/
inference.py
281 lines (223 loc) · 8.7 KB
/
inference.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
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
import glob
import os.path
from pprint import pprint
import warnings
import numpy as np
import pytorch_lightning as pl
import torch
from core import LightningSystem
from utils.config import read_nested_yaml
torch.set_float32_matmul_precision("medium")
import torchaudio as ta
from typing import List, Optional, Dict, Union
from tqdm import tqdm
def inference(
ckpt_path: str,
file_path: str,
model_name: str,
output_dir: Optional[str] = None,
include_track_name: Optional[bool] = None,
get_residual: bool = True,
get_no_vox_combinations: bool = True,
channel_filter: Optional[Union[int, List[int]]] = None,
) -> None:
if os.path.isdir(ckpt_path):
ckpt_dir = ckpt_path
if os.path.exists(os.path.join(ckpt_path, "checkpoints", "last.ckpt")):
ckpt = os.path.join(ckpt_path, "checkpoints", "last.ckpt")
else:
ckpts = glob.glob(os.path.join(ckpt_path, "checkpoints", "*.ckpt"))
epochs = [int(os.path.basename(c).split("-")[0].split("=")[-1]) for
c in ckpts]
max_epochs = np.argmax(epochs)
ckpt = ckpts[max_epochs]
else:
ckpt_dir = os.path.dirname(os.path.dirname(ckpt_path))
ckpt = ckpt_path
config: Dict = read_nested_yaml(os.path.join(ckpt_dir, "hparams.yaml"))
pprint(config)
pl.seed_everything(seed=config["seed"], workers=True)
if output_dir is None:
output_dir = os.path.join(
os.path.dirname(file_path),
"separated",
model_name
)
cuda_available = torch.cuda.is_available()
if cuda_available:
map_location = None
else:
map_location = torch.device('cpu')
model: LightningSystem = LightningSystem.load_from_checkpoint(
ckpt,
config=config["system"],
map_location=map_location
)
model.set_predict_output_path(output_dir)
# restore the fader to intended value
model.fader.__init__( # type: ignore[misc]
**config["system"]["inference"]["fader"]["kwargs"]
)
model.fader.to(model.device) # type: ignore[arg-type]
model.eval()
audio, fs = ta.load(file_path)
if fs != config["system"]["model"]["kwargs"]["fs"]:
audio = ta.functional.resample(
audio,
fs,
config["system"]["model"]["kwargs"]["fs"]
)
fs = config["system"]["model"]["kwargs"]["fs"]
track_name = os.path.basename(file_path).split(".")[0]
track = [track_name]
treat_batch_as_channels = False
if channel_filter is not None:
if isinstance(channel_filter, int):
channel_filter = [channel_filter]
audio = audio[channel_filter, :]
in_channel_audio = audio.shape[0]
in_channel_model = config["system"]["model"]["kwargs"]["in_channel"]
if in_channel_audio != in_channel_model:
if in_channel_audio == 1 and in_channel_model > 1:
audio = audio.repeat(in_channel_model, 1)
elif in_channel_audio > 1 and in_channel_model == 1:
audio = audio[:, None, :]
treat_batch_as_channels = True
track = [track_name + f"_{i}" for i in range(in_channel_audio)]
else:
raise ValueError(
f"Cannot handle in_channel_audio={in_channel_audio} "
f"and in_channel_model={in_channel_model}"
)
if in_channel_audio == 1 and in_channel_model == 1:
audio = audio[None, ...]
audio = audio.to(model.device)
with torch.inference_mode():
model.predict_step(
{
"audio": {
"mixture": audio,
},
"track": track,
},
get_residual=get_residual,
get_no_vox_combinations=get_no_vox_combinations,
include_track_name=include_track_name,
treat_batch_as_channels=treat_batch_as_channels,
fs=fs,
)
def inference_multiple(
ckpt_path: str,
file_glob: str,
model_name: str,
output_dir: Optional[str],
include_track_name: Optional[bool] = False,
get_residual: bool = False,
get_no_vox_combinations: bool = False,
channel_filter: Optional[Union[int, List[int]]] = None,
) -> None:
if os.path.isdir(ckpt_path):
ckpt_dir = ckpt_path
if os.path.exists(os.path.join(ckpt_path, "checkpoints", "last.ckpt")):
ckpt = os.path.join(ckpt_path, "checkpoints", "last.ckpt")
else:
ckpts = glob.glob(os.path.join(ckpt_path, "checkpoints", "*.ckpt"))
epochs = [int(os.path.basename(c).split("-")[0].split("=")[-1]) for
c in ckpts]
max_epochs = np.argmax(epochs)
ckpt = ckpts[max_epochs]
else:
ckpt_dir = os.path.dirname(os.path.dirname(ckpt_path))
ckpt = ckpt_path
if model_name is None:
model_name = os.path.basename(ckpt_dir)
print(model_name)
config: Dict = read_nested_yaml(os.path.join(ckpt_dir, "hparams.yaml"))
pprint(config)
pl.seed_everything(seed=config["seed"], workers=True)
files = sorted(glob.glob(file_glob, recursive=True))
cuda_available = torch.cuda.is_available()
if cuda_available:
map_location = None
else:
map_location = torch.device('cpu')
model: LightningSystem = LightningSystem.load_from_checkpoint(
ckpt,
config=config["system"],
map_location=map_location
)
model.set_predict_output_path(output_dir)
# restore the fader to intended value
model.fader.__init__( # type: ignore[misc]
**config["system"]["inference"]["fader"]["kwargs"]
)
model.fader.to(model.device) # type: ignore[arg-type]
model.eval()
output_dir_ = output_dir
for file_path in tqdm(files):
if output_dir_ is None:
output_dir = os.path.join(
os.path.dirname(file_path),
"separated",
model_name
)
else:
output_dir = os.path.join(output_dir_, model_name, os.path.basename(os.path.dirname(file_path)))
os.makedirs(output_dir, exist_ok=True)
audio, fs = ta.load(file_path)
if fs != config["system"]["model"]["kwargs"]["fs"]:
audio = ta.functional.resample(
audio,
fs,
config["system"]["model"]["kwargs"]["fs"]
)
fs = config["system"]["model"]["kwargs"]["fs"]
track_name = os.path.basename(file_path).split(".")[0]
track = [track_name]
treat_batch_as_channels = False
if channel_filter is not None:
if isinstance(channel_filter, int):
channel_filter = [channel_filter]
audio = audio[channel_filter, :]
in_channel_audio = audio.shape[0]
kwargs = config["system"]["model"]["kwargs"]
in_channel_model = kwargs.get("in_channel", None)
if in_channel_model is None:
# for demucs
in_channel_model = kwargs.get("audio_channels", None)
if in_channel_model is None:
warnings.warn("in_channel_model is not specified, using 1")
in_channel_model = 1
if in_channel_audio != in_channel_model:
if in_channel_audio == 1 and in_channel_model > 1:
audio = audio.repeat(in_channel_model, 1)
elif in_channel_audio > 1 and in_channel_model == 1:
audio = audio[:, None, :]
treat_batch_as_channels = True
track = [track_name + f"_{i}" for i in range(in_channel_audio)]
else:
raise ValueError(
f"Cannot handle in_channel_audio={in_channel_audio} "
f"and in_channel_model={in_channel_model}"
)
if in_channel_audio == 1 and in_channel_model == 1:
audio = audio[None, ...]
audio = audio.to(model.device)
model.set_predict_output_path(output_dir)
with torch.inference_mode():
model.predict_step(
{
"audio": {
"mixture": audio,
},
"track": track,
},
get_residual=get_residual,
get_no_vox_combinations=get_no_vox_combinations,
include_track_name=include_track_name,
treat_batch_as_channels=treat_batch_as_channels,
fs=fs,
)
if __name__ == "__main__":
import fire
fire.Fire()