文件列表:

AdjustWeights.m (547, 2010-09-23)
barbi.p (1951, 2011-07-20)
befica.p (3479, 2011-07-20)
bssdeconv.p (1602, 2011-07-20)
compweights4d.p (458, 2011-07-20)
efica.p (3274, 2011-07-20)
gfilter.mat (1019, 2009-08-11)
hclus.p (792, 2011-07-20)
maxxcorr2.p (284, 2011-07-20)
playrec2.dll (126976, 2009-07-03)
projections4.p (638, 2011-07-20)
projections7.p (352, 2011-07-20)
projections8.p (343, 2011-07-20)
qmf.p (150, 2011-07-20)
qmfanalysis.p (163, 2011-07-20)
qmfsynthesis.p (210, 2011-07-20)
RFCM2.p (1570, 2011-07-20)
SBpermute3.p (596, 2011-07-20)
tabcd.p (2646, 2011-08-04)
tbgsep.p (1122, 2011-07-20)
tddeconv.p (4770, 2011-07-20)
treeanalysis.p (218, 2011-07-20)
treesynthesis.p (231, 2011-07-20)
zoomfft.p (629, 2011-07-20)

This is the time-domain blind audio source separation method called T-ABCD by Zbynìk Koldovsk, Petr Tichavsk and Jií Málek. release: July 20, 2011 Instalation and execution ------------------------- unzip the package into a directory, run Matlab, load microphone recordings into rows of a matrix x, and type (in Matlab) >> tddeconv(x) Usage of GUI ------------ Click signal to listen to either in the "Mixed signals" plot or in the "Separated signals" plot. Note that the correct sampling frequency should be selected. "Mixed signals" plot > Click left and right button to select, respectively, the offset of the block of data for ICA and the end of the block (the green are in the plot). "SEPARATION" button > Start the separation. "Save results" button > Save last separated signals into variable "shat". Filter length > Corresponds to the length of the separation filter in case that mu=1. This number times number of input channels equals the dimension of the observation space to that the ICA algorithm is applied [0]. The higher the length, the higher the computational burden. Select between 3-40. mu parameter > Parameter of Laguerre separating filters. mu=1 => FIR, mu~=1 => IIR. The values should be from 0 to 2 [0,9]. ICA method > Method used for the ICA decomposition. EFICA exploits nonGaussianity of sources while BGSEP (formerly reffered to as BGL) utilizes their nonstationarity. BGSEP is faster then EFICA since only second-order statistics are used; see [4],[6]. Block EFICA combines both approaches [5], thus, is computationally more expensive. Analogously, BARBI combines block-stationarity and block-spectral-diversity of sources [7]. The applicability of BARBI within T-ABCD is questionable: it might fail as well as perform best; see [8]. Similarity measure > The criterion of similarity of independent components. Projections were described in [0-3]. GCC-PHAT coefficients were first used in [11]. Clustering method > The Hierarchical clst. algorithm [1,2] or the Fuzzy C-Means algorithm [3]. Weighting type > Normal [1], Toeplitz-structure-based [12], binary (hard) [1,2], fuzzy [3], "constrained" (adjustable by user through AdjustWeights.m). Level of weighting > The alpha parameter in (10) in [1,3] that controls "hardness" of weighting in the fuzzy reconstruction mechanism. The higher the value, the higher interference suppression but at the cost of higher spectral distortion. Default value is 2. The best Signal-to-Distortion ratio (SDR) is usually achieved for alpha close to 1. # Subbands > The number of subbands. One means fullband separation (default). The subband processing approach was described in [10]. "Record 5 secs" button > Record 5 seconds of stereo signal from a standard audio device (works well with MS Windows XP; not tested elsewhere), and restart tddeconv with the recorded data. Save results > Store the separated signals into 'shat' and the estimated microphone responses (images) into 'est_resp'. Sampling frequency > sampling frequency for playback and recording Length of plot > range of x-axis in the "Mixed signals" plot Offset and Length of block for ICA > The beginning and the length of block of data where ICA decomposition is computed. References ---------- [0] Z. Koldovsk and P. Tichavsk, "Time-Domain Blind Separation of Audio Sources on the basis of a Complete ICA Decomposition of an Observation Space", accepted for publication in IEEE Trans. on Speech, Audio and Language Processing, April 2010. [1] Z. Koldovsk and P. Tichavsk, "Time-domain Blind Audio Source Separation Using Advanced Component Clustering and Reconstruction", The Joint Workshop on Hands-free Speech Communication and Microphone Arrays (HSCMA 2008), May 6-8, Trento, Italy, 2008. [2] Z. Koldovsk and P. Tichavsk, "Time-Domain Blind Audio Source Separation Using Advanced ICA Methods", Proceedings of 8th Annual Conference of the International Speech Communication Association (Interspeech 2007), pp. 846-849, August 2007. [3] J. Málek, Z. Koldovsk, J. ánsk and J. Nouza, "Enhancement of Noisy Speech Recordings via Blind Source Separation", Proceedings of the 9th Annual Conference of the International Speech Communication Association (Interspeech 2008), pp. 159-162, ISSN: 1990-9772, September 22-26, Brisbane, Australia, 2008. EFICA [4] Z. Koldovsk, P. Tichavsk and E. Oja, "Efficient Variant of Algorithm FastICA for Independent Component Analysis Attaining the Cramér-Rao Lower Bound", IEEE Trans. on Neural Networks, Vol. 17, No. 5, Sept 2006. Block EFICA [5] Z. Koldovsk, J. Málek, P. Tichavsk, Y. Deville, and S. Hosseini, "Blind Separation of Piecewise Stationary NonGaussian Sources", accepted for publication in Signal Processing, 2009. BGSEP [6] P. Tichavsk and A. Yeredor, Fast Approximate Joint Diagonalization Incorporating Weight Matrices, IEEE Tr. on Signal Processing, Vol. 57, No. 3, pp. 878-891, March 2009. BARBI [7] P. Tichavsk, A. Yeredor, and Z. Koldovsk, "A Fast Asymptotically Efficient Algorithm for Blind Separation of a Linear Mixture of Block-Wise Stationary Autoregressive Processes, ICASSP 2009, pp. 3133-3136, Taipei, Taiwan, April 2009. Further extensions and comparisons [8] Z. Koldovsk and P. Tichavsk, "A Comparison of Independent Component and Independent Subspace Analysis Algorithms," EUSIPCO 2009 , pp. 1447-1451, Glasgow, Scotland, August 24-28, 2009. [9] Z. Koldovsk, P. Tichavsk, and J. Málek, "Time-domain Blind Audio Source Separation Method Producing Separating Filters of Generalized Feedforward Structure," Proc. of LVA/ICA 2010, St. Malo, France, Sept. 2010. [10] Z. Koldovsk, P. Tichavsk, and J. Málek, "Subband Blind Audio Source Separation Using a Time-Domain Algorithm and Tree-Structured QMF Filter Bank," Proc. of LVA/ICA 2010, St. Malo, France, Sept. 2010. [11] J. Málek, Z. Koldovsk, and P. Tichavsk, "Adaptive Time-Domain Blind Separation of Speech Signals, " Proc. of LVA/ICA 2010, St. Malo, France, Sept. 2010. [12] Z. Koldovsk, J. Málek, and P. Tichavsk, "Blind Speech Separation in Time-Domain Using Block-Toeplitz Structure of Reconstructed Signal Matrices," Interspeech 2011, Florence, Italy, Aug. 2011. Links ----- http://itakura.ite.tul.cz/zbynek/downloads.htm http://si.utia.cas.cz/downloadPT.htm Contacts -------- zbynek.koldovsky_at_tul.cz tichavsk_at_utia.cas.cz jiri.malek_at_tul.cz (_at_ = @) Enjoy it! :)

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：