forked from Rayhane-mamah/Tacotron-2
-
Notifications
You must be signed in to change notification settings - Fork 9
/
test_wavenet_feeder.py
113 lines (90 loc) · 3.47 KB
/
test_wavenet_feeder.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
import numpy as np
import os
import argparse
from hparams import hparams
from datasets import audio
from tqdm import tqdm
def _limit_time(hparams):
'''Limit time resolution to save GPU memory.
'''
if hparams.max_time_sec is not None:
return int(hparams.max_time_sec * hparams.sample_rate)
elif hparams.max_time_steps is not None:
return hparams.max_time_steps
else:
return None
def get_groups(args, hparams, meta, local_condition):
if hparams.train_with_GTA:
mel_file = meta[2]
else:
mel_file = meta[1]
audio_file = meta[0]
input_data = np.load(os.path.join(args.base_dir, audio_file))
if local_condition:
local_condition_features = np.load(os.path.join(args.base_dir, mel_file))
else:
local_condition_features = None
return (input_data, local_condition_features, None, len(input_data))
def _adjust_time_resolution(hparams, batch, local_condition, max_time_steps):
'''Adjust time resolution between audio and local condition
'''
if local_condition:
new_batch = []
for b in batch:
x, c, g, l = b
_assert_ready_for_upsample(hparams, x, c)
if max_time_steps is not None:
max_steps = _ensure_divisible(max_time_steps, audio.get_hop_size(hparams), True)
if len(x) > max_time_steps:
max_time_frames = max_steps // audio.get_hop_size(hparams)
start = np.random.randint(0, len(c) - max_time_frames)
time_start = start * audio.get_hop_size(hparams)
x = x[time_start: time_start + max_time_frames * audio.get_hop_size(hparams)]
c = c[start: start + max_time_frames, :]
_assert_ready_for_upsample(hparams, x, c)
new_batch.append((x, c, g, l))
return new_batch
else:
new_batch = []
for b in batch:
x, c, g, l = b
x = audio.trim_silence(x, hparams)
if max_time_steps is not None and len(x) > max_time_steps:
start = np.random.randint(0, len(c) - max_time_steps)
x = x[start: start + max_time_steps]
new_batch.append((x, c, g, l))
return new_batch
def _assert_ready_for_upsample(hparams, x, c):
assert len(x) % len(c) == 0 and len(x) // len(c) == audio.get_hop_size(hparams)
def check_time_alignment(hparams, batch, local_condition):
#No need to check beyond this step when preparing data
#Limit time steps to save GPU Memory usage
max_time_steps = _limit_time(hparams)
#Adjust time resolution for upsampling
batch = _adjust_time_resolution(hparams, batch, local_condition, max_time_steps)
def _ensure_divisible(length, divisible_by=256, lower=True):
if length % divisible_by == 0:
return length
if lower:
return length - length % divisible_by
else:
return length + (divisible_by - length % divisible_by)
def run(args, hparams):
with open(args.metadata, 'r') as file:
metadata = [line.strip().split('|') for line in file]
local_condition = hparams.cin_channels > 0
examples = [get_groups(args, hparams, meta, local_condition) for meta in metadata]
batches = [examples[i: i+hparams.wavenet_batch_size] for i in range(0, len(examples), hparams.wavenet_batch_size)]
for batch in tqdm(batches):
check_time_alignment(hparams, batch, local_condition)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--base_dir', default='')
parser.add_argument('--hparams', default='',
help='Hyperparameter overrides as a comma-separated list of name=value pairs')
parser.add_argument('--metadata', default='tacotron_output/gta/map.txt')
args = parser.parse_args()
modified_hparams = hparams.parse(args.hparams)
run(args, modified_hparams)
if __name__ == '__main__':
main()