-
Notifications
You must be signed in to change notification settings - Fork 9
/
generate_Oct10_FORMATTED.py
179 lines (150 loc) · 5.73 KB
/
generate_Oct10_FORMATTED.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
167
168
169
170
171
172
173
174
175
176
177
178
179
from __future__ import division
from __future__ import print_function
import math
import argparse
from datetime import datetime
import json
import os
import numpy as np
import tensorflow as tf
from wavenet import WaveNetModel, text_reader
SAMPLES = 16000
LOGDIR = './logdir'
WINDOW = 8000
WAVENET_PARAMS = './wavenet_params.json'
SAVE_EVERY = None
def get_arguments():
def _str_to_bool(s):
"""Convert string to bool (in argparse context)."""
if s.lower() not in ['true', 'false']:
raise ValueError('Argument needs to be a '
'boolean, got {}'.format(s))
return {'true': True, 'false': False}[s.lower()]
parser = argparse.ArgumentParser(description='WaveNet generation script')
parser.add_argument(
'checkpoint', type=str, help='Which model checkpoint to generate from')
parser.add_argument(
'--samples',
type=int,
default=SAMPLES,
help='How many waveform samples to generate')
parser.add_argument(
'--logdir',
type=str,
default=LOGDIR,
help='Directory in which to store the logging '
'information for TensorBoard.')
parser.add_argument(
'--window',
type=int,
default=WINDOW,
help='The number of past samples to take into '
'account at each step')
parser.add_argument(
'--wavenet_params',
type=str,
default=WAVENET_PARAMS,
help='JSON file with the network parameters')
parser.add_argument(
'--text_out_path',
type=str,
default=None,
help='Path to output txt file')
parser.add_argument(
'--save_every',
type=int,
default=SAVE_EVERY,
help='How many samples before saving in-progress wav')
parser.add_argument(
'--fast_generation',
type=_str_to_bool,
default=True,
help='Use fast generation')
return parser.parse_args()
def write_text(waveform, filename):
text = waveform
y = []
for index, item in enumerate(text):
y.append(chr(text[index]))
print('Prediction is: ', ''.join(str(e) for e in y))
y = np.array(y)
np.savetxt(filename, y.reshape(1, y.shape[0]), delimiter="", newline="\n", fmt="%s")
print('\n\nUpdated text file at {}'.format(filename))
print('\n**********\n\n')
def main(checkpoint=None):
args = get_arguments()
started_datestring = "{0:%Y-%m-%dT%H-%M-%S}".format(datetime.now())
logdir = os.path.join(args.logdir, 'generate', started_datestring)
with open(args.wavenet_params, 'r') as config_file:
wavenet_params = json.load(config_file)
sess = tf.Session()
net = WaveNetModel(
batch_size=1,
dilations=wavenet_params['dilations'],
filter_width=wavenet_params['filter_width'],
residual_channels=wavenet_params['residual_channels'],
dilation_channels=wavenet_params['dilation_channels'],
quantization_channels=wavenet_params['quantization_channels'],
skip_channels=wavenet_params['skip_channels'],
use_biases=wavenet_params['use_biases'])
samples = tf.placeholder(tf.int32)
if args.fast_generation:
next_sample = net.predict_proba_incremental(samples)
else:
next_sample = net.predict_proba(samples)
if args.fast_generation:
sess.run(tf.initialize_all_variables())
sess.run(net.init_ops)
variables_to_restore = {
var.name[:-2]: var for var in tf.all_variables()
if not ('state_buffer' in var.name or 'pointer' in var.name)}
saver = tf.train.Saver(variables_to_restore)
if checkpoint==None:
print('Restoring model from {}'.format(args.checkpoint))
saver.restore(sess, args.checkpoint)
else:
print('Restoring model from PARAMETER {}'.format(checkpoint))
saver.restore(sess, args.checkpoint)
decode = samples
quantization_channels = wavenet_params['quantization_channels']
waveform = [32.]
last_sample_timestamp = datetime.now()
for step in range(args.samples):
if args.fast_generation:
outputs = [next_sample]
outputs.extend(net.push_ops)
window = waveform[-1]
else:
if len(waveform) > args.window:
window = waveform[-args.window:]
else:
window = waveform
outputs = [next_sample]
# Run the WaveNet to predict the next sample.
prediction = sess.run(outputs, feed_dict={samples: window})[0]
sample = np.random.choice(
np.arange(quantization_channels), p=prediction)
waveform.append(sample)
# Show progress only once per second.
current_sample_timestamp = datetime.now()
time_since_print = current_sample_timestamp - last_sample_timestamp
if time_since_print.total_seconds() > 1.:
print('Sample {:3<d}/{:3<d}'.format(step + 1, args.samples),
end='\r')
last_sample_timestamp = current_sample_timestamp
if args.text_out_path == None:
args.text_out_path="GENERATED/"+args.checkpoint.split("/")[-2]+"_Model_"+args.checkpoint.split("-")[-1]+".txt"
# If we have partial writing, save the result so far.
if (args.text_out_path and args.save_every and
(step + 1) % args.save_every == 0):
out = sess.run(decode, feed_dict={samples: waveform})
write_text(out, args.text_out_path)
# Introduce a newline to clear the carriage return from the progress.
print()
# Save the result as a wav file.
if args.text_out_path:
out = sess.run(decode, feed_dict={samples: waveform})
write_text(out, args.text_out_path)
print('Finished generating.')
if __name__ == '__main__':
main()