文件列表:

New Folder\diagdual.m (1093, 2001-05-24)
New Folder\exp_gg.m (5900, 2002-07-31)
New Folder\gabmat.m (2427, 2002-04-12)
New Folder\gabt_fft.m (869, 2002-05-26)
New Folder\gengamma.m (906, 2002-06-28)
New Folder\gibbsgab_kr.m (5698, 2002-07-31)
New Folder\gibbsgab_w.m (4985, 2002-07-01)
New Folder\gmat.m (1855, 2002-06-14)
New Folder\nips_gabreg.tar.gz (0, 2002-07-31)
New Folder\snrmeas.m (994, 2001-07-31)
New Folder\snr_test.m (1881, 2002-07-30)
New Folder\stsa_comp.m (4521, 2002-07-30)
New Folder\stsa_init.m (2091, 2002-07-02)
New Folder\test.wav (10284, 2002-07-30)
New Folder\wilmat.m (1324, 2002-05-28)
New Folder\wmat.m (640, 2002-06-09)
New Folder\w_eval.m (1320, 2002-07-30)
New Folder (0, 2006-06-30)

To reproduce the NIPS 2002 Submission Results, run the following: E = exp_gg('test',[1 512*10],10,2000,1000); is the main call, for example for 10 dB SNR degradation. To reproduce the reported results, of course, you will have to also include the appropriate random number generator seeds, obtained from the posted data. See the code for exp_gg ('help exp_gg' or 'type exp_gg') for details. ******* Tests of Noise Reduction Gain ******* Note: All of the code in this section is set to write files to d:\temp\ snr_test - in script form, you will have to modify the code slightly to ensure that the correct random number seeds are used. Alternatively, you can reproduce the individual results one-by-one using exp_gg as specified above. Set the algorithm choice to 'kr' in exp_gg.m, as well as the initialisation desired ('rand', 'noisy', etc.). Then check that the priors are set appropriately: kap = zeros(M_max,N); kap(:) = 10^-12; ups = kap; for diffuse priors and kap = zeros(M_max,N); kap(:) = 2*10^0; ups_m = sca*(1*(kap(1)-1)).*[1./(1:M_max)']; ups = repmat(ups_m,1,N); for frequency-dependent priors. To compare with STSA-based techniques, you will need to download and install the STSA toolbox, http://www-sigproc.eng.cam.ac.uk/~pjw47/ftp/STSA-Toolbox-0-01.tar.gz . Then stsa_comp.m will load in the appropriate files generated by gibbstest.m from d:\temp. (Assuming that the stsa files are now in your matlab path.) You will have to run this manually for each algorithm you want to compare; at the moment it is set to stsa_algorithm = 'specsub', or spectral subtraction, and method = 'wiener'. I also compared it with `specsub' and method = 'magnitude', as well as stsa_algorithm = `ephraim'. ******* Wilson Basis Experiments ******* Set the algorithm choice to 'w' in exp_gg.m, as well as the initialisation desired ('rand', 'noisy', etc.). Then check that the priors are set appropriately, as above. E = exp_gg('test',[1 512*10],10,2000,1000); will then run a Wilson experiment. (To reproduce the reported results, of course, you will have to also include the appropriate random number generator seeds, obtained from the posted data.) w_eval(E,F), where F is another run (say with frequency-dependent priors) will then plot some results from these experiments, as in the NIPS submission paper. Patrick J. Wolfe, p.wolfe@ieee.org, July 2002

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：