forked from BUTSpeechFIT/VBx
-
Notifications
You must be signed in to change notification settings - Fork 0
/
AMI_run_2pass.sh
executable file
·187 lines (167 loc) · 7.27 KB
/
AMI_run_2pass.sh
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
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
#!/bin/bash
INSTRUCTION=$1
METHOD=$2 # AHC or AHC+VB or AHC+GMM
exp_dir=$3 # output experiment directory
xvec_dir_base=$4 # output xvectors directory
WAV_DIR=$5 # wav files directory
FILE_LIST=$6 # txt list of files to process
LAB_DIR=$7 # lab files directory with VAD segments
RTTM_DIR=$8 # reference rttm files directory
NUM_PASS=${9:-1}
QUEUE=${10:-none}
num_iter=${11:-30}
M=${12:-7}
r=${13:-0.9}
N0_firstpass=${14:-50}
N0_secondpass=${15:-50}
k_means_only=${16:-0}
model_type=${17:-"wb"} # can be nb [narrowband] or wb [wideband]
DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" >/dev/null 2>&1 && pwd)"
# define models and configurations for each pass
if [[ $model_type == "wb" ]]; then
XVEC_PLDA_MODEL1="/expscratch/kkarra/xvec_trained_models/wb/resnet_wb_sc0.5.pt"
XVEC_PLDA_MODEL2="/expscratch/kkarra/xvec_trained_models/wb/resnet_wb_sc0.5.pt"
KALDI_FBANK_CONF="/expscratch/kkarra/xvec_trained_models/wb/fbank_16k.conf"
PASS1_SEG_JUMP=200
PASS1_SEG_LEN=200
PASS2_SEG_JUMP=25
PASS2_SEG_LEN=125
elif [[ $model_type == "nb" ]]; then
XVEC_PLDA_MODEL1="/expscratch/kkarra/xvec_trained_models/nb/tdnn_nb_2s.pt"
XVEC_PLDA_MODEL2="/expscratch/kkarra/xvec_trained_models/nb/tdnn_nb_1_25s.pt"
KALDI_FBANK_CONF="/expscratch/kkarra/xvec_trained_models/nb/fbank_8k.conf"
PASS1_SEG_JUMP=200 # corresponds to ovlp=0
PASS1_SEG_LEN=200 # corresponds to 2 sec
PASS2_SEG_JUMP=25
PASS2_SEG_LEN=125
else
echo "invalid model type: $model_type"
exit 1
fi
FEAT_EXTRACT_ENGINE=kaldi
XVECTOR_EXTRACTION_ENGINE=coe
EMBED_DIM=128
if (("$NUM_PASS" < 1)); then
echo "NUM_PASS must be >= 1"
exit 1
fi
passes=($(seq 1 1 $NUM_PASS))
# verify models/seg_jump/seg_len for each pass is defined
for pass in "${passes[@]}"; do
# model, seg_jump, and seg_len are dynamic bash variables.
# See: https://stackoverflow.com/a/65021258/1057098
model="XVEC_PLDA_MODEL$pass"
seg_jump="PASS${pass}_SEG_JUMP"
seg_len="PASS${pass}_SEG_LEN"
if [ -z ${!model} ]; then
echo "$model is undefined!"
exit 1
fi
if [ -z ${!seg_jump} ]; then
echo "$seg_jump is undefined!"
exit 1
fi
if [ -z ${!seg_len} ]; then
echo "$seg_len is undefined!"
exit 1
fi
done
if [[ $INSTRUCTION == "xvectors" ]]; then
EXTRACT_SCRIPT=$DIR/VBx/extract.sh
DEVICE=cpu
for pass in "${passes[@]}"; do
xvec_dir="$xvec_dir_base"_"$pass"
mkdir -p $xvec_dir
##################################
# model, seg_jump, and seg_len are dynamic bash variables.
# See: https://stackoverflow.com/a/65021258/1057098
model="XVEC_PLDA_MODEL$pass"
seg_jump="PASS${pass}_SEG_JUMP"
seg_len="PASS${pass}_SEG_LEN"
##################################
$EXTRACT_SCRIPT ${!model} None $WAV_DIR $LAB_DIR $FILE_LIST $xvec_dir $DEVICE \
$FEAT_EXTRACT_ENGINE $XVECTOR_EXTRACTION_ENGINE $KALDI_FBANK_CONF $EMBED_DIM \
${!seg_jump} ${!seg_len}
if [ "$QUEUE" = "none" ]; then
bash $xvec_dir/xv_task
else
nl=$(wc -l <$FILE_LIST)
qsub -cwd -l num_proc=1,mem_free=4G,h_rt=400:00:00 -q $QUEUE -t 1:$nl -sync y -o $xvec_dir/extract.log/ -e $xvec_dir/extract.err/ $xvec_dir/uge_xv_task.sh
fi
done
fi
if [[ $INSTRUCTION == "diarization" ]]; then
TASKFILE=$exp_dir/diar_"$METHOD"_task
UGE_TASKFILE=$exp_dir/diar_"$METHOD"_uge_task
OUTFILE=$exp_dir/diar_"$METHOD"_out
rm -f $TASKFILE $OUTFILE $UGE_TASKFILE
mkdir -p $exp_dir/lists
echo "#!/bin/bash" >>$UGE_TASKFILE
echo ". /etc/profile.d/modules.sh" >>$UGE_TASKFILE
echo "module load cuda11.0/blas/11.0.3 cuda11.0/toolkit/11.0.3 cudnn/8.0.2_cuda11.0"
printf "flist=(" >>$UGE_TASKFILE
thr=-0.015
tareng=0.3
smooth=7.0
Fa=0.4
Fb=64
loopP=0.65
OUT_DIR=$exp_dir/out_dir_"$METHOD"
if [[ ! -d $OUT_DIR ]]; then
mkdir -p $OUT_DIR
while IFS= read -r line; do
grep $line $FILE_LIST >$exp_dir/lists/$line".txt"
#echo "python $DIR/VBx/vbhmm.py --init $METHOD --out-rttm-dir $OUT_DIR/rttms --xvec-ark-file $xvec_dir/xvectors/$line.ark --segments-file $xvec_dir/segments/$line --plda-file $XVEC_PLDA_MODEL --plda-format pytorch --threshold $thr --target-energy $tareng --init-smoothing $smooth --Fa $Fa --Fb $Fb --loopP $loopP" >>$TASKFILE
cmd_str="python $DIR/VBx/vbhmm.py --init $METHOD --out-rttm-dir $OUT_DIR/rttms --xvec-ark-file"
xvec_inputarg=""
for pass in "${passes[@]}"; do
xvec_dir="$xvec_dir_base"_"$pass"
xvec_inputarg="$xvec_inputarg ${xvec_dir}/xvectors/${line}.ark"
done
cmd_str="$cmd_str $xvec_inputarg --segments-file"
segment_inputarg=""
for pass in "${passes[@]}"; do
xvec_dir="$xvec_dir_base"_"$pass"
segment_inputarg="$segment_inputarg ${xvec_dir}/segments/${line}"
done
cmd_str="$cmd_str $segment_inputarg --plda-file"
plda_inputarg=""
for pass in "${passes[@]}"; do
plda_pass_model="XVEC_PLDA_MODEL$pass"
plda_inputarg="$plda_inputarg ${!plda_pass_model}"
done
cmd_str="$cmd_str $plda_inputarg --plda-format pytorch --threshold $thr --target-energy $tareng --init-smoothing $smooth --Fa $Fa --Fb $Fb --loopP $loopP --num-iter $num_iter --M $M --r $r --N0 $N0_firstpass $N0_secondpass --kmeans-only $k_means_only"
echo $cmd_str >> $TASKFILE
printf "$line " >>$UGE_TASKFILE
done <$FILE_LIST
printf ")\n\n" >>$UGE_TASKFILE
if [ "$QUEUE" = "none" ]; then
bash $TASKFILE >$OUTFILE
else
# run it on the grid
# TODO: add GPU support
xvec_inputarg=""
segment_inputarg=""
plda_inputarg=""
for pass in "${passes[@]}"; do
xvec_dir="$xvec_dir_base"_"$pass"
xvec_inputarg="$xvec_inputarg ${xvec_dir}/xvectors/\${file_uge}.ark"
segment_inputarg="$segment_inputarg ${xvec_dir}/segments/\${file_uge}"
plda_model="XVEC_PLDA_MODEL$pass"
plda_inputarg="$plda_inputarg ${!plda_model}"
done
echo "file_uge=\${flist[\$((\${SGE_TASK_ID}-1))]}" >>$UGE_TASKFILE
#echo "python $DIR/VBx/vbhmm.py --init $METHOD --out-rttm-dir $OUT_DIR/rttms --xvec-ark-file $xvec_dir/xvectors/\${file_uge}.ark --segments-file $xvec_dir/segments/\${file_uge} --plda-file $XVEC_PLDA_MODEL --plda-format pytorch --threshold $thr --target-energy $tareng --init-smoothing $smooth --Fa $Fa --Fb $Fb --loopP $loopP" >>$UGE_TASKFILE
cmd_str="python $DIR/VBx/vbhmm.py --init $METHOD --out-rttm-dir $OUT_DIR/rttms --xvec-ark-file $xvec_inputarg --segments-file $segment_inputarg --plda-file $plda_inputarg --plda-format pytorch --threshold $thr --target-energy $tareng --init-smoothing $smooth --Fa $Fa --Fb $Fb --loopP $loopP --num-iter $num_iter --M $M --r $r --N0 $N0_firstpass $N0_secondpass --kmeans-only $k_means_only"
echo $cmd_str >> $UGE_TASKFILE
nl=$(wc -l <$FILE_LIST)
qsub -cwd -l num_proc=1,mem_free=4G,h_rt=400:00:00 -N vbxhmm -q $QUEUE -t 1:$nl -sync y -o $OUT_DIR/vbhmm.log/ -e $OUT_DIR/vbhmm.err/ $UGE_TASKFILE
fi
fi
## Score
cat $OUT_DIR/rttms/*.rttm >$OUT_DIR/sys.rttm
cat $RTTM_DIR/*.rttm > $OUT_DIR/ref.rttm
$DIR/dscore/score.py --collar 0.25 --ignore_overlaps -r $OUT_DIR/ref.rttm -s $OUT_DIR/sys.rttm > $OUT_DIR/result_forgiving
$DIR/dscore/score.py --collar 0.25 -r $OUT_DIR/ref.rttm -s $OUT_DIR/sys.rttm >$OUT_DIR/result_fair
$DIR/dscore/score.py --collar 0.0 -r $OUT_DIR/ref.rttm -s $OUT_DIR/sys.rttm >$OUT_DIR/result_full
fi