From 18b0e3d993ec5ce8e97a6affb533c9ace940bfff Mon Sep 17 00:00:00 2001 From: Determinant Date: Tue, 1 Mar 2016 00:33:28 +0800 Subject: ... --- kaldi_decode/Makefile | 2 +- kaldi_decode/README.timit | 2 +- kaldi_io/tools/convert_from_kaldi_pretrain.sh | 4 +- tutorial/howto_pretrain_from_kaldi.rst | 105 +++++++++++++++----------- 4 files changed, 68 insertions(+), 45 deletions(-) diff --git a/kaldi_decode/Makefile b/kaldi_decode/Makefile index e3a7c2d..2ccedfb 100644 --- a/kaldi_decode/Makefile +++ b/kaldi_decode/Makefile @@ -25,7 +25,7 @@ OBJ_SUBDIR := $(addprefix $(OBJ_DIR)/,$(SUBDIR)) KL := $(KDIR)/src/feat/kaldi-feat.a $(KDIR)/src/cudamatrix/kaldi-cudamatrix.a $(KDIR)/src/matrix/kaldi-matrix.a $(KDIR)/src/base/kaldi-base.a $(KDIR)/src/util/kaldi-util.a $(KDIR)/src/hmm/kaldi-hmm.a $(KDIR)/src/tree/kaldi-tree.a $(KDIR)/src/nnet/kaldi-nnet.a $(BLAS_LDFLAGS) -build: $(OBJ_DIR) $(LUA_DIR) $(OBJ_SUBDIR) $(OBJS) +build: $(OBJ_DIR) $(OBJ_SUBDIR) $(OBJS) $(OBJ_DIR)/%.o: %.cc g++ -c -o $@ $< -Wall $(KALDIINCLUDE) -DHAVE_ATLAS -DKALDI_DOUBLEPRECISION=0 -DHAVE_POSIX_MEMALIGN -DLUA_USE_APICHECK -I $(LUA_INCDIR) -I $(INC_PATH) $(CFLAGS) $(OBJ_DIR)/nnet-forward: $(OBJ_DIR)/src/nnet-forward.o diff --git a/kaldi_decode/README.timit b/kaldi_decode/README.timit index 7fac918..0a3e33a 100755 --- a/kaldi_decode/README.timit +++ b/kaldi_decode/README.timit @@ -4,7 +4,7 @@ source cmd.sh gmmdir=/speechlab/users/mfy43/timit/s5/exp/tri3/ data_fmllr=/speechlab/users/mfy43/timit/s5/data-fmllr-tri3/ -dir=/speechlab/users/mfy43/timit/s5/exp/dnn4_nerv_prepare/ +dir=/speechlab/users/mfy43/timit/s5/exp/dnn4_nerv_dnn/ nerv_config=/speechlab/users/mfy43/nerv/nerv/examples/timit_baseline2.lua decode=/speechlab/users/mfy43/nerv/install/bin/decode_with_nerv.sh diff --git a/kaldi_io/tools/convert_from_kaldi_pretrain.sh b/kaldi_io/tools/convert_from_kaldi_pretrain.sh index 78f532f..81fe840 100755 --- a/kaldi_io/tools/convert_from_kaldi_pretrain.sh +++ b/kaldi_io/tools/convert_from_kaldi_pretrain.sh @@ -17,12 +17,13 @@ dir=$6 [[ -z $data_fmllr ]] && data_fmllr=data-fmllr-tri3 [[ -z $alidir ]] && alidir=exp/tri3_ali -[[ -z $dir ]] && dir=exp/dnn4_nerv_prepare +[[ -z $dir ]] && dir=exp/dnn4_nerv_dnn [[ -z $data ]] && data=$data_fmllr/train_tr90 [[ -z $data_cv ]] && data_cv=$data_fmllr/train_cv10 kaldi_to_nerv=$nerv_kaldi/tools/kaldi_to_nerv mkdir $dir -p mkdir $dir/log -p + ###### PREPARE DATASETS ###### cp $data/feats.scp $dir/train_sorted.scp cp $data_cv/feats.scp $dir/cv.scp @@ -44,6 +45,7 @@ nnet-initialize --binary=false $nnet_proto $nnet_init $kaldi_to_nerv $nnet_init $dir/nnet_output.nerv $hid_num $kaldi_to_nerv <(nnet-copy --binary=false $pretrain_dir/${hid_num}.dbn -) $dir/nnet_init.nerv $kaldi_to_nerv <(nnet-copy --binary=false $pretrain_dir/final.feature_transform -) $dir/nnet_trans.nerv + ###### PREPARE FOR DECODING ##### echo "Using PDF targets from dirs '$alidir' '$alidir_cv'" # training targets in posterior format, diff --git a/tutorial/howto_pretrain_from_kaldi.rst b/tutorial/howto_pretrain_from_kaldi.rst index ff6ef3d..6b8253a 100644 --- a/tutorial/howto_pretrain_from_kaldi.rst +++ b/tutorial/howto_pretrain_from_kaldi.rst @@ -7,38 +7,53 @@ How to Use a Pretrained nnet Model from Kaldi NERV finetune. Finally it shows two possible ways to decode the finetuned model in Kaldi framework. -- Locate the egs/timit inside Kaldi trunk directory. +- Note: in this tutorial, we use the following notations to denote the directory prefix: -- Configure ``cmd.sh`` and ``path.sh`` according to your machine setting. + - ````: the path of NERV (the location of outer most directory ``nerv``) -- Open the ``run.sh`` and locate the line saying ``exit 0 # From this point - you can run Karel's DNN: local/nnet/run_dnn.sh``. Uncomment this line. This - is because in this tutorial, we only want to train a basic tri-phone DNN, - so we simply don't do MMI training, system combination or fancy things like - these. + - ````: the working directory of timit (the location of directory ``timit/s5``) -- Run ``./run.sh`` to start the training stages. After that, we will get +- Locate the ``egs/timit`` inside Kaldi trunk directory. + +- Configure ``/cmd.sh`` and ``/path.sh`` according to your machine setting. + +- Open the ``/run.sh`` and locate the line saying + + :: + + exit 0 # From this point you can run Karel's DNN: local/nnet/run_dnn.sh + . + Uncomment this line. This is because in this tutorial, we only want to train + a basic tri-phone DNN, so we simply don't do MMI training, system combination + or fancy things like these. + +- Run ``./run.sh`` (at ````) to start the training stages. After that, we will get tri-phone GMM-HMM trained and the aligned labels. Let's move forward to pretrain a DNN. -- Open ``local/nnet/run_dnn.sh``, there are again several stages. Note that - the first stage is what we actually need (pretraining the DNN), since in - this tutorial we want to demonstrate how to get the pretrained model from - stage 1, replace stage 2 with NERV (finetune per-frame cross-entropy), and - decode using the finetuned network. However, here we add a line ``exit 0`` - after stage 2 to preserve stage 2 in order to compare the NERV result - against the standard one (the decode result using finetuned model produced - by the original stage 2). +- Open ``/local/nnet/run_dnn.sh``, there are again several stages. + Note that the first stage is what we actually need (pretraining the DNN), + since in this tutorial we want to demonstrate how to get the pretrained model + from stage 1, replace stage 2 with NERV (finetune per-frame cross-entropy), + and decode using the finetuned network. However, here we add a line ``exit + 0`` after stage 2 to preserve stage 2 in order to compare the NERV result + against the standard one (the decode result using finetuned model produced by + the original stage 2). -- Run ``local/nnet/run_dnn.sh`` (first two stages). +- Run ``local/nnet/run_dnn.sh`` (at ````, for first two stages). - You'll find directory like ``dnn4_pretrain-dbn`` and - ``dnn4_pretrain-dbn_dnn`` inside the ``exp/``. They correspond to stage 1 and - stage 2 respectively. To use NERV to do stage 2 instead, we need the - pretrained network and the global transformation from stage 1: + ``dnn4_pretrain-dbn_dnn`` inside the ``/exp/``. They correspond + to stage 1 and stage 2 respectively. To use NERV to do stage 2 instead, we + need the pretrained network and the global transformation from stage 1: - - Check the file ``exp/dnn4_pretrain-dbn/6.dbn`` exists. (pretrained network) - - Check the file ``exp/dnn4_pretrain-dbn/tr_splice5_cmvn-g.nnet`` exists. (global transformation) - - Run script from ``kaldi_io/tools/convert_from_kaldi_pretrain.sh`` to + - Check the file ``/exp/dnn4_pretrain-dbn/6.dbn`` exists. + (pretrained network) + + - Check the file + ``/exp/dnn4_pretrain-dbn/tr_splice5_cmvn-g.nnet`` exists. + (global transformation) + + - Run script from ``/speech/kaldi_io/tools/convert_from_kaldi_pretrain.sh`` to generate the parameters for the output layer and the script files for training and cross-validation set. @@ -47,18 +62,25 @@ How to Use a Pretrained nnet Model from Kaldi example, ``affine0_ltp`` and ``bias0``. These names should correspond to the identifiers used in the declaration of the network. Luckily, this tutorial comes with a written network declaration at - ``nerv/examples/timit_baseline2.lua``. + ``/nerv/examples/timit_baseline2.lua``. + +- Copy the file ``/nerv/examples/timit_baseline2.lua`` to + ``/timit_mybaseline.lua``, and change the line containing + ``/speechlab`` to your own setting. -- Copy the file ``nerv/examples/timit_baseline2.lua`` to - ``timit_mybaseline.lua``, and change the line containing ``/speechlab`` to - your own setting. +- Start the NERV training by + + :: + + /install/bin/nerv /nerv/examples/asr_trainer.lua timit_mybaseline.lua -- Start the NERV training by ``install/bin/nerv nerv/examples/asr_trainer.lua timit_mybaseline.lua``. + (at ````). - - ``install/bin/nerv`` is the program which sets up the NERV environment, + - ``/install/bin/nerv`` is the program which sets up the NERV environment, - - followed by an argument ``nerv/examples/asr_trainer.lua`` which is the script - you actually want to run (the general DNN training scheduler), + - followed by an argument ``/nerv/examples/asr_trainer.lua`` which + is the script you actually want to run (the general DNN training + scheduler), - followed by an argument ``timit_mybaseline.lua`` to the scheduler, specifying the network you want to train and some relevant settings, such @@ -74,19 +96,17 @@ How to Use a Pretrained nnet Model from Kaldi global transformation chunk file once used in training. This part lets the decoder know about the set of parameters for decoding. - - Copy the script ``nerv/speech/kaldi_io/README.timit`` to your Kaldi - working directory (``timit/s5``) and modify the paths listed in the - script. + - Copy the script ``/nerv/speech/kaldi_io/README.timit`` to + ```` and modify the paths listed in the script. - - Run the modified ``README.timit`` in ``s5`` directory (where there is the - ``path.sh``). + - Run the modified ``README.timit`` (at ````). - - After decoding, run ``bash RESULT exp/dnn4_nerv`` to see the results. + - After decoding, run ``bash RESULT exp/dnn4_nerv_dnn`` to see the results. - Plan B: In this plan, we manually convert the trained model back to Kaldi nnet format, and use Kaldi to decode. - - Create a copy of ``nerv/speech/kaldi_io/tools/nerv_to_kaldi.lua``. + - Create a copy of ``/nerv/speech/kaldi_io/tools/nerv_to_kaldi.lua``. - Modify the list named ``lnames`` to list the name of layers you want to put into the output Kaldi parameter file in order. (You don't actually @@ -103,14 +123,15 @@ How to Use a Pretrained nnet Model from Kaldi :: cat your_trained_params.nerv your_global_trans.nerv > all.nerv - install/bin/nerv nerv_to_kaldi.lua timit_mybaseline.lua all.nerv your_kaldi_output.nnet + /install/bin/nerv nerv_to_kaldi.lua timit_mybaseline.lua all.nerv your_kaldi_output.nnet - - Finally, locate the directory of stage 2: ``exp/dnn4_pretrain-dbn_dnn`` - and temporarily change the symbolic link for the final network file to the converted one: + - Finally, locate the directory of stage 2: + ``/exp/dnn4_pretrain-dbn_dnn`` and temporarily change the + symbolic link for the final network file to the converted one: :: - cd exp/dnn4_pretrain-dbn_dnn + cd /exp/dnn4_pretrain-dbn_dnn mv final.nnet final.nnet.orig ln -sv your_kaldi_output.nnet final.nnet -- cgit v1.2.3