run_autoencoder.sh
2.72 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
#!/bin/bash
# this is an example to show a "tdnn" system in raw nnet configuration
# i.e. without a transition model
# It uses corrupted (reverberation + noise) speech as input and clean speech
# as output.
. ./cmd.sh
stage=0
affix=
train_stage=-10
common_egs_dir=
egs_opts=
num_data_reps=10
remove_egs=true
. ./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
dir=exp/nnet3/tdnn_raw
dir=$dir${affix:+_$affix}
clean_data_dir=data/train
data_dir=data/train_rvb
targets_scp=$dir/targets.scp
mkdir -p $dir
# Create copies of clean feats with prefix "rev$x_" to match utterance names of
# the noisy feats
for x in `seq 1 $num_data_reps`; do
awk -v x=$x '{print "rev"x"_"$0}' $clean_data_dir/feats.scp | sort -k1,1 > $targets_scp
done
if [ $stage -le 9 ]; then
echo "$0: creating neural net configs";
num_targets=`feat-to-dim scp:$targets_scp - 2>/dev/null` || exit 1
feat_dim=`feat-to-dim scp:$data_dir/feats.scp - 2>/dev/null` || exit 1
mkdir -p $dir/configs
cat <<EOF > $dir/configs/network.xconfig
input dim=$feat_dim name=input
relu-renorm-layer name=tdnn1 dim=1024 input=Append(-2,-1,0,1,2)
relu-renorm-layer name=tdnn2 dim=1024 input=Append(-1,2)
relu-renorm-layer name=tdnn3 dim=1024 input=Append(-3,3)
relu-renorm-layer name=tdnn4 dim=1024 input=Append(-7,2)
relu-renorm-layer name=tdnn5 dim=1024
output-layer name=output dim=$num_targets max-change=1.5 objective-type=quadratic include-log-softmax=false
EOF
steps/nnet3/xconfig_to_configs.py --xconfig-file $dir/configs/network.xconfig --config-dir $dir/configs/
fi
if [ $stage -le 10 ]; then
if [[ $(hostname -f) == *.clsp.jhu.edu ]] && [ ! -d $dir/egs/storage ]; then
utils/create_split_dir.pl \
/export/b0{3,4,5,6}/$USER/kaldi-data/egs/aspire-$(date +'%m_%d_%H_%M')/s5/$dir/egs/storage $dir/egs/storage
fi
steps/nnet3/train_raw_dnn.py --stage=$train_stage \
--cmd="$decode_cmd" \
--feat.cmvn-opts "--norm-means=false --norm-vars=false" \
--trainer.num-epochs 2 \
--trainer.optimization.num-jobs-initial 3 \
--trainer.optimization.num-jobs-final 16 \
--trainer.optimization.initial-effective-lrate 0.0017 \
--trainer.optimization.final-effective-lrate 0.00017 \
--trainer.optimization.minibatch-size 512 \
--egs.dir "$common_egs_dir" --egs.opts "$egs_opts" \
--cleanup.remove-egs $remove_egs \
--cleanup.preserve-model-interval 50 \
--nj=30 \
--use-dense-targets=true \
--feat-dir=${data_dir} \
--targets-scp=$targets_scp \
--dir=$dir || exit 1;
fi