forked from aishoot/LSTM_PIT_Speech_Separation
-
Notifications
You must be signed in to change notification settings - Fork 0
/
run.sh
117 lines (100 loc) · 5.23 KB
/
run.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
#! /bin/bash
# Author:
# Sining Sun (Northwestern Polytechnical University, China)
# Chao Peng, EECS, Peking University, Beijing, China.
# This recipe is used to do NN-PIT (LSTM, DNN or BLSTM)
# Program timer with bash
time_program(){
echo `date --date='0 days ago' "+%Y-%m-%d %H:%M:%S"`
}
echo "The program starts at `time_program`."
step=0
speaker_num="2speaker"
lists_dir=storage/lists # lists_dir is used to store some necessary files lists
tfrecords_dir=storage/tfrecords
mkdir -p $lists_dir
wav_dir=～Dataset/WSJ0-mix/mix/data/2speakers_0dB/wav8k/min
# audio and stft configs
sample_rate=8000
sample_rate_name=`expr $sample_rate / 1000`KHz
window_size=256
window_shift=128
# net config
rnn_num_layers=3
rnn_size=496
model_type=LSTM
input_size=129
output_size=129
keep_prob=0.8
learning_rate=0.0005
halving_factor=0.7
gpu_id='0'
TF_CPP_MIN_LOG_LEVEL=1
# train opts
prefix=PIT
assignment=def
name=${prefix}_${model_type}_${rnn_num_layers}_${rnn_size}_${speaker_num}_${sample_rate_name}_"0dB"
TF_save_dir=storage/TFCheckpoint/$name
separated_dir=storage/separated/${name}_${assignment}
resume_training=true
# note: we want to use gender information, but we didn't use in this version.
# but when we prepared our data, we stored the gender information (maybe useful in the future).
# wsj-train-spkrinfo.txt: https://catalog.ldc.upenn.edu/docs/LDC93S6A/wsj0-train-spkrinfo.txt
#####################################################################################################
# NOTE for STEP 0: ###
# 1. Extract STFT features, and convert them to the format of tfrecords ###
#####################################################################################################
if [ $step -le 0 ]; then
echo -e "\nGenerate wav file to .lst file at `time_program`."
for x in tr cv tt; do
python local/make_wav_list.py --wav_dir ${wav_dir}/$x/mix --output_lst ${lists_dir}/${x}_wav.lst
done
# tfrecords are stored in data/tfrecords/{tr, cv, tt}_tfrecord/
echo -e "\nfrom wav file generate {tr, cv, tt}_tfrecord/*.tfrecords at `time_program`."
for x in tr cv tt; do
python -u local/gen_tfrecords.py --gender_list local/wsj0-train-spkrinfo.txt \
--window_size $window_size --window_shift $window_shift --sample_rate $sample_rate \
${wav_dir}/$x/ ${lists_dir}/${x}_wav.lst ${tfrecords_dir}/${x}_tfrecord &
done
wait
# Here, we have made tfrecords list file for tr, cv and tt data,
# Make sure you have generated tfrecords files in $tfrecords_dir/{tr, cv, tt}_tfrecord/
# The list files name must be tr_tf.lst, cv_tf.lst and tt_tf.lst.
echo -e "\nFrom {tr, cv, tt}_tfrecord/*.tfrecords generate {tr, cv, tt}_tf.lst at `time_program`."
for x in tr tt cv; do
find $tfrecords_dir/${x}_tfrecord/ -iname "*.tfrecords" > $lists_dir/${x}_tf.lst
# storage/tfrecords/tr_tfrecord/aa.tfrecords
# storage/tfrecords/tr_tfrecord/bb.tfrecords
done
fi
#####################################################################################################
# NOTE for STEP 1: Train ###
# 1. Make sure that you configure the RNN/data_dir/model_dir/ all rights ###
#####################################################################################################
if [ $step -le 1 ]; then
echo -e "\nStart Traing RNN(LSTM or BLSTM) model at `time_program`.\n"
decode=0
batch_size=25
tr_cmd="python -u run_lstm.py --lists_dir=$lists_dir --rnn_num_layers=$rnn_num_layers --batch_size=$batch_size \
--rnn_size=$rnn_size --decode=$decode --learning_rate=$learning_rate --TF_save_dir=$TF_save_dir \
--separated_dir=$separated_dir --keep_prob=$keep_prob --input_size=$input_size --output_size=$output_size \
--assign=$assignment --resume_training=$resume_training --model_type=$model_type --halving_factor=$halving_factor"
#echo $tr_cmd
CUDA_VISIBLE_DEVICES=$gpu_id TF_CPP_MIN_LOG_LEVEL=$TF_CPP_MIN_LOG_LEVEL $tr_cmd
fi
#####################################################################################################
# NOTE for STEP 2: Decode ###
# 1. Make sure that you configure the RNN/data_dir/model_dir/ all rights ###
#####################################################################################################
if [ $step -le 2 ]; then
echo -e "\nStart Decoding at `time_program`.\n"
decode=1
batch_size=30
tr_cmd="python -u run_lstm.py --lists_dir=$lists_dir --rnn_num_layers=$rnn_num_layers --batch_size=$batch_size \
--decode=$decode --learning_rate=$learning_rate --TF_save_dir=$TF_save_dir --separated_dir=$separated_dir --keep_prob=$keep_prob \
--input_size=$input_size --output_size=$output_size --assign=$assignment --resume_training=$resume_training --rnn_size=$rnn_size \
--model_type=$model_type --czt_dim=128 --window_size=$window_size --window_shift=$window_shift --sample_rate=$sample_rate"
#echo $tr_cmd
CUDA_VISIBLE_DEVICES=$gpu_id TF_CPP_MIN_LOG_LEVEL=$TF_CPP_MIN_LOG_LEVEL $tr_cmd
fi
echo -e "\nAll programs finish running at `time_program`."