run_ctc_phn.sh

#!/bin/bash

. ./cmd.sh ## You'll want to change cmd.sh to something that will work on your system.
           ## This relates to the queue.

stage=1
wsj0=/path/to/LDC93S6B
wsj1=/path/to/LDC94S13B

. utils/parse_options.sh

# add check for IRSTLM prune-lm
if ! prune-lm > /dev/null 2>&1; then
    echo "Error: prune-lm (part of IRSTLM) is not in path"
    echo "Make sure that you run tools/extras/install_irstlm.sh in the main Eesen directory;"
    echo " this is no longer installed by default."
    exit 1
fi


if [ $stage -le 1 ]; then
  echo =====================================================================
  echo "             Data Preparation and FST Construction                 "
  echo =====================================================================
  # Use the same datap prepatation script from Kaldi
  local/wsj_data_prep.sh $wsj0/??-{?,??}.? $wsj1/??-{?,??}.?  || exit 1;

  # Construct the phoneme-based lexicon from the CMU dict
  local/wsj_prepare_phn_dict.sh || exit 1;

  # Compile the lexicon and token FSTs
  utils/ctc_compile_dict_token.sh data/local/dict_phn data/local/lang_phn_tmp data/lang_phn || exit 1;

  # Compile the language-model FST and the final decoding graph TLG.fst
  local/wsj_decode_graph.sh data/lang_phn || exit 1;
fi

if [ $stage -le 2 ]; then
  echo =====================================================================
  echo "                    FBank Feature Generation                       "
  echo =====================================================================
  # Split the whole training data into training (95%) and cross-validation (5%) sets
  utils/subset_data_dir_tr_cv.sh --cv-spk-percent 5 data/train_si284 data/train_tr95 data/train_cv05 || exit 1

  # Generate the fbank features; by default 40-dimensional fbanks on each frame
  fbankdir=fbank
  for set in train_tr95 train_cv05; do
    steps/make_fbank.sh --cmd "$train_cmd" --nj 14 data/$set exp/make_fbank/$set $fbankdir || exit 1;
    utils/fix_data_dir.sh data/$set || exit;
    steps/compute_cmvn_stats.sh data/$set exp/make_fbank/$set $fbankdir || exit 1;
  done

  for set in test_dev93 test_eval92; do
    steps/make_fbank.sh --cmd "$train_cmd" --nj 8 data/$set exp/make_fbank/$set $fbankdir || exit 1;
    utils/fix_data_dir.sh data/$set || exit;
    steps/compute_cmvn_stats.sh data/$set exp/make_fbank/$set $fbankdir || exit 1;
  done
fi

if [ $stage -le 3 ]; then
  echo =====================================================================
  echo "                        Network Training                           "
  echo =====================================================================
  # Specify network structure and generate the network topology
  input_feat_dim=120   # dimension of the input features; we will use 40-dimensional fbanks with deltas and double deltas
  lstm_layer_num=4     # number of LSTM layers
  lstm_cell_dim=320    # number of memory cells in every LSTM layer

  dir=exp/train_phn_l${lstm_layer_num}_c${lstm_cell_dim}
  mkdir -p $dir

  target_num=`cat data/local/dict_phn/units.txt | wc -l`; target_num=$[$target_num+1]; # the number of targets 
                                                         # equals [the number of labels] + 1 (the blank)

  # Output the network topology
  utils/model_topo.py --input-feat-dim $input_feat_dim --lstm-layer-num $lstm_layer_num \
    --lstm-cell-dim $lstm_cell_dim --target-num $target_num > $dir/nnet.proto || exit 1;

  # Label sequences; simply convert words into their label indices
  utils/prep_ctc_trans.py data/lang_phn/lexicon_numbers.txt data/train_tr95/text "<UNK>" | gzip -c - > $dir/labels.tr.gz
  utils/prep_ctc_trans.py data/lang_phn/lexicon_numbers.txt data/train_cv05/text "<UNK>" | gzip -c - > $dir/labels.cv.gz

  # Train the network with CTC. Refer to the script for details about the arguments
  steps/train_ctc_parallel.sh --add-deltas true --num-sequence 10 --frame-num-limit 25000 \
    --learn-rate 0.00004 --report-step 1000 \
    data/train_tr95 data/train_cv05 $dir || exit 1;

  echo =====================================================================
  echo "                            Decoding                               "
  echo =====================================================================
  # Config for the basic decoding: --beam 30.0 --max-active 5000 --acoustic-scales "0.7 0.8 0.9"
  for lm_suffix in tgpr tg; do
    steps/decode_ctc_lat.sh --cmd "$decode_cmd" --nj 10 --beam 17.0 --lattice_beam 8.0 --max-active 5000 --acwt 0.9 \
      data/lang_phn_test_${lm_suffix} data/test_dev93 $dir/decode_dev93_${lm_suffix} || exit 1;
    steps/decode_ctc_lat.sh --cmd "$decode_cmd" --nj 8 --beam 17.0 --lattice_beam 8.0 --max-active 5000 --acwt 0.9 \
      data/lang_phn_test_${lm_suffix} data/test_eval92 $dir/decode_eval92_${lm_suffix} || exit 1;
  done
fi