plateform
stringclasses
1 value
repo_name
stringlengths
13
113
name
stringlengths
3
74
ext
stringclasses
1 value
path
stringlengths
12
229
size
int64
23
843k
source_encoding
stringclasses
9 values
md5
stringlengths
32
32
text
stringlengths
23
843k
github
singaxiong/SignalGraph-master
B_frame_select.m
.m
SignalGraph-master/graph/B_frame_select.m
1,077
utf_8
1f59f4e6b016ee90119561fcb7e17d2c
% select frames from the input stream % function grad = B_frame_select(input_layer, future_layers, curr_layer) input = input_layer.a; [D,T,N] = size(input); future_grad = GetFutureGrad(future_layers, curr_layer); words = ExtractWordsFromString_v2(curr_layer.frameSelect); selectionType = words{1}; switch selectionTyp...
github
singaxiong/SignalGraph-master
B_real_imag2BFweight_beamforming_power.m
.m
SignalGraph-master/graph/B_real_imag2BFweight_beamforming_power.m
2,023
utf_8
fe6cf83395ed0445ef901a41c80b981d
function grad = B_real_imag2BFweight_beamforming_power(X, beamform_layer, after_power_layer, weight_layer, real_imag_weight) % X is the multichannel complex spectrum inputs [N,C,T,nSent] = size(X); % Y is the beamforming's output Y = beamform_layer.a; % weight is the beamforming weight weight = weight_layer.a; % future...
github
singaxiong/SignalGraph-master
B_permute.m
.m
SignalGraph-master/graph/B_permute.m
332
utf_8
db3c6c08fb377d217aaa65eae5462060
% % function [grad] = B_permute(future_layers, curr_layer) future_grad = GetFutureGrad(future_layers, curr_layer); %decide the permute order to undo the permutation of forward pass permute_order = curr_layer.permute_order; [~,reverse_permute_order] = sort(permute_order); grad = permute(future_grad, reverse_permute_o...
github
singaxiong/SignalGraph-master
B_LSTM_back2.m
.m
SignalGraph-master/graph/B_LSTM_back2.m
5,937
utf_8
811b80a3d0c792eedf01fa4adf8b8465
% Implement the back propagation of an LSTM layer. % Author: Xiong Xiao, Temasek Labs, NTU, Singapore. % Last modified: 13 Oct 2015 % function [grad, grad_W, grad_b] = B_LSTM(input_layer, LSTM_layer, future_layers) input = input_layer.a; W = LSTM_layer.W; if strcmpi(class(input), 'gpuArray'); useGPU=1; else useGPU = 0;...
github
singaxiong/SignalGraph-master
F_transpose.m
.m
SignalGraph-master/graph/F_transpose.m
125
utf_8
abca4e955980bd2bdd9d17b6caa81061
% repeat a matrix % function [output] = F_transpose(input_layer, curr_layer) input = input_layer.a; output = input'; end
github
singaxiong/SignalGraph-master
F_MVDR_spatialCov.m
.m
SignalGraph-master/graph/F_MVDR_spatialCov.m
2,538
utf_8
a4e835b6d4e66f2c5876eb469bec1751
% Estimate spatial covariance matrix for sentences using a mask. The mask % specifies speech presense probability at all time frequency locations, % with a 1 means speech present and 0 means speech absent. % function curr_layer = F_MVDR_spatialCov(input_layer, curr_layer) input = input_layer.a; fs = curr_layer.fs; fre...
github
singaxiong/SignalGraph-master
F_LSTM.m
.m
SignalGraph-master/graph/F_LSTM.m
6,371
utf_8
ae4fd89cbecfbeb277172ce98d2b4567
% Implement the forward pass of an LSTM layer. % Author: Xiong Xiao, Temasek Labs, NTU, Singapore. % Last modified: 13 Oct 2015 % function [LSTM_layer] = F_LSTM(input_layer, LSTM_layer) % % The weight matrix of LSTM is organized as follows: % W = [ W_cf W_hf W_xf; % W_cc W_hc W_xc; % W_ci W_hi W_xi; % ...
github
singaxiong/SignalGraph-master
DNN_cost_wrapper.m
.m
SignalGraph-master/graph/DNN_cost_wrapper.m
666
utf_8
bcdfa02155fd92ff73b909492421a192
% convert the trainable parameters in the graph into a vector such that we % can call standard optimization packages to optimize the network % parameters in batch mode. % function [cost, grad] = DNN_cost_wrapper(W, layer, data, para, mode) % retrieve the weights from W and assign it to the correct layers layer = NetWe...
github
singaxiong/SignalGraph-master
F_power_spectrum_split.m
.m
SignalGraph-master/graph/F_power_spectrum_split.m
402
utf_8
6e518b9588fa93e56287839546c4430c
function output = F_power_spectrum_split(input) % assume the input is a 2DxT matrix of real and imaginary parts of the % complex spectrum, where D is the number of frequency bins, % T is the number of frames in the minibatch or utterance % ral and imagineary parts of the complex Fourier spectrum is concatenated. % ...
github
singaxiong/SignalGraph-master
F_tdoa2weight.m
.m
SignalGraph-master/graph/F_tdoa2weight.m
392
utf_8
ad9435397bfdb60fa9ba377f12783a7c
function output = F_tdoa2weight(input, freq_bin) % assume input is an array of time delay of C microphone channels. % freq_bin is an array of center frequencies of N FFT bins. [D,T,N] = size(input); nCh = D+1; delay = [zeros(1,T); input]; delay = reshape(delay, 1, nCh, T, N); j = sqrt(-1); output = exp(-j * bsxfun...
github
singaxiong/SignalGraph-master
F_minmax_norm.m
.m
SignalGraph-master/graph/F_minmax_norm.m
920
utf_8
b4f4fff1a3727c65dced6487d50d4fdd
function [output,validFrameMask] = F_minmax_norm(input_layer, curr_layer) input = input_layer.a; [D,T,N] = size(input); if isfield(curr_layer, 'minmax') minmax = curr_layer.minmax; else minmax = [-1 1]; end if N==1 output = MinMaxNorm(input,minmax(1), minmax(2)); validFrameMask = []; else [validF...
github
singaxiong/SignalGraph-master
F_repmat.m
.m
SignalGraph-master/graph/F_repmat.m
291
utf_8
e9b057f17dc81908d88ce98da59a3de8
% repeat a matrix % function [output] = F_repmat(input_layer, curr_layer) input = input_layer.a; sourceDims = curr_layer.sourceDims; targetDims = curr_layer.targetDims; if length(sourceDims)==1 output = repmat(input, targetDims(1),targetDims(2)); else % to be implemented end end
github
singaxiong/SignalGraph-master
AddSpMatMat.m
.m
SignalGraph-master/graph/AddSpMatMat.m
700
utf_8
02957dfcf636093d6d1e5999df3245f9
% This function add a full matrix with a sparse matrix function out = AddSpMatMat(w1,spMat, w2, Mat, sp_elements_only) [m,n] = size(spMat); if w1==-1 spMat = -spMat; elseif w1~=1 spMat = spMat * w1; end if w2==-1 Mat = -Mat; elseif w2~=1 Mat = Mat * w2; end out = Mat; if 0 idx = find(spMat); ...
github
singaxiong/SignalGraph-master
PostprocessCostEvaluation.m
.m
SignalGraph-master/graph/PostprocessCostEvaluation.m
1,707
utf_8
53288559cac79e59c83df9d22a5694a7
% This function postprocess gradient to undo whatever we did in % prepareCostEvaluation, such as labelDelay, costFrameSelection, etc. % Author: Xiong Xiao, Temasek Labs, NTU, Singapore. % Last modified: 28 Jun 2016 % function grad = PostprocessCostEvaluation(grad, output, mask, nSeg, nFrOrig, CostLayer) if ~isempty(m...
github
singaxiong/SignalGraph-master
TestDereverbNet_Regression_Rpi.m
.m
SignalGraph-master/examples/dereverb/TestDereverbNet_Regression_Rpi.m
2,691
utf_8
97f6df47f02e5757d81d2dcd639a65c3
function TestDereverbNet_Regression_Rpi addpath('local'); dnn1 = load('nnet/Dereverb.noCMN.DeltaByEqn.MbSize20.U52013.771-LSTM-2048-771.L2_3E-4.LR_5E-3/nnet.itr10.LR4.39E-4.CV1630.073.mat'); dnn2 = load('nnet/DereverbMask.noCMN.DeltaByEqn.MbSize20.U52013.771-LSTM-1500-771.L2_3E-4.LR_5E-3/nnet.itr11.LR1.46E-4.CV1801.50...
github
singaxiong/SignalGraph-master
TrainDereverbNet_Regression.m
.m
SignalGraph-master/examples/dereverb/TrainDereverbNet_Regression.m
4,298
utf_8
618a47fffd689ced65d5e8758264ba14
% This script train an LSTM or DNN based clean speech log spectrogram % predictor, using simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University, Singapore % Last modified: 08 Feb 2016. % %clear function TrainDereverbNet_Regression(modelType, hiddenLayerSize, hiddenLayerSizeFF, DeltaGen...
github
singaxiong/SignalGraph-master
TrainDereverbNet_RegressionComplex.m
.m
SignalGraph-master/examples/dereverb/TrainDereverbNet_RegressionComplex.m
4,419
utf_8
6abb5763a71112ca11c0f705432c307d
% This script train an LSTM or DNN to predict complex Fourier transform % domain temporal filters, which will be used to filter the noisy and % reverberant Fourier coefficients to produce enhanced speech. The script % is based on the simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University...
github
singaxiong/SignalGraph-master
TrainDereverbNet_FilterPredictionSubnet.m
.m
SignalGraph-master/examples/dereverb/TrainDereverbNet_FilterPredictionSubnet.m
4,609
utf_8
ae4cfe99c137672bfaef477c6e49420f
% This script train an LSTM or DNN to predict complex Fourier transform % domain temporal filters, which will be used to filter the noisy and % reverberant Fourier coefficients to produce enhanced speech. The script % is based on the simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University...
github
singaxiong/SignalGraph-master
ResaveWSJCAM0WAV.m
.m
SignalGraph-master/examples/dereverb/ResaveWSJCAM0WAV.m
922
utf_8
7e909e683ded47cded11de95eadac4ad
% WSJCAM0 audio files are in SPHERE format, while the rest of REVERB % Challenge data are in wav format. To make it easier to read audio files, % we first save the WSJCAM0 audio into wav files. % Xiong Xiao, Nanyang Technological University, Singapore. % Feb 9, 2017 % function ResaveWSJCAM0WAV() wsjcam0root = Choose...
github
singaxiong/SignalGraph-master
TrainDereverbNet_Masking.m
.m
SignalGraph-master/examples/dereverb/TrainDereverbNet_Masking.m
4,265
utf_8
95d1e3a19866387d3d0309e4a91ca440
% This script train an LSTM or DNN based clean speech log spectrogram % predictor, using simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University, Singapore % Last modified: 08 Feb 2016. % %clear function TrainDereverbNet_Masking(modelType, hiddenLayerSize, hiddenLayerSizeFF, DeltaGenera...
github
singaxiong/SignalGraph-master
TrainDereverbNet_FilterPrediction.m
.m
SignalGraph-master/examples/dereverb/TrainDereverbNet_FilterPrediction.m
4,889
utf_8
cf3cd0113414e2a84d31a6f1d9746fda
% This script train an LSTM or DNN to predict complex Fourier transform % domain temporal filters, which will be used to filter the noisy and % reverberant Fourier coefficients to produce enhanced speech. The script % is based on the simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University...
github
singaxiong/SignalGraph-master
TestDereverbNet_Regression.m
.m
SignalGraph-master/examples/dereverb/TestDereverbNet_Regression.m
4,843
utf_8
5201a4d886856e1c41edabf32b9a9c33
function TestDereverbNet_Regression addpath('local'); % dnn = load('nnet/Dereverb.U7861.771-LSTM-1024-257.L2_3E-4.LR_1E-2/nnet.itr6.LR9.58E-4.CV1679.563.mat'); % dnn = load('nnet/Dereverb.noCMN.U7861.771-LSTM-1024-771.L2_3E-4.LR_1E-2/nnet.itr10.LR2E-4.CV1760.704.mat'); % dnn = load('nnet/Dereverb.noCMN.DeltaByEqn.MbSi...
github
singaxiong/SignalGraph-master
genNetworkDereverb_Gaussian.m
.m
SignalGraph-master/examples/dereverb/local/genNetworkDereverb_Gaussian.m
7,033
utf_8
6162c43e85f7ff06670748459dee9c1f
% This file create a simple regression based network for speech % dereverberation or enhancement % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Feb 2017 % function layer = genNetworkDereverb_Gaussian(para, stage) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, where f is ...
github
singaxiong/SignalGraph-master
genNetworkDereverb_FilterPrediction.m
.m
SignalGraph-master/examples/dereverb/local/genNetworkDereverb_FilterPrediction.m
8,071
utf_8
33e78a00471b01ebb95b239bffb0c2f5
% This file create a simple regression based network for speech % dereverberation or enhancement % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Feb 2017 % function layer = genNetworkDereverb_FilterPrediction(para, type) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, wher...
github
singaxiong/SignalGraph-master
LoadWavFilter_Reverb.m
.m
SignalGraph-master/examples/dereverb/local/LoadWavFilter_Reverb.m
3,375
utf_8
6707e178580ff942674d6c4c97afc722
% Load far talk, close talk, and frame label. function [Data, para, vocab] = LoadWavFilter_Reverb(para, step, dataset, datatype, distance) nCh = para.topology.useChannel; wavlist = []; % note that training data do not contain real recordings. % real dev and eval data do not have clean version for type_i = 1:length(da...
github
singaxiong/SignalGraph-master
genNetworkDereverb_Regression.m
.m
SignalGraph-master/examples/dereverb/local/genNetworkDereverb_Regression.m
6,187
utf_8
52a210fa17183093797c9e97fc94b852
% This file create a simple regression based network for speech % dereverberation or enhancement % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Feb 2017 % function layer = genNetworkDereverb_Regression(para) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, where f is the n...
github
singaxiong/SignalGraph-master
ConfigDereverbNet_Regression.m
.m
SignalGraph-master/examples/dereverb/local/ConfigDereverbNet_Regression.m
1,808
utf_8
c4ff2edcba275c199f3025d3bc3f2964
% This file serves as a template of defining the topology of the % beamforming network with cross entropy training. % You should create a copy for each of your experiments and name them % differently. % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 29 Nov 2016 % function para = Confi...
github
singaxiong/SignalGraph-master
genNetworkDereverb_Masking.m
.m
SignalGraph-master/examples/dereverb/local/genNetworkDereverb_Masking.m
6,931
utf_8
8343a990bbd1ff8b11f7dd8858bd955c
% This file create a simple regression based network for speech % dereverberation or enhancement % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Feb 2017 % function layer = genNetworkDereverb_Masking(para) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, where f is the norm...
github
singaxiong/SignalGraph-master
Build_DereverbNet_Regression.m
.m
SignalGraph-master/examples/dereverb/local/Build_DereverbNet_Regression.m
2,470
utf_8
ac9c3179e28682ec4b0a5716b49db535
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_DereverbNet_Regression(Data_tr, para) para.output = 'tmp'; if para.topology.useMasking layer = genNetworkDereverb_Masking(para.topology); % generate the network graph else ...
github
singaxiong/SignalGraph-master
Build_DereverbNet_Masking.m
.m
SignalGraph-master/examples/dereverb/local/Build_DereverbNet_Masking.m
2,367
utf_8
6605ae53b277026bc200c3fb399796c4
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_DereverbNet_Masking(Data_tr, para) para.output = 'tmp'; layer = genNetworkDereverb_Masking(para.topology); % generate the network graph para.preprocessing{1} = {}; ...
github
singaxiong/SignalGraph-master
Build_DereverbNet_FilterPrediction.m
.m
SignalGraph-master/examples/dereverb/local/Build_DereverbNet_FilterPrediction.m
2,555
utf_8
355e3ec85a980aa040786fc2516b7aa8
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_DereverbNet_FilterPrediction(Data_tr, para, type) para.output = 'tmp'; if nargin<3 type = 'fullnet'; end layer = genNetworkDereverb_FilterPrediction(para.topology, type); % ...
github
singaxiong/SignalGraph-master
LoadParallelWavLabel_Reverb.m
.m
SignalGraph-master/examples/dereverb/local/LoadParallelWavLabel_Reverb.m
5,528
utf_8
c4b0b8c379102fb1a99f4c29881a49b2
% Load far talk, close talk, and frame label. function [Data, para, vocab] = LoadParallelWavLabel_Reverb(para, step, dataset, datatype, distance) nCh = para.topology.useChannel; wavlist = []; wavlistClean = []; % note that training data do not contain real recordings. % real dev and eval data do not have clean versio...
github
singaxiong/SignalGraph-master
Build_DereverbNet_RegressionComplex.m
.m
SignalGraph-master/examples/dereverb/local/Build_DereverbNet_RegressionComplex.m
763
utf_8
0cecf692fcc5e93ec5d3eadfbc683dc1
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_DereverbNet_RegressionComplex(Data_tr, para) para.output = 'tmp'; layer = genNetworkDereverb_RegressionComplex(para.topology); para.preprocessing{1} = {}; % opt...
github
singaxiong/SignalGraph-master
genNetworkDereverb_RegressionComplex.m
.m
SignalGraph-master/examples/dereverb/local/genNetworkDereverb_RegressionComplex.m
3,804
utf_8
9fcca4820dd2ff9c4c96afb0d40967e8
% This file create a simple regression based network for speech % dereverberation or enhancement % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Feb 2017 % function layer = genNetworkDereverb_RegressionComplex(para) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, where f i...
github
singaxiong/SignalGraph-master
GenFbankFeatures.m
.m
SignalGraph-master/examples/beamforming/GenFbankFeatures.m
1,650
utf_8
7f2b953f9209cf9d2e6ed2213cf2e8ee
% This function generate the log Mel filterbanks features of the training % and evaluation data. Note that the filterbanks features generated here is % slightly different from that by other toolkits, such as Kaldi. % This recipe will always use the filterbank features generated in the same % way as in this function, s...
github
singaxiong/SignalGraph-master
ConfigBasicSTFT.m
.m
SignalGraph-master/examples/beamforming/lib/ConfigBasicSTFT.m
577
utf_8
1c4a89411432bd03d73c01c00f231d19
% Add default configurations to STFT % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 29 Jun 2016 % function para = ConfigBasicSTFT(para) para.topology.fs = 16000; % sampling rate para.topology.fft_len = 512; para.topology.nFbank = 40; % define the parameters for extracting Fourier c...
github
singaxiong/SignalGraph-master
LoadParallelWavLabel_CHiME4.m
.m
SignalGraph-master/examples/beamforming/lib/LoadParallelWavLabel_CHiME4.m
6,682
utf_8
c3e6af65c1e30496a291f68b3846cf31
% Load far talk, close talk, and frame label. function [Data, para, vocab, wavlist] = LoadParallelWavLabel_CHiME4(para, step, dataset) switch lower(dataset) case {'dt05','et05'} wavlist = ['../Kaldi/data/' dataset '_multi_noisy/wav.scp']; ali_file = [para.local.aliDir '_' dataset '/ali.txt']; c...
github
singaxiong/SignalGraph-master
HandleSTFT.m
.m
SignalGraph-master/examples/beamforming/lib/HandleSTFT.m
469
utf_8
f8c8f0c1d7e2c8f706b685bac652f97c
function [layer, scm_idx, split] = HandleSTFT(layer, STFT_layer_idx, nPass, para) [scm_idx, split] = GetScmLayer(layer); bf_idx = ReturnLayerIdxByName(layer, 'Beamforming'); for i=1:nPass if split layer{scm_idx(i)}.prev(3) = STFT_layer_idx - scm_idx(i); else layer{scm_idx(i)}.prev(2) = STFT_lay...
github
singaxiong/SignalGraph-master
BuildFbankExtractionNet.m
.m
SignalGraph-master/examples/beamforming/lib/BuildFbankExtractionNet.m
955
utf_8
222b18afdcbbf00aafd2a0bf836f735b
% Build and initialize the computational graph for extracting log Mel % filterbank features % function [layer, para] = BuildFbankExtractionNet() para.output = 'tmp'; para.IO.nStream = 1; para.NET.sequential = 1; para.cost_func.layer_idx = []; para = ConfigBasicSTFT(para); layer = genNetworkFbankExtraction(para.topolog...
github
singaxiong/SignalGraph-master
ConvertMaskBF2Split.m
.m
SignalGraph-master/examples/beamforming/mask_prediction/local/ConvertMaskBF2Split.m
2,090
utf_8
02cad0c62d9fa3bb364cfffa889fc1eb
% The network generated by genNetworkMaskBF_CE.m only predicts speech mask. % The function will convert the network to predict speech and noise masks % independently. % Currently, this function assume there is only 1 LSTM layer for mask % prediction. Extension needed if we use multiple LSTM layers. % function [layer,...
github
singaxiong/SignalGraph-master
HandleSTFTReference.m
.m
SignalGraph-master/examples/beamforming/mask_prediction/local/HandleSTFTReference.m
478
utf_8
63b04e4c0301db90cbc9356c04a8ae40
function [layer, scm_idx, split] = HandleSTFTReference(layer, STFT_layer_idx, nPass, para) [scm_idx, split] = GetScmLayer(layer); bf_idx = ReturnLayerIdxByName(layer, 'Beamforming'); for i=1:nPass if split layer{scm_idx(i)}.prev(3) = STFT_layer_idx - scm_idx(i); else layer{scm_idx(i)}.prev(2) =...
github
singaxiong/SignalGraph-master
genNetworkMaskBF_CE.m
.m
SignalGraph-master/examples/beamforming/mask_prediction/local/genNetworkMaskBF_CE.m
7,648
utf_8
ad72fb40d4f88e9e7aa645f4a8003af4
% This file create the joint mask predicting, beamforming weight % predicting, and acoustic model networks consists of a linked list % of computing layers. % % References: % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Jul 2016 % function layer = genNetworkMaskBF_CE(para) para.f...
github
singaxiong/SignalGraph-master
ConvertMaskBF2pooling.m
.m
SignalGraph-master/examples/beamforming/mask_prediction/local/ConvertMaskBF2pooling.m
2,547
utf_8
78c9253e889584b8879e29acce34493f
% The network generated by genNetworkMaskBF_CE.m always use the first % channel to predict the mask. This function changes the network such that % it uses all the 6 channels for mask prediction. For each channel, a % speech and noise masks are predicted based on only the log spectrum of % current channel. Then, the 6 s...
github
singaxiong/SignalGraph-master
ConfigMaskBFnetCE.m
.m
SignalGraph-master/examples/beamforming/mask_prediction/local/ConfigMaskBFnetCE.m
5,260
utf_8
a4d5a487e9b702aa930eb1f8baaa03cd
% This file serves as a template of defining the topology of the % beamforming network with cross entropy training. % You should create a copy for each of your experiments and name them % differently. % % References: % [1] Xiong Xiao, Shinji Watanabe, Hakan Erdogan, Liang Lu, John Hershey, % Michael L. Seltzer, Guogu...
github
singaxiong/SignalGraph-master
Build_MaskBFnet_CE.m
.m
SignalGraph-master/examples/beamforming/mask_prediction/local/Build_MaskBFnet_CE.m
7,612
utf_8
1f20ce209dabeea2d69018c0789dfaec
% Build and initialize the computational graph for mask-based beamforming % for speech recognition. % stage: % 1 - only initialize mask prediction subnet % 2 - only initiliaze mask prediction and weight prediction subnets % 3 - initialize also the acoustic model subnet % function [layer, para] = Build_MaskBFnet...
github
singaxiong/SignalGraph-master
genNetworkMVDR.m
.m
SignalGraph-master/examples/beamforming/mvdr/local/genNetworkMVDR.m
1,656
utf_8
2e2ff27f1fe3c77d4c96e6eb0f8f6a20
% This function create a network of MVDR beamforming. You need to supply % the noise and speech masks. % % Created by Xiong Xiao % Last Modified: 30 Jun 2017 % function layer = genNetworkMVDR(para) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, where f is the normalized frequency k/N is from 0 to 0.5. nFre...
github
singaxiong/SignalGraph-master
TrainEnhanceNet_Masking.m
.m
SignalGraph-master/examples/enhancement/TrainEnhanceNet_Masking.m
5,794
utf_8
c163d0753ae0a228dd8a8275443beec7
% This script train an LSTM or DNN based clean speech log spectrogram % predictor, using simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University, Singapore % Last modified: 08 Feb 2016. % %clear function TrainEnhanceNet_Masking(modelType, hiddenLayerSize, hiddenLayerSizeFF, learning_rat...
github
singaxiong/SignalGraph-master
TrainEnhanceNet_Gaussian.m
.m
SignalGraph-master/examples/enhancement/TrainEnhanceNet_Gaussian.m
6,226
utf_8
01a08d76b6e93b43912fbd3d684a59ec
% This script train an LSTM or DNN based clean speech log spectrogram % predictor, using simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University, Singapore % Last modified: 08 Feb 2016. % %clear function TrainEnhanceNet_Gaussian(modelType, hiddenLayerSizeShared, hiddenLayerSizeMu, hidde...
github
singaxiong/SignalGraph-master
BatchTestEnhanceNetByCategory.m
.m
SignalGraph-master/examples/enhancement/BatchTestEnhanceNetByCategory.m
4,792
utf_8
ad92d72c1890a6430ca7d89772a0135f
function BatchTestEnhanceNetByCategory(testSet, T60, noise, SNR, DEBUG) addpath('local'); dnn_files{1} = 'nnet/EnhanceRegression.noCMN.DeltaByEqn.MbSize40.U28539.771-LSTM-2048-771.L2_3E-4.LR_3E-3/nnet.itr37.LR1E-5.CV2453.828.mat'; dnn_files{2} = 'nnet/EnhanceRegression.noCMN.DeltaByEqn.MbSize20.U28539.771-LSTM-2048-7...
github
singaxiong/SignalGraph-master
TrainEnhanceNet_Regression.m
.m
SignalGraph-master/examples/enhancement/TrainEnhanceNet_Regression.m
5,657
utf_8
7afce091ac7adaa42225dbf74def1c8e
% This script train an LSTM or DNN based clean speech log spectrogram % predictor, using simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University, Singapore % Last modified: 08 Feb 2016. % %clear function TrainEnhanceNet_Regression(modelType, hiddenLayerSize, hiddenLayerSizeFF, DeltaGene...
github
singaxiong/SignalGraph-master
Build_EnhanceNet_Masking.m
.m
SignalGraph-master/examples/enhancement/local/Build_EnhanceNet_Masking.m
2,730
utf_8
c6f0e5b590d80435ff8079f6d370bfa4
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_EnhanceNet_Masking(Data_tr, para) para.output = 'tmp'; layer = genNetworkDereverb_Masking(para.topology); % generate the network graph para.preprocessing{1} = {}; ...
github
singaxiong/SignalGraph-master
LoadParallelWav_Libri.m
.m
SignalGraph-master/examples/enhancement/local/LoadParallelWav_Libri.m
2,885
utf_8
3075aff75c0c170f7c6b2b43d041ac3a
% Load far talk, close talk, and frame label. function [Data, para, vocab] = LoadParallelWav_Libri(para, step, precision) if nargin<3 precision = 'int16'; end if isfield(para.local, 'cv_wav_root_clean') wavlistClean = findFiles(para.local.cv_wav_root_clean, para.local.cv_wav_clean_ext); wavlist = findFiles...
github
singaxiong/SignalGraph-master
RunEnhanceNN.m
.m
SignalGraph-master/examples/enhancement/local/RunEnhanceNN.m
880
utf_8
30cbd8fd458cede80dc319d59beb1088
function [noisy, enhanced, clean, enhanced_wav, noisySTFT, mask, variance] = RunEnhanceNN(Data, layer, para) output = FeatureTree2(Data, para, layer); noisySTFT = gather(output{1}{1}); noisy = gather(output{1}{2}); enhanced = gather(output{1}{3}); enhanced_wav = abs2wav(exp(enhanced(1:257,:)/2)', angle(noisySTFT)',...
github
singaxiong/SignalGraph-master
LoadWavRIRNoise_Libri.m
.m
SignalGraph-master/examples/enhancement/local/LoadWavRIRNoise_Libri.m
2,333
utf_8
da3583c83217c35e5565597916be83fc
% Load far talk, close talk, and frame label. function [Data, para] = LoadWavRIRNoise_Libri(para, step) wavreader.name = 'wavfile'; wavreader.array = 0; wavreader.precision = 'int16'; % load clean speech list if isfield(para.local, 'clean_wav_files') clean_list = para.local.clean_wav_files; else clean_list =...
github
singaxiong/SignalGraph-master
Build_EnhanceNet_Regression.m
.m
SignalGraph-master/examples/enhancement/local/Build_EnhanceNet_Regression.m
2,736
utf_8
66626b9a541e5268a2a6dce0de76c5c4
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_EnhanceNet_Regression(Data_tr, para) para.output = 'tmp'; layer = genNetworkDereverb_Regression(para.topology); % generate the network graph para.preprocessing{1} = {}; ...
github
singaxiong/SignalGraph-master
TestEnhanceNetByCategory.m
.m
SignalGraph-master/examples/enhancement/local/TestEnhanceNetByCategory.m
10,674
utf_8
7983196d81cbc282e485e40a9f6fad4d
function TestEnhanceNetByCategory(dnn_files, testSet, T60, noise, SNR, IsMaskNet, measures, useGPU, DEBUG) addpath('local'); % dnn_files{1} = 'nnet/EnhanceRegression.noCMN.DeltaByEqn.MbSize40.U28539.771-LSTM-2048-771.L2_3E-4.LR_3E-3/nnet.itr37.LR1E-5.CV2453.828.mat'; % dnn_files{2} = 'nnet/EnhanceRegression.noCMN.Del...
github
singaxiong/SignalGraph-master
Build_EnhanceNet_Gaussian.m
.m
SignalGraph-master/examples/enhancement/local/Build_EnhanceNet_Gaussian.m
4,207
utf_8
d0cb00274834041101f01d9a6a3fa793
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_EnhanceNet_Gaussian(Data_tr, para, stage) para.output = 'tmp'; layer = genNetworkDereverb_Gaussian(para.topology, stage); % generate the network graph para.preprocessing{1} = {}...
github
singaxiong/SignalGraph-master
PrepareNetwork4Enhancement.m
.m
SignalGraph-master/examples/enhancement/local/PrepareNetwork4Enhancement.m
1,566
utf_8
8441a9a7f49d36d009ddc21274591af6
function [model] = PrepareNetwork4Enhancement(dnnFile, hasClean, useMasking, useGPU) dnn = load(dnnFile); layer = dnn.layer; para = dnn.para; para.local.useFileName = 1; % para.topology.useFileName = 1; % para.local.seglen = 100; % para.local.segshift = 100; para.useGPU = useGPU; % noisy STFT stft_idx = ReturnLayerI...
github
singaxiong/SignalGraph-master
TestEnhanceNetByCategory.m
.m
SignalGraph-master/examples/enhancement/local/back/TestEnhanceNetByCategory.m
9,792
utf_8
c83053d59787830b11baf36e0b3d506b
function TestEnhanceNetByCategory(testSet, T60, noise, SNR, playSound) addpath('..\..\..\..\Enhancement\Loizou\MATLAB_code\objective_measures\quality'); addpath('..\..\..\..\Enhancement\Loizou\MATLAB_code\statistical_based'); hasClean = 1; addpath('local'); % dnn1 = load('nnet/EnhanceRegression.noCMN.DeltaByEqn.MbSiz...
github
singaxiong/SignalGraph-master
TestEnhanceNet.m
.m
SignalGraph-master/examples/enhancement/local/back/TestEnhanceNet.m
7,997
utf_8
3207ec507ef79b288f6ed919dbb85ac0
function TestEnhanceNet addpath('..\..\..\..\Enhancement\Loizou\MATLAB_code\objective_measures\quality'); addpath('..\..\..\..\Enhancement\Loizou\MATLAB_code\statistical_based'); hasClean = 1; addpath('local'); dnn1 = load('nnet/EnhanceRegression.noCMN.DeltaByEqn.MbSize20.U28539.771-LSTM-2048-771.L2_3E-4.LR_1E-4/nnet...
github
singaxiong/SignalGraph-master
TrainSeparationNet_MaskingMagnitude.m
.m
SignalGraph-master/examples/separation/TrainSeparationNet_MaskingMagnitude.m
6,590
utf_8
7242dc7868c9de8109a4be8ce9e95f02
% This script train an LSTM or DNN based clean speech log spectrogram % predictor, using simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University, Singapore % Last modified: 08 Feb 2016. % %clear function TrainSeparationNet_MaskingMagnitude(hiddenLayerSize, hiddenLayerSizeFF, learning_ra...
github
singaxiong/SignalGraph-master
TrainSeparationNet_Masking.m
.m
SignalGraph-master/examples/separation/TrainSeparationNet_Masking.m
6,565
utf_8
d09e598a1849fd920c7491167bd1cd16
% This script train an LSTM or DNN based clean speech log spectrogram % predictor, using simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University, Singapore % Last modified: 08 Feb 2016. % %clear function TrainSeparationNet_Masking(hiddenLayerSize, hiddenLayerSizeFF, learning_rate, nUtt4...
github
singaxiong/SignalGraph-master
TrainSeparationNet_Regression.m
.m
SignalGraph-master/examples/separation/TrainSeparationNet_Regression.m
6,584
utf_8
a89a4b6cdc1d98f0d5bc2b66372c209e
% This script train an LSTM or DNN based clean speech log spectrogram % predictor, using simulated data of Reverb challenge. % % Xiong Xiao, Nanyang Technological University, Singapore % Last modified: 08 Feb 2016. % %clear function TrainSeparationNet_Regression(hiddenLayerSize, hiddenLayerSizeFF, learning_rate, nU...
github
singaxiong/SignalGraph-master
Build_SeparationNet_Masking.m
.m
SignalGraph-master/examples/separation/local/Build_SeparationNet_Masking.m
3,006
utf_8
fdffe613c078a0d50f7ef1a11a34bbe9
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_SeparationNet_Masking(Data_tr, para) para.output = 'tmp'; layer = genNetworkSeparation_Masking(para.topology); % generate the network graph para.preprocessing{1} = {}; ...
github
singaxiong/SignalGraph-master
LoadSeparationWav_Libri.m
.m
SignalGraph-master/examples/separation/local/LoadSeparationWav_Libri.m
3,561
utf_8
7d1fc3f8d7f4cbc060aecb5c9733969f
% Load far talk, close talk, and frame label. function [Data, para, vocab] = LoadSeparationWav_Libri(para, step) if isfield(para.local, 'cv_wav_root_clean') wavlistClean = findFiles(para.local.cv_wav_root_clean, para.local.cv_wav_clean_ext); wavlist = findFiles([para.local.cv_wav_root], 'wav'); wavIndexCle...
github
singaxiong/SignalGraph-master
genNetworkSeparation_Regression.m
.m
SignalGraph-master/examples/separation/local/genNetworkSeparation_Regression.m
6,777
utf_8
1fec0ab573ca3f24046b13b451039214
% This file create a simple regression based network for speech % dereverberation or enhancement % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Feb 2017 % function layer = genNetworkSeparation_Regression(para) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, where f is the...
github
singaxiong/SignalGraph-master
Build_SeparationNet_MaskingMagnitude.m
.m
SignalGraph-master/examples/separation/local/Build_SeparationNet_MaskingMagnitude.m
2,829
utf_8
74693f02a16d8f7207eabbcbb7f901b4
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_SeparationNet_Masking(Data_tr, para) para.output = 'tmp'; layer = genNetworkSeparation_MaskingMagnitude(para.topology); % generate the network graph para.preprocessing{1} = {}; ...
github
singaxiong/SignalGraph-master
genNetworkSeparation_Masking_ShareProjection.m
.m
SignalGraph-master/examples/separation/local/genNetworkSeparation_Masking_ShareProjection.m
7,851
utf_8
16e98cab0f3fcae86492e307aeca75ba
% This file create a simple regression based network for speech % dereverberation or enhancement % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Feb 2017 % function layer = genNetworkSeparation_Masking_ShareProjection(para) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, w...
github
singaxiong/SignalGraph-master
RunSeparationNN.m
.m
SignalGraph-master/examples/separation/local/RunSeparationNN.m
820
utf_8
dccc91f33692dcdbf3be5a2d6e56c639
function [mixture, separated, clean, separated_wav, mixtureSTFT, mask] = RunSeparationNN(Data, layer, para) output = FeatureTree2(Data, para, layer); for i=1:length(output{1}) output{1}{i} = gather(output{1}{i}); end mixtureSTFT = gather(output{1}{1}); mixture = gather(output{1}{2}); separated{1} = gather(output...
github
singaxiong/SignalGraph-master
genNetworkSeparation_Masking.m
.m
SignalGraph-master/examples/separation/local/genNetworkSeparation_Masking.m
7,512
utf_8
4937c8cde937bd9d9ed5aa4f78cc8c2a
% This file create a simple regression based network for speech % dereverberation or enhancement % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Feb 2017 % function layer = genNetworkSeparation_Masking(para) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, where f is the no...
github
singaxiong/SignalGraph-master
PrepareNetwork4Separation.m
.m
SignalGraph-master/examples/separation/local/PrepareNetwork4Separation.m
1,779
utf_8
2792d42f2bf7a47581051fd7dfb83840
function [model] = PrepareNetwork4Separation(dnnFile, hasClean, useGPU) dnn = load(dnnFile); layer = dnn.layer; para = dnn.para; para.local.useFileName = 1; % para.topology.useFileName = 1; % para.local.seglen = 100; % para.local.segshift = 100; para.useGPU = useGPU; % mixture STFT stft_idx = ReturnLayerIdxByName(la...
github
singaxiong/SignalGraph-master
Build_SeparationNet_Regression.m
.m
SignalGraph-master/examples/separation/local/Build_SeparationNet_Regression.m
3,012
utf_8
82b8b14e35991df59db21d60187202be
% Build and initialize the computational graph for regression based speech % enhancement/dereverberation % function [layer, para] = Build_SeparationNet_Regression(Data_tr, para) para.output = 'tmp'; layer = genNetworkSeparation_Regression(para.topology); % generate the network graph para.preprocessing{1} = {}; ...
github
singaxiong/SignalGraph-master
genNetworkSeparation_MaskingMagnitude.m
.m
SignalGraph-master/examples/separation/local/genNetworkSeparation_MaskingMagnitude.m
5,862
utf_8
3f49ca46b38d984f118fcead4834bd1e
% This file create a simple regression based network for speech % dereverberation or enhancement % % Created by Xiong Xiao, Temasek Laboratories, NTU, Singapore. % Last Modified: 08 Feb 2017 % function layer = genNetworkSeparation_MaskingMagnitude(para) para.freqBin = (0:1/para.fft_len:0.5)*2*pi; % w = 2*pi*f, where f ...
github
singaxiong/SignalGraph-master
TestSeparationNetByCategory.m
.m
SignalGraph-master/examples/separation/local/TestSeparationNetByCategory.m
3,694
utf_8
8c4710a1c7d22860500675215ebe6ce6
function TestSeparationNetByCategory() addpath('local'); % dnn_files{1} = 'nnet/SeparationRegression.CMN.DeltaByEqn.MbSize40.U28539.771-LSTM-2048-771.L2_0.LR_1E-3/nnet.itr19.LR1.29E-4.CV5039.393.mat'; dnn_files{1} = 'nnet/SepReg.CMN.DeltaByEqn.MbSize40.SPR0.seg100.U28539.771-LSTM-2048-771.L2_0.LR_1E-3/nnet.itr2.LR8.3...
github
singaxiong/SignalGraph-master
TIMIT_map39to2.m
.m
SignalGraph-master/examples/classification_framewise/TIMIT_map39to2.m
454
utf_8
66121b7066d69d9c75c4041e85d7cc52
% The mapping from 61 phones set to 48 phonse set for TIMIT. The mapping % table is as follows: % % cl vcl epi --> sil % el --> l % en --> n % zh --> sh % aa --> ao % ix --> ih % ax --> ah % % Author: Xiong Xiao, NTU, Singapore % Date: 1 Jun 2016 function phone_seq_2 = TIMIT_map39to2(phone_seq_39) found = strcmpi(pho...
github
singaxiong/SignalGraph-master
TIMIT_map61to48.m
.m
SignalGraph-master/examples/classification_framewise/TIMIT_map61to48.m
959
utf_8
ea9fd20929cec8222d926670f02dc794
% The mapping from 61 phones set to 48 phonse set for TIMIT. The mapping % table is as follows: % % pcl tcl kcl qcl q --> cl % bcl dcl gcl --> vcl % h# #h pau --> sil % ux --> uw % axr --> er % em --> m % nx --> n % eng --> ng % hv --> hh % ax-h --> ax % % Author: Xiong Xiao, NTU, Singapore % Date: 1 Jun 2016 function...
github
singaxiong/SignalGraph-master
TIMIT_map48to39.m
.m
SignalGraph-master/examples/classification_framewise/TIMIT_map48to39.m
747
utf_8
0331399c6fe423ca626830ce372c4b4b
% The mapping from 61 phones set to 48 phonse set for TIMIT. The mapping % table is as follows: % % cl vcl epi --> sil % el --> l % en --> n % zh --> sh % aa --> ao % ix --> ih % ax --> ah % % Author: Xiong Xiao, NTU, Singapore % Date: 1 Jun 2016 function phone_seq_39 = TIMIT_map48to39(phone_seq_48) mapping_table = {...
github
singaxiong/SignalGraph-master
genNetworkTemporalConvDeep.m
.m
SignalGraph-master/prototypes/genNetworkTemporalConvDeep.m
1,980
utf_8
353155580424437fb3bdcc20778d9f18
% poolType = 'mean' or 'max' % poolLayer = integer layer number after which we do the pooling function layer = genNetworkTemporalConvDeep(para) if isfield(para, 'LastActivation4MSE')==0 para.LastActivation4MSE = 'linear'; end inputDim = double(para.inputDim); layer{1}.name = 'Input'; % this is an input laye...
github
singaxiong/SignalGraph-master
computeGlobalCMVN_obj.m
.m
SignalGraph-master/prototypes/computeGlobalCMVN_obj.m
1,287
utf_8
e5d7987f4aaadbbe9d60a7cec8d154c3
% given a network and some data, compute the mean and variance of the % network output. This function is usually used to determine the global % mean and variance normalization parameters in the preprocessing. % Xiong Xiao % function [W, b] = computeGlobalCMVN_obj(Visible, nUttUsed, para, layer) if exist('nUttUsed')==0...
github
singaxiong/SignalGraph-master
FinishLayer.m
.m
SignalGraph-master/prototypes/FinishLayer.m
360
utf_8
8d7ac11ade26c4ef64cd7dc82ba519de
% automatically derive the list of layers that the output of the current layer goes. function layer = FinishLayer(layer) for i=1:length(layer); layer{i}.next = []; end for i=length(layer):-1:1 if isfield(layer{i}, 'prev') for j=1:length(layer{i}.prev) layer{i+layer{i}.prev(j)}.next(end+1) = -la...
github
singaxiong/SignalGraph-master
genNetworkLinearStructure.m
.m
SignalGraph-master/prototypes/genNetworkLinearStructure.m
1,399
utf_8
100253a2fd771ce28973d22bd8456766
% generate a network with linear structure using descriptive config function layer = genNetworkLinearStructure(config) for i=1:length(config) currConfig = config{i}; end inputDim = double(inputDim); inputStreamIdx = 1; layer{1} = InputNode(inputStreamIdx,inputDim); for i=1:length(hiddenLayerSize) ...
github
singaxiong/SignalGraph-master
genNetworkTemporalConv.m
.m
SignalGraph-master/prototypes/genNetworkTemporalConv.m
1,353
utf_8
f0ecb857ef8c92d9460cd274912edf21
% poolType = 'mean' or 'max' % poolLayer = integer layer number after which we do the pooling function layer = genNetworkTemporalConv(inputDim, nFilter, filterLen, hiddenLayerSize, outputDim, costFn, LastActivation4MSE) if nargin<7 LastActivation4MSE = 'linear'; end inputDim = double(inputDim); layer{1}.name = 'In...
github
singaxiong/SignalGraph-master
FinishLayer_obj.m
.m
SignalGraph-master/prototypes/FinishLayer_obj.m
370
utf_8
2b0a1ea5b121c7671f01d2abd773f41f
% automatically derive the list of layers that the output of the current layer goes. function layer = FinishLayer_obj(layer) for i=1:length(layer); layer{i}.next = []; end for i=length(layer):-1:1 if strcmpi(layer{i}.name, 'input'); continue; end for j=1:length(layer{i}.prev) layer{i+layer{i}.prev(j)}....
github
singaxiong/SignalGraph-master
computeGlobalPCA.m
.m
SignalGraph-master/prototypes/computeGlobalPCA.m
1,428
utf_8
f16c87ae38b517424173cb5b278a4cdd
% given a network and some data, compute the mean and variance of the % network output. This function is usually used to determine the global % mean and variance normalization parameters in the preprocessing. % Xiong Xiao % function [W, b] = computeGlobalPCA(Visible, nUttUsed, dimUsed, para, layer) if exist('nUttUsed'...
github
singaxiong/SignalGraph-master
ConnectLinearGraph.m
.m
SignalGraph-master/prototypes/ConnectLinearGraph.m
624
utf_8
792e50d1d9a64f9746fb1d8ad535cd61
% giving an array of nodes, assume they are connected from the first node % to the last node. Automatically derive the connections parameters and % dimensions function layer = ConnectLinearGraph(layer) % generate the prev and next properties for i=1:length(layer) if strcmpi(class(layer{i}), 'InputNode') %...
github
singaxiong/SignalGraph-master
genNetworkFF_Pairwise2.m
.m
SignalGraph-master/prototypes/genNetworkFF_Pairwise2.m
2,034
utf_8
532bc4217b66ffe76dc20d8cd6d478b7
% generate network prototype that use feedforward (FF) networks for % predicting feature representation and frame weights. Then the feature % representation are weighted by the frame weights and summed together to % produce a single feature vector for each input sequence. function [layer, WeightTyingSet] = genNetworkF...
github
singaxiong/SignalGraph-master
genNetworkMTL_DML_CE.m
.m
SignalGraph-master/prototypes/genNetworkMTL_DML_CE.m
993
utf_8
1a9f528f2b7019ae3048b30a22ecd22f
% generate network prototype that use feedforward (FF) networks for % predicting feature representation and frame weights. Then the feature % representation are weighted by the frame weights and summed together to % produce a single feature vector for each input sequence. function [layer, WeightTyingSet] = genNetworkM...
github
singaxiong/SignalGraph-master
genNetworkTemporalConv2.m
.m
SignalGraph-master/prototypes/genNetworkTemporalConv2.m
1,654
utf_8
74020d75660f1b2fc7b37a544c037baa
% poolType = 'mean' or 'max' % poolLayer = integer layer number after which we do the pooling function layer = genNetworkTemporalConv2(para) if isfield(para, 'LastActivation4MSE')==0 para.LastActivation4MSE = 'linear'; end inputDim = double(para.inputDim); layer{1}.name = 'Input'; % this is an input layer l...
github
singaxiong/SignalGraph-master
genNetworkFF_Pairwise.m
.m
SignalGraph-master/prototypes/genNetworkFF_Pairwise.m
2,010
utf_8
c2d3a7ac5bb8debd86c91a513da8274c
% generate network prototype that use feedforward (FF) networks for % predicting feature representation and frame weights. Then the feature % representation are weighted by the frame weights and summed together to % produce a single feature vector for each input sequence. function [layer, WeightTyingSet] = genNetworkF...
github
singaxiong/SignalGraph-master
genNetworkSTFT2LogSpec.m
.m
SignalGraph-master/prototypes/genNetworkSTFT2LogSpec.m
422
utf_8
57ed6a733407a6ede9e1ea07662a2ca1
% create a sub network that takes in waveforms and produces fourier % coefficients function layer = genNetworkSTFT2LogSpec(stftLayer, useLog, logConst) layer = stftLayer; layer{end+1} = PowerNode(stftLayer{end}.dim(1)); if useLog if nargin<3 logConst = 0.00; end layer{end+1} = LogarithmNode(layer...
github
singaxiong/SignalGraph-master
VerifyPreprocessingTree_obj.m
.m
SignalGraph-master/prototypes/VerifyPreprocessingTree_obj.m
325
utf_8
98cb8544c1ce7ba3e24e27f3207135fa
% verify that the global MVN is correct. % Xiong Xiao function [processing] = VerifyPreprocessingTree_obj(layer, Visible, para, nUttUsed) if exist('nUttUsed')==0 || isempty(nUttUsed) nUttUsed = 500; end [W, b] = computeGlobalCMVN_obj(Visible, nUttUsed, para, layer); plot(-b); hold on; plot(1./diag(W)); hold off ...
github
singaxiong/SignalGraph-master
genNetworkFF_WeightedAverage.m
.m
SignalGraph-master/prototypes/genNetworkFF_WeightedAverage.m
1,729
utf_8
89510b30ee47410c0d65f7c4afe5eccb
% generate network prototype that use feedforward (FF) networks for % predicting feature representation and frame weights. Then the feature % representation are weighted by the frame weights and summed together to % produce a single feature vector for each input sequence. function layer = genNetworkFF_WeightedAverage(...
github
singaxiong/SignalGraph-master
genNetworkTemporalConvLSTM.m
.m
SignalGraph-master/prototypes/genNetworkTemporalConvLSTM.m
396
utf_8
da9f67355a0fd9efe7a4055d4896240d
% poolType = 'mean' or 'max' % poolLayer = integer layer number after which we do the pooling function layer = genNetworkTemporalConvLSTM(para) layer = genNetworkTemporalConv2(para); useFirstNLayer = length(para.nFilter)*3+1; layer = layer(1:useFirstNLayer); paraLSTM = para; paraLSTM.inputDim = layer{end}.dim(1); lay...
github
singaxiong/SignalGraph-master
computeGlobalCMVN.m
.m
SignalGraph-master/prototypes/computeGlobalCMVN.m
1,256
utf_8
9678e1a367a055ab8889774be79e90c7
% given a network and some data, compute the mean and variance of the % network output. This function is usually used to determine the global % mean and variance normalization parameters in the preprocessing. % Xiong Xiao % function [W, b] = computeGlobalCMVN(Visible, nUttUsed, para, layer) if exist('nUttUsed')==0 || ...
github
singaxiong/SignalGraph-master
ConnectGraph.m
.m
SignalGraph-master/prototypes/ConnectGraph.m
1,516
utf_8
5a5894f351280c13c4cbe1387d3201d9
% giving an array of nodes, assume they are connected from the first node % to the last node. Automatically derive the connections parameters and % dimensions function layer = ConnectGraph(layer) % generate connections to the immediate parent nodes (prev) for i=1:length(layer) if strcmpi(class(layer{i}), 'InputNo...
github
singaxiong/SignalGraph-master
genNetworkFeedForward_pool.m
.m
SignalGraph-master/prototypes/genNetworkFeedForward_pool.m
915
utf_8
eb83a49827fd227ed4321a3293624e3a
% poolType = 'mean' or 'max' % poolLayer = integer layer number after which we do the pooling function layer = genNetworkFeedForward_pool(inputDim, hiddenLayerSize, outputDim, costFn, poolType, poolAfterNlayer, LastActivation4MSE) if nargin<7 LastActivation4MSE = 'linear'; end layer = genNetworkFeedForward_v2(inpu...
github
singaxiong/SignalGraph-master
VerifyPreprocessingTree.m
.m
SignalGraph-master/prototypes/VerifyPreprocessingTree.m
317
utf_8
d3e4aaedd5e7b60e5f618556d7fbfce0
% verify that the global MVN is correct. % Xiong Xiao function [processing] = VerifyPreprocessingTree(layer, Visible, para, nUttUsed) if exist('nUttUsed')==0 || isempty(nUttUsed) nUttUsed = 500; end [W, b] = computeGlobalCMVN(Visible, nUttUsed, para, layer); plot(-b); hold on; plot(1./diag(W)); hold off end
github
singaxiong/SignalGraph-master
genNetworkSTFT.m
.m
SignalGraph-master/prototypes/genNetworkSTFT.m
585
utf_8
74bc5df118d851e14b923b1b7af1d053
% create a sub network that takes in waveforms and produces fourier % coefficients function layer = genNetworkSTFT(input_idx, nCh, usedChannel, nFFT) layer{1} = InputNode(input_idx, nCh); nUsedChannel = length(usedChannel); if nUsedChannel < nCh layer{end+1} = ElementSelectNode(usedChannel); end featDim = nUsedC...
github
francois-a/llsmtools-master
readtiff.m
.m
llsmtools-master/iofunc/readtiff.m
2,822
utf_8
2034b18fd357c37154f6a81467fa4064
%[s] = readtiff(filepath) loads a tiff file or stack using libtiff % This function is ~1.5-2x faster than imread, useful for large stacks, % and supports a wider range of TIFF formats (see below) % % Inputs: % filepath : path to TIFF file to read from % % Optional inputs % range : range of pages to read from...
github
francois-a/llsmtools-master
writetiff.m
.m
llsmtools-master/iofunc/writetiff.m
1,981
utf_8
4f832daddca9584d3f26b367ec08acc0
%writetiff(img, filepath, varargin) writes a TIFF stack using libtiff % Stores TIFFs as 64-bit % Francois Aguet, 05/21/2013 function writetiff(img, filepath, varargin) ip = inputParser; ip.CaseSensitive = false; ip.addParamValue('Compression', 'lzw', @(x) any(strcmpi(x, {'none', 'lzw'}))); ip.addParamValue('Mode', '...