Collapse model

egs/ami/s5/cmd.sh

@@ -12,26 +12,19 @@
 
 export train_cmd="queue.pl --mem 1G"
 export decode_cmd="queue.pl --mem 2G"
-# the use of cuda_cmd is deprecated but it is sometimes still used in nnet1
-# scripts.
+# the use of cuda_cmd is deprecated, used only in 'nnet1',
 export cuda_cmd="queue.pl --gpu 1 --mem 20G"
 
-# the rest of this file is present for historical reasons.
-# In general it's best to rely on conf/queue.conf for cluster-specific
-# configuration.
+if [ "$(hostname -d)" == "fit.vutbr.cz" ]; then
+  queue_conf=$HOME/queue_conf/default.conf # see example /homes/kazi/iveselyk/queue_conf/default.conf,
+  export train_cmd="queue.pl --config $queue_conf --mem 2G --matylda 0.2"
+  export decode_cmd="queue.pl --config $queue_conf --mem 3G --matylda 0.1"
+  export cuda_cmd="queue.pl --config $queue_conf --gpu 1 --mem 10G --tmp 40G"
+fi
 
 # On Eddie use:
 #export train_cmd="queue.pl -P inf_hcrc_cstr_nst -l h_rt=08:00:00"
 #export decode_cmd="queue.pl -P inf_hcrc_cstr_nst  -l h_rt=05:00:00 -pe memory-2G 4"
 #export highmem_cmd="queue.pl -P inf_hcrc_cstr_nst -l h_rt=05:00:00 -pe memory-2G 4"
 #export scoring_cmd="queue.pl -P inf_hcrc_cstr_nst  -l h_rt=00:20:00"
 
-if [ "$(hostname -d)" == "fit.vutbr.cz" ]; then
-  # BUT cluster:
-  queue="all.q@@blade,all.q@@speech"
-  storage="matylda5"
-  export train_cmd="queue.pl -q $queue -l ram_free=1.5G,mem_free=1.5G,${storage}=1"
-  export decode_cmd="queue.pl -q $queue -l ram_free=2.5G,mem_free=2.5G,${storage}=0.5"
-  export cuda_cmd="queue.pl -q long.q -l gpu=1"
-fi
-

egs/ami/s5/run_ihm.sh

@@ -30,7 +30,7 @@ LM=$final_lm.pr1-7
 # Avoiding re-download, using 'wget --continue ...',
 if [ $stage -le 0 ]; then
   [ -e data/local/downloads/wget_${mic}.sh ] && \
-    echo "$data/local/downloads/wget_${mic}.sh already exists, better quit than re-download... (use --stage N)" && \
+    echo "data/local/downloads/wget_${mic}.sh already exists, better quit than re-download... (use --stage N)" && \
     exit 1
   local/ami_download.sh $mic $AMI_DIR
 fi

egs/ami/s5b/cmd.sh

@@ -12,25 +12,17 @@
 
 export train_cmd="queue.pl --mem 1G"
 export decode_cmd="queue.pl --mem 2G"
-# the use of cuda_cmd is deprecated but it is sometimes still used in nnet1
-# scripts.
+# the use of cuda_cmd is deprecated, used only in 'nnet1',
 export cuda_cmd="queue.pl --gpu 1 --mem 20G"
 
-# the rest of this file is present for historical reasons.
-# In general it's best to rely on conf/queue.conf for cluster-specific
-# configuration.
+if [ "$(hostname -d)" == "fit.vutbr.cz" ]; then
+  queue_conf=$HOME/queue_conf/default.conf # see example /homes/kazi/iveselyk/queue_conf/default.conf,
+  export train_cmd="queue.pl --config $queue_conf --mem 2G --matylda 0.2"
+  export decode_cmd="queue.pl --config $queue_conf --mem 3G --matylda 0.1"
+  export cuda_cmd="queue.pl --config $queue_conf --gpu 1 --mem 10G --tmp 40G"
+fi
 
 # On Eddie use:
 #export train_cmd="queue.pl -P inf_hcrc_cstr_nst -l h_rt=08:00:00"
 #export decode_cmd="queue.pl -P inf_hcrc_cstr_nst  -l h_rt=05:00:00 -pe memory-2G 4"
 
-if [ "$(hostname -d)" == "fit.vutbr.cz" ]; then
-  # BUT cluster:
-  queue="all.q@@blade,all.q@@speech"
-  gpu_queue="long.q@@gpu"
-  storage="matylda5"
-  export train_cmd="queue.pl -q $queue -l ram_free=1.5G,mem_free=1.5G,${storage}=1"
-  export decode_cmd="queue.pl -q $queue -l ram_free=2.5G,mem_free=2.5G,${storage}=0.5"
-  export cuda_cmd="queue.pl -q $gpu_queue -l gpu=1"
-fi
-

egs/ami/s5b/local/chain/tuning/run_tdnn_1f.sh

@@ -0,0 +1,266 @@
+#!/bin/bash
+
+# same as 1e but uses batchnorm components instead of renorm also adding
+# proportional-shrink 10, trained with 4 epochs
+
+# System               tdnn1e_sp_bi_ihmali tdnn1f_sp_bi_ihmali
+# WER on dev        39.2      37.5
+# WER on eval        42.8      41.3
+# Final train prob      -0.235518 -0.195525
+# Final valid prob      -0.275605 -0.258708
+# Final train prob (xent)      -2.75633  -2.42821
+# Final valid prob (xent)      -2.88854  -2.63458
+
+set -e -o pipefail
+# First the options that are passed through to run_ivector_common.sh
+# (some of which are also used in this script directly).
+stage=0
+mic=ihm
+nj=30
+min_seg_len=1.55
+use_ihm_ali=false
+train_set=train_cleaned
+gmm=tri3_cleaned  # the gmm for the target data
+ihm_gmm=tri3  # the gmm for the IHM system (if --use-ihm-ali true).
+num_threads_ubm=32
+ivector_transform_type=pca
+nnet3_affix=_cleaned  # cleanup affix for nnet3 and chain dirs, e.g. _cleaned
+
+# The rest are configs specific to this script.  Most of the parameters
+# are just hardcoded at this level, in the commands below.
+train_stage=-10
+tree_affix=  # affix for tree directory, e.g. "a" or "b", in case we change the configuration.
+tdnn_affix=1f  #affix for TDNN directory, e.g. "a" or "b", in case we change the configuration.
+common_egs_dir=  # you can set this to use previously dumped egs.
+
+# End configuration section.
+echo "$0 $@"  # Print the command line for logging
+
+. ./cmd.sh
+. ./path.sh
+. ./utils/parse_options.sh
+
+
+if ! cuda-compiled; then
+  cat <<EOF && exit 1
+This script is intended to be used with GPUs but you have not compiled Kaldi with CUDA
+If you want to use GPUs (and have them), go to src/, and configure and make on a machine
+where "nvcc" is installed.
+EOF
+fi
+
+local/nnet3/run_ivector_common.sh --stage $stage \
+                                  --mic $mic \
+                                  --nj $nj \
+                                  --min-seg-len $min_seg_len \
+                                  --train-set $train_set \
+                                  --gmm $gmm \
+                                  --num-threads-ubm $num_threads_ubm \
+                                  --ivector-transform-type "$ivector_transform_type" \
+                                  --nnet3-affix "$nnet3_affix"
+
+# Note: the first stage of the following script is stage 8.
+local/nnet3/prepare_lores_feats.sh --stage $stage \
+                                   --mic $mic \
+                                   --nj $nj \
+                                   --min-seg-len $min_seg_len \
+                                   --use-ihm-ali $use_ihm_ali \
+                                   --train-set $train_set
+
+if $use_ihm_ali; then
+  gmm_dir=exp/ihm/${ihm_gmm}
+  ali_dir=exp/${mic}/${ihm_gmm}_ali_${train_set}_sp_comb_ihmdata
+  lores_train_data_dir=data/$mic/${train_set}_ihmdata_sp_comb
+  tree_dir=exp/$mic/chain${nnet3_affix}/tree_bi${tree_affix}_ihmdata
+  lat_dir=exp/$mic/chain${nnet3_affix}/${gmm}_${train_set}_sp_comb_lats_ihmdata
+  dir=exp/$mic/chain${nnet3_affix}/tdnn${tdnn_affix}_sp_bi_ihmali
+  # note: the distinction between when we use the 'ihmdata' suffix versus
+  # 'ihmali' is pretty arbitrary.
+else
+  gmm_dir=exp/${mic}/$gmm
+  ali_dir=exp/${mic}/${gmm}_ali_${train_set}_sp_comb
+  lores_train_data_dir=data/$mic/${train_set}_sp_comb
+  tree_dir=exp/$mic/chain${nnet3_affix}/tree_bi${tree_affix}
+  lat_dir=exp/$mic/chain${nnet3_affix}/${gmm}_${train_set}_sp_comb_lats
+  dir=exp/$mic/chain${nnet3_affix}/tdnn${tdnn_affix}_sp_bi
+fi
+
+train_data_dir=data/$mic/${train_set}_sp_hires_comb
+train_ivector_dir=exp/$mic/nnet3${nnet3_affix}/ivectors_${train_set}_sp_hires_comb
+final_lm=`cat data/local/lm/final_lm`
+LM=$final_lm.pr1-7
+
+
+for f in $gmm_dir/final.mdl $lores_train_data_dir/feats.scp \
+   $train_data_dir/feats.scp $train_ivector_dir/ivector_online.scp; do
+  [ ! -f $f ] && echo "$0: expected file $f to exist" && exit 1
+done
+
+
+if [ $stage -le 11 ]; then
+  if [ -f $ali_dir/ali.1.gz ]; then
+    echo "$0: alignments in $ali_dir appear to already exist.  Please either remove them "
+    echo " ... or use a later --stage option."
+    exit 1
+  fi
+  echo "$0: aligning perturbed, short-segment-combined ${maybe_ihm}data"
+  steps/align_fmllr.sh --nj $nj --cmd "$train_cmd" \
+     ${lores_train_data_dir} data/lang $gmm_dir $ali_dir
+fi
+
+[ ! -f $ali_dir/ali.1.gz ] && echo  "$0: expected $ali_dir/ali.1.gz to exist" && exit 1
+
+if [ $stage -le 12 ]; then
+  echo "$0: creating lang directory with one state per phone."
+  # Create a version of the lang/ directory that has one state per phone in the
+  # topo file. [note, it really has two states.. the first one is only repeated
+  # once, the second one has zero or more repeats.]
+  if [ -d data/lang_chain ]; then
+    if [ data/lang_chain/L.fst -nt data/lang/L.fst ]; then
+      echo "$0: data/lang_chain already exists, not overwriting it; continuing"
+    else
+      echo "$0: data/lang_chain already exists and seems to be older than data/lang..."
+      echo " ... not sure what to do.  Exiting."
+      exit 1;
+    fi
+  else
+    cp -r data/lang data/lang_chain
+    silphonelist=$(cat data/lang_chain/phones/silence.csl) || exit 1;
+    nonsilphonelist=$(cat data/lang_chain/phones/nonsilence.csl) || exit 1;
+    # Use our special topology... note that later on may have to tune this
+    # topology.
+    steps/nnet3/chain/gen_topo.py $nonsilphonelist $silphonelist >data/lang_chain/topo
+  fi
+fi
+
+if [ $stage -le 13 ]; then
+  # Get the alignments as lattices (gives the chain training more freedom).
+  # use the same num-jobs as the alignments
+  steps/align_fmllr_lats.sh --nj 100 --cmd "$train_cmd" ${lores_train_data_dir} \
+    data/lang $gmm_dir $lat_dir
+  rm $lat_dir/fsts.*.gz # save space
+fi
+
+if [ $stage -le 14 ]; then
+  # Build a tree using our new topology.  We know we have alignments for the
+  # speed-perturbed data (local/nnet3/run_ivector_common.sh made them), so use
+  # those.
+  if [ -f $tree_dir/final.mdl ]; then
+    echo "$0: $tree_dir/final.mdl already exists, refusing to overwrite it."
+    exit 1;
+  fi
+  steps/nnet3/chain/build_tree.sh --frame-subsampling-factor 3 \
+      --context-opts "--context-width=2 --central-position=1" \
+      --leftmost-questions-truncate -1 \
+      --cmd "$train_cmd" 4200 ${lores_train_data_dir} data/lang_chain $ali_dir $tree_dir
+fi
+
+xent_regularize=0.1
+
+if [ $stage -le 15 ]; then
+  echo "$0: creating neural net configs using the xconfig parser";
+
+  num_targets=$(tree-info $tree_dir/tree |grep num-pdfs|awk '{print $2}')
+  learning_rate_factor=$(echo "print 0.5/$xent_regularize" | python)
+
+  mkdir -p $dir/configs
+  cat <<EOF > $dir/configs/network.xconfig
+  input dim=100 name=ivector
+  input dim=40 name=input
+
+  # please note that it is important to have input layer with the name=input
+  # as the layer immediately preceding the fixed-affine-layer to enable
+  # the use of short notation for the descriptor
+  fixed-affine-layer name=lda input=Append(-1,0,1,ReplaceIndex(ivector, t, 0)) affine-transform-file=$dir/configs/lda.mat
+
+  # the first splicing is moved before the lda layer, so no splicing here
+  relu-batchnorm-layer name=tdnn1 dim=450
+  relu-batchnorm-layer name=tdnn2 input=Append(-1,0,1) dim=450
+  relu-batchnorm-layer name=tdnn3 input=Append(-1,0,1) dim=450
+  relu-batchnorm-layer name=tdnn4 input=Append(-3,0,3) dim=450
+  relu-batchnorm-layer name=tdnn5 input=Append(-3,0,3) dim=450
+  relu-batchnorm-layer name=tdnn6 input=Append(-3,0,3) dim=450
+  relu-batchnorm-layer name=tdnn7 input=Append(-3,0,3) dim=450
+
+  ## adding the layers for chain branch
+  relu-batchnorm-layer name=prefinal-chain input=tdnn7 dim=450 target-rms=0.5
+  output-layer name=output include-log-softmax=false dim=$num_targets max-change=1.5
+
+  # adding the layers for xent branch
+  # This block prints the configs for a separate output that will be
+  # trained with a cross-entropy objective in the 'chain' models... this
+  # has the effect of regularizing the hidden parts of the model.  we use
+  # 0.5 / args.xent_regularize as the learning rate factor- the factor of
+  # 0.5 / args.xent_regularize is suitable as it means the xent
+  # final-layer learns at a rate independent of the regularization
+  # constant; and the 0.5 was tuned so as to make the relative progress
+  # similar in the xent and regular final layers.
+  relu-batchnorm-layer name=prefinal-xent input=tdnn7 dim=450 target-rms=0.5
+  output-layer name=output-xent dim=$num_targets learning-rate-factor=$learning_rate_factor max-change=1.5
+
+EOF
+
+  steps/nnet3/xconfig_to_configs.py --xconfig-file $dir/configs/network.xconfig --config-dir $dir/configs/
+fi
+
+if [ $stage -le 16 ]; then
+  if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $dir/egs/storage ]; then
+    utils/create_split_dir.pl \
+     /export/b0{5,6,7,8}/$USER/kaldi-data/egs/ami-$(date +'%m_%d_%H_%M')/s5b/$dir/egs/storage $dir/egs/storage
+  fi
+
+  steps/nnet3/chain/train.py --stage $train_stage \
+    --cmd "$decode_cmd" \
+    --feat.online-ivector-dir $train_ivector_dir \
+    --feat.cmvn-opts "--norm-means=false --norm-vars=false" \
+    --chain.xent-regularize $xent_regularize \
+    --chain.leaky-hmm-coefficient 0.1 \
+    --chain.l2-regularize 0.00005 \
+    --chain.apply-deriv-weights false \
+    --chain.lm-opts="--num-extra-lm-states=2000" \
+    --egs.dir "$common_egs_dir" \
+    --egs.opts "--frames-overlap-per-eg 0" \
+    --egs.chunk-width 150 \
+    --trainer.num-chunk-per-minibatch 128 \
+    --trainer.frames-per-iter 1500000 \
+    --trainer.num-epochs 4 \
+    --trainer.optimization.num-jobs-initial 2 \
+    --trainer.optimization.num-jobs-final 12 \
+    --trainer.optimization.initial-effective-lrate 0.001 \
+    --trainer.optimization.final-effective-lrate 0.0001 \
+    --trainer.optimization.proportional-shrink 10 \
+    --trainer.max-param-change 2.0 \
+    --cleanup.remove-egs true \
+    --feat-dir $train_data_dir \
+    --tree-dir $tree_dir \
+    --lat-dir $lat_dir \
+    --dir $dir
+fi
+
+
+graph_dir=$dir/graph_${LM}
+if [ $stage -le 17 ]; then
+  # Note: it might appear that this data/lang_chain directory is mismatched, and it is as
+  # far as the 'topo' is concerned, but this script doesn't read the 'topo' from
+  # the lang directory.
+  utils/mkgraph.sh --self-loop-scale 1.0 data/lang_${LM} $dir $graph_dir
+fi
+
+if [ $stage -le 18 ]; then
+  rm $dir/.error 2>/dev/null || true
+  for decode_set in dev eval; do
+      (
+      steps/nnet3/decode.sh --acwt 1.0 --post-decode-acwt 10.0 \
+          --nj $nj --cmd "$decode_cmd" \
+          --online-ivector-dir exp/$mic/nnet3${nnet3_affix}/ivectors_${decode_set}_hires \
+          --scoring-opts "--min-lmwt 5 " \
+         $graph_dir data/$mic/${decode_set}_hires $dir/decode_${decode_set} || exit 1;
+      ) || touch $dir/.error &
+  done
+  wait
+  if [ -f $dir/.error ]; then
+    echo "$0: something went wrong in decoding"
+    exit 1
+  fi
+fi
+exit 0