Activity for Karel Vesely
-
Karel Vesely posted a comment on discussion Open Discussion
Hi, it sholud be working well now! Thanks for finding the bug! K.
-
trunk,nnet1,mmi : bugfix in inital data filteri...
-
Yes it is. We should both thank to 'Lukas Burget', he is the original author of the...
-
Ok, I'll fix that. Thanks for finding the bug! K. Dne 15. 7. 2015 v 13:32 Daniel...
-
Hi, the problem is that you pre-trained only 5 layers, while the DNN training script...
-
trunk,nnet1: fixing 'blocksoftmax' example, add...
-
trunk,nnet1: updating 'blocksoftmax' example (r...
-
Hi, Yes it should work like that. K. Dne 4. 7. 2015 v 19:56 tfpeach napsal(a): Hi,...
-
trunk,ami : refactoring the ami recipe which wi...
-
trunk,nnet1 : bugfix in cuda-gpu detection (mmi...
-
Hi, no, the adaptation of NN weights to individual speakers is not supported. The...
-
trunk,nnet1: cosmetic changes, allowing initial...
-
Hi, you need to put whole <ParallelComponent> description to single line in the config...
-
trunk,nnet1 : adding BLSTM code from Ni Chongji...
-
Hi, most likely you have run out of RAM (you'll see this with utility 'top'), by...
-
trunk,nnet1: adding tool 'cuda-gpu-available', ...
-
trunk,nnet1 : lstm, changing indentation to be ...
-
trunk,nnet1 : cosmetic changes to 2D-CNN script,
-
trunk,nnet1: updating the 'usage' messages in s...
-
Hello, no, in order to put 2 NNs in parallel you'll need to create a "prototype"...
-
trunk,nnet1 : cosmetic change in MultiTaskLoss
-
You don't need to split the input features '<ParallelComponent>' will do it for you....
-
Hi, this can be done easily with nnet1, there is a Component '<ParallelComponent>'...
-
trunk,nnet1 : implementing multi-task loss func...
-
Hi, most likely the input-feature dimension is wrong, i.e. different from the RM...
-
trunk,nnet1: adding the LSTM equation diagram f...
-
sandbox/cudnn : syncing branch with trunk, will...
-
Hi, I have noticed the pipeline with 'paste-post' consumed quite a lot of memory,...
-
Hi, I have noticed the pipeline with 'paste-post' consumed quite a lot of memory,...
-
Hi, I have noticed the pipeline with 'paste-post' consumed quite a lot of memory,...
-
trunk,nnet1: updating paste-post.cc, improving ...
-
OK, I have tried to fix it, but I haven't run the example script to check. Dan
-
No, there has not, and since it relates to the nnet1 setup which Karel maintains...
-
It looks like that program was added recently by Karel (Karel, please run cpplint.py...
-
Hi, the variable 'cmvn_opts' is contains options for tool 'apply-cmvn'. If you go...
-
You will have to see how data-fbank-multisoftmax/train/ was created, but IIRC in...
-
trunk,tools: beamformit, updating the patch (th...
-
trunk,tools : adding patch to beamformit tool, ...
-
Hi, hard to tell where the problem is, the assert in 'nnet-forward.cc:158' is right...
-
trunk,cudamatrix : removing compute capability ...
-
How did this happen? Someone else gave you a trained neural network or did you train...
-
trunk,nnet1 : changing magic constant in 'nnet-...
-
Hi, the affine transforms contain variable "<LearnRateCoef> 1" you can set it to...
-
Hi, do you need to use the nnet2 setup? If not, it should be easy, you build an RBM...
-
trunk,nnet1 : adding prior re-estimation after ...
-
trunk,nnet1 : sMBR bugfix (accidentally flipped...
-
trunk,nnet1: tedlium, adding DNN baseline for L...
-
trunk,nnet1 : tedlium, updating DNN recipe with...
-
trunk,nnet1: removing scrtipt (use run_dnn_bn.sh)
-
trunk,nnet1 : bugfix in mmi/mpe script (unshuff...
-
trunk,nnet1 : adding support to --one-silence-c...
-
trunk,nnet1: various enhancements, minor update...
-
dummy : just a test if svn commit work
-
trunk,nnet1 : bugfix in LSTM forwarding (missin...
-
Hi, looks like lots of insertions appeared, hard to say why, you can try: steps/nnet/train_mpe.sh...
-
Hi Lucian, something got wrong, the difference must be (<0.5%), typically (<0.2%)....
-
trunk,nnet1 : adding lstm example for tedlium (...
-
trunk,nnet1 : adding --time-shift to 'nnet-forw...
-
trunk,nnet1 : adding lstm example (needs to be ...
-
trunk,nnet1: adding the LSTM C++ code from Jiay...
-
trunk,nnet1 : refactoring nnet/nnet-loss.{h,cc}...
-
trunk,dnn1.dox : updating the doc based on the ...
-
trunk,dnn1.dox : updating the docs (incorporate...
-
trunk,nnet1: steps/nnet/make_bn_feats.sh, HTK f...
-
utils/copy_data_dir.sh - suppress message when ...
-
trunk,nnet1 : bugfix
-
Hi, seems like you are using very old GPU, which has 10x lower computing power than...
-
Okay I have the code already, it was in email from the author... Thanks, K. On 01/22/2015...
-
Hi Alim, thanks for the info, with classical feedforward DNNs, the frame accuracy...
-
trunk,nnet1: moving run_dnn.sh scripts to be co...
-
trunk,nnet1 : adding the tandem recipe,
-
trunk,bugfix : fixing bug in decode script with...
-
trunk,steps: adding the 'delta_opts' so there c...
-
trunk,nnet1: adding some bacward-compatibility ...
-
Hi, Gaussians with diagonal covariance matrix cannot fit well the data with correlations,...
-
Hi, the autoencoder example is in the trunk: trunk/egs/timit/s5/local/nnet/run_autoencoder.sh...
-
Hi, the autoencoder example is in the trunk: trunk/egs/timit/s5/local/nnet/run_autoencoder.sh...
-
trunk,nnet1: adding autoencoder example, demons...
-
trunk,nnet1: adding tool 'feat-to-post', Conver...
-
Hi, thanks to Dan for forwarding. What you need to do is very similar to training...
-
... Also you may have to replace the softmax nonlinearity with another one, such...
-
In nnet-loss.h, there are child classes of Component called Xent and Mse. You have...
-
Look for the functions Propagate and Backpropagate of virtual class Component in...
-
Karel would have to answer, since you're using the nnet1 setup. Dan
-
Thanks for your reply. Yes, maximizing the -MSE is also one of method for minimization...
-
Instead of modification the update rule to allow for minimization, you can \max -1...
-
I think the toolkit is probably set up to maximize the objective function, not minimize...
-
It is done beforehead, when preparing the input features, there are long vectors...
-
WARNING, TIMIT BASELINES CHANGE BY THIS COMMIT!
-
Dear Xinquan, which script do you refer to? If nnet1, there is example in ^/trunk/egs/rm/s5/local/nnet/run_cnn.sh...
-
Dear Ben, please have a look at the tool : nnet1-to-raw-nnet, and the script "steps/nnet2/convert_nnet1_to_nnet2.sh"....
-
trunk,openblas : sync of linux_openblas.mk with...
-
trunk,matrix: adding speed test which can be us...
-
trunk,nnet1: adding workaround to disable GPU c...
-
Hi Rayava, Sorry, I try not to focus on TNet anymore and there was no conversion...
-
sandbox/cudnn : adding demonstration of Sigmoid...
-
Creating branch for cuDNN and Thrust code examp...
-
Yes, the tool was long ago replaced by "nnet-copy", which accepts options --remove-{first,last}-layers,...
-
trunk,nnet: updating the training scripts,
-
trunk,nnet1: changing comment to the --label op...
1 >