-
Notifications
You must be signed in to change notification settings - Fork 12
/
synthesize.py
executable file
·166 lines (137 loc) · 5.79 KB
/
synthesize.py
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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# Synthesize waveform using converted features.
# By Wen-Chin Huang 2019.06
import json
import os
import tensorflow as tf
import numpy as np
from datetime import datetime
from importlib import import_module
import pysptk
import pyworld as pw
from scipy.io import loadmat, savemat
from scipy.io import wavfile
import argparse
import logging
import multiprocessing as mp
import sys
from preprocessing.vcc2018.feature_reader import Whole_feature_reader
from util.synthesizer import world_synthesis
from util.misc import read_hdf5
from util.postfilter import fast_MLGV
from util.f0transformation import log_linear_transformation
def read_and_synthesize(file_list, arch, stats, input_feat, output_feat):
for i, (bin_path, feat_path) in enumerate(file_list):
input_feat_dim = arch['feat_param']['dim'][input_feat]
# define paths
output_dir = os.path.dirname(bin_path).replace('converted-' + output_feat, 'converted-wav')
basename = os.path.splitext(os.path.split(bin_path)[-1])[0]
wav_name = os.path.join(output_dir, basename + '.wav')
gv_wav_name = os.path.join(output_dir, basename + '-gv.wav')
# read source features and converted spectral features
src_data = Whole_feature_reader(feat_path, arch['feat_param'])
cvt = np.fromfile(bin_path, dtype = np.float32).reshape([-1, input_feat_dim])
# f0 conversion
lf0 = log_linear_transformation(src_data['f0'], stats)
# apply gv post filtering to converted
cvt_gv = fast_MLGV(cvt, stats['gv_t'])
# energy compensation
if output_feat == 'mcc':
en_cvt = np.c_[src_data['en_mcc'], cvt]
en_cvt_gv = np.c_[src_data['en_mcc'], cvt_gv]
elif output_feat == 'sp':
cvt = np.power(10., cvt)
en_cvt = np.expand_dims(src_data['en_sp'], 1) * cvt
cvt_gv = np.power(10., cvt_gv)
en_cvt_gv = np.expand_dims(src_data['en_sp'], 1) * cvt_gv
# synthesis
world_synthesis(wav_name, arch['feat_param'],
lf0.astype(np.float64).copy(order='C'),
src_data['ap'].astype(np.float64).copy(order='C'),
en_cvt.astype(np.float64).copy(order='C'),
output_feat)
world_synthesis(gv_wav_name, arch['feat_param'],
lf0.astype(np.float64).copy(order='C'),
src_data['ap'].astype(np.float64).copy(order='C'),
en_cvt_gv.astype(np.float64).copy(order='C'),
output_feat)
def main():
parser = argparse.ArgumentParser(
description="synthesize waveforms using converted files.")
parser.add_argument(
"--logdir", required=True, type=str,
help="path of log directory")
parser.add_argument(
"--type", default='test', type=str,
help="test or valid (default is test)")
parser.add_argument(
"--input_feat", required=True,
type=str, help="input feature type")
parser.add_argument(
"--output_feat", required=True,
type=str, help="output feature type")
parser.add_argument(
"--n_jobs", default=12,
type=int, help="number of parallel jobs")
args = parser.parse_args()
# set log level
fmt = '%(asctime)s %(message)s'
datefmt = '%m/%d/%Y %I:%M:%S'
logFormatter = logging.Formatter(fmt, datefmt=datefmt)
logging.basicConfig(
level=logging.INFO,
filename=os.path.join(args.logdir, 'exp.log'),
format=fmt,
datefmt=datefmt,
)
consoleHandler = logging.StreamHandler()
consoleHandler.setFormatter(logFormatter)
logging.getLogger().addHandler(consoleHandler)
logging.info('====================')
logging.info('Synthesize start')
logging.info(args)
train_dir = os.sep.join(os.path.normpath(args.logdir).split(os.sep)[:-1])
output_dir = os.path.basename(os.path.normpath(args.logdir))
src, trg = output_dir.split('-')[-2:]
# Load architecture
arch = tf.gfile.Glob(os.path.join(train_dir, 'architecture*.json'))[0] # should only be 1 file
with open(arch) as fp:
arch = json.load(fp)
input_feat = args.input_feat
output_feat = args.output_feat
# Load statistics
stats = {
'mu_s' : read_hdf5(arch['stats'], '/f0/' + src + '/mean'),
'std_s' : read_hdf5(arch['stats'], '/f0/' + src + '/std'),
'mu_t' : read_hdf5(arch['stats'], '/f0/' + trg + '/mean'),
'std_t' : read_hdf5(arch['stats'], '/f0/' + trg + '/std'),
'gv_t' : read_hdf5(arch['stats'], '/gv_{}/'.format(output_feat) + trg),
}
# Make directory
tf.gfile.MakeDirs(os.path.join(args.logdir, 'converted-wav'))
# Get and divide list
bin_list = sorted(tf.gfile.Glob(os.path.join(args.logdir, 'converted-{}'.format(output_feat), '*.bin')))
if args.type == 'test':
feat_list = sorted(tf.gfile.Glob(arch['conversion']['test_file_pattern'].format(src)))
elif args.type == 'valid':
feat_list = []
for p in arch['training']['valid_file_pattern']:
feat_list.extend(tf.gfile.Glob(p.replace('*', src)))
feat_list = sorted(feat_list)
assert(len(bin_list) == len(feat_list))
file_list = list(zip(bin_list, feat_list))
logging.info("number of utterances = %d" % len(file_list))
file_lists = np.array_split(file_list, args.n_jobs)
file_lists = [f_list.tolist() for f_list in file_lists]
# multi processing
processes = []
for f in file_lists:
p = mp.Process(target=read_and_synthesize, args=(f, arch, stats, input_feat, output_feat))
p.start()
processes.append(p)
# wait for all process
for p in processes:
p.join()
if __name__ == '__main__':
main()