#!/bin/bash
  
  # This is as run_sgmm.sh but using the "sgmm2" code, which uses "state-clustered tied mixtures"
  # and the symmetric SGMM, and one or two other small changes (e.g. no updating of M for a few
  # iterations.)
  
  . ./cmd.sh
  
  # Begin configuration section.
  nj=8
  # End configuration section.
  
  echo "$0 $@"  # Print the command line for logging
  
  . parse_options.sh || exit 1;
  
  ## SGMM on top of LDA+MLLT+SAT features.
  if [ ! -f exp/ubm4a/final.mdl ]; then
    steps/train_ubm.sh --silence-weight 0.5 --cmd "$train_cmd" 400 data/train data/lang exp/tri3b_ali exp/ubm4a || exit 1;
  fi
  steps/train_sgmm2.sh  --cmd "$train_cmd" 8000 19000 data/train data/lang exp/tri3b_ali exp/ubm4a/final.ubm exp/sgmm2_4a || exit 1;
  
  utils/mkgraph.sh data/lang_test exp/sgmm2_4a exp/sgmm2_4a/graph || exit 1;
  
  steps/decode_sgmm2.sh --config conf/decode.config --nj $nj --cmd "$decode_cmd" \
    --transform-dir exp/tri3b/decode  exp/sgmm2_4a/graph data/test exp/sgmm2_4a/decode || exit 1;
  
  steps/decode_sgmm2.sh --use-fmllr true --config conf/decode.config --nj $nj --cmd "$decode_cmd" \
    --transform-dir exp/tri3b/decode  exp/sgmm2_4a/graph data/test exp/sgmm2_4a/decode_fmllr || exit 1;
  
   # Now we'll align the SGMM system to prepare for discriminative training.
   steps/align_sgmm2.sh --nj $nj --cmd "$train_cmd" --transform-dir exp/tri3b \
      --use-graphs true --use-gselect true data/train data/lang exp/sgmm2_4a exp/sgmm2_4a_ali || exit 1;
   steps/make_denlats_sgmm2.sh --nj $nj --sub-split 20 --cmd "$decode_cmd" --transform-dir exp/tri3b \
     data/train data/lang exp/sgmm2_4a_ali exp/sgmm2_4a_denlats
   steps/train_mmi_sgmm2.sh --cmd "$decode_cmd" --transform-dir exp/tri3b --boost 0.15 \
     data/train data/lang exp/sgmm2_4a_ali exp/sgmm2_4a_denlats exp/sgmm2_4a_mmi_b0.15
  
   for iter in 1 2 3 4; do
    steps/decode_sgmm2_rescore.sh --cmd "$decode_cmd" --iter $iter \
      --transform-dir exp/tri3b/decode data/lang data/test exp/sgmm2_4a/decode exp/sgmm2_4a_mmi_b0.15/decode_it$iter &
   done
  (
   steps/train_mmi_sgmm2.sh --cmd "$decode_cmd" --transform-dir exp/tri3b --boost 0.15 --drop-frames true \
     data/train data/lang exp/sgmm2_4a_ali exp/sgmm2_4a_denlats exp/sgmm2_4a_mmi_b0.15_x
  
   for iter in 1 2 3 4; do
    steps/decode_sgmm2_rescore.sh --cmd "$decode_cmd" --iter $iter \
      --transform-dir exp/tri3b/decode data/lang data/test exp/sgmm2_4a/decode exp/sgmm2_4a_mmi_b0.15_x/decode_it$iter &
   done
  )
  wait
  steps/decode_combine.sh data/test data/lang exp/tri1/decode exp/tri2a/decode exp/combine_1_2a/decode || exit 1;
  steps/decode_combine.sh data/test data/lang exp/sgmm2_4a/decode exp/tri3b_mmi/decode exp/combine_sgmm2_4a_3b/decode || exit 1;
  # combining the sgmm run and the best MMI+fMMI run.
  steps/decode_combine.sh data/test data/lang exp/sgmm2_4a/decode exp/tri3b_fmmi_c/decode_it5 exp/combine_sgmm2_4a_3b_fmmic5/decode || exit 1;
  
  steps/decode_combine.sh data/test data/lang exp/sgmm2_4a_mmi_b0.15/decode_it4 exp/tri3b_fmmi_c/decode_it5 exp/combine_sgmm2_4a_mmi_3b_fmmic5/decode || exit 1;