-
Notifications
You must be signed in to change notification settings - Fork 0
/
Load_nsynth_data_for_FCAC.py
239 lines (180 loc) · 8.36 KB
/
Load_nsynth_data_for_FCAC.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
"""
-------------------------------File info-------------------------
% - File name: Load_nsynth_data_for_FCAC.py
% - Description: Load audio samples from the nsynth dataset and convert them online to time-frequency
% - features according to the problem setting of FCAC
% -
% - Input: 1)path to audio file folder 2) path to Nsynth_meta_for_FCAC folder
% - Output: Fbank feature of each audio sample
% - Calls:
% - usage:
% - Version： V1.0
% - Last update: 2022-11-05
% Copyright (C) PRMI, South China university of technology; 2022
% ------For Educational and Academic Purposes Only ------
% - Author : Chester.Wei.Xie, PRMI, SCUT/ GXU
% - Contact: chester.w.xie@gmail.com
------------------------------------------------------------------
"""
import argparse
import os
import numpy as np
import random
import pickle
import torch
from torch.utils.data import Dataset
from collections import defaultdict
import torchaudio
import pandas as pd
import json
def build_label_index(label_unique_list):
label2inds = defaultdict(list)
num_labels = len(label_unique_list)
for idxs, label_unique in enumerate(label_unique_list):
for label in label_unique: #
if label not in label2inds:
label2inds[label] = []
label2inds[label].append(idxs)
return label2inds
def load_meta(file):
with open(file, 'rb') as fo:
meta = pickle.load(fo)
return meta
def wave_to_tfr(audio_path):
waveform, sr = torchaudio.load(audio_path)
waveform = waveform - waveform.mean()
fbank = torchaudio.compliance.kaldi.fbank(waveform, htk_compat=True, sample_frequency=sr, use_energy=False,
window_type='hanning', num_mel_bins=128, dither=0.0,
frame_shift=10)
fbank = fbank.view(1, fbank.shape[0], fbank.shape[1])
return fbank
class NsynthDatasets(Dataset):
def __init__(self, _args, phase=None):
self.phase = phase
self.audio_dir = _args.audiopath
self.meta_dir = os.path.join(_args.metapath, 'nsynth-' + str(_args.num_class) + '-fs-meta')
with open(os.path.join(self.meta_dir, 'nsynth-' + str(_args.num_class) + '-fs_vocab.json')) as vocab_json_file:
label_to_ix = json.load(vocab_json_file)
if self.phase == 'train':
meta_info = pd.read_csv(os.path.join(self.meta_dir, 'nsynth-' + str(_args.num_class) + '-fs_train.csv'))
elif self.phase == 'val':
meta_info = pd.read_csv(os.path.join(self.meta_dir, 'nsynth-' + str(_args.num_class) + '-fs_val.csv'))
elif self.phase == 'test':
meta_info = pd.read_csv(os.path.join(self.meta_dir, 'nsynth-' + str(_args.num_class) + '-fs_test.csv'))
else:
raise Exception('No such phase {0}, only support train, val and test'.format(phase))
self.filenames = meta_info['filename']
self.labels = meta_info['instrument']
self.audio_source = meta_info['audio_source']
label_code = []
for i in range(len(self.labels)):
label_code.append(label_to_ix[self.labels[i]])
self.label_codes = np.array(label_code) # -
self.sub_indexes = defaultdict(list)
target_max = np.max(self.label_codes) # -
for i in range(target_max + 1):
self.sub_indexes[i] = np.where(self.label_codes == i)[0] # -
def __getitem__(self, index):
audio_feature = wave_to_tfr(os.path.join(self.audio_dir, self.audio_source[index], 'audio',
self.filenames[index] + '.wav'))
label_out = self.label_codes[index]
return audio_feature, label_out
def __len__(self):
return len(self.filenames)
def nsynth_dataset_for_fscil(args_):
label_per_session = [list(np.array(range(args_.base_class)))] + \
[list(np.array(range(args_.way)) + args_.way * task_id + args_.base_class)
for task_id in range(args_.tasks)]
dataset_train = NsynthDatasets(args_, phase='train')
dataset_val = NsynthDatasets(args_, phase='val')
dataset_test = NsynthDatasets(args_, phase='test')
train_datasets = []
test_datasets = []
all_datasets = {}
for session_id in range(args_.session):
train_datasets.append(SubDatasetTrain(dataset_train, label_per_session, args_, session_id))
test_datasets.append(SubDatasetTest(dataset_test, label_per_session, session_id))
all_datasets['train'] = train_datasets
all_datasets['val'] = dataset_val #
all_datasets['test'] = test_datasets
return all_datasets
class SubDatasetTrain(Dataset):
def __init__(self, dataset, sublabels, args__, task_ids):
self.ds = dataset
self.indexes = []
self.sub_indexes = defaultdict(list)
if task_ids == 0:
sublabel = sublabels[task_ids] # -
for label in sublabel:
self.indexes.extend(dataset.sub_indexes[int(label)]) # -
self.sub_indexes[label] = dataset.sub_indexes[int(label)] # -
else:
sublabel = sublabels[task_ids]
# -
for label in sublabel:
shot_sample = random.sample(list(dataset.sub_indexes[int(label)]), args__.shot)
self.indexes.extend(shot_sample)
self.sub_indexes[label] = shot_sample
def __getitem__(self, item):
return self.ds[self.indexes[item]] # -
def __len__(self):
return len(self.indexes)
class SubDatasetTest(Dataset):
def __init__(self, dataset, sublabels, task_ids):
self.ds = dataset
self.sub_indexes = []
for task in range(task_ids + 1):
sublabel = sublabels[task]
for label in sublabel:
self.sub_indexes.extend(dataset.sub_indexes[int(label)])
def __getitem__(self, item):
return self.ds[self.sub_indexes[item]]
def __len__(self):
return len(self.sub_indexes)
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--metapath', type=str, required=True, help='path to nsynth-xxx-fs-meta folder')
parser.add_argument('--audiopath', type=str, required=True, help='path to The NSynth Dataset folder)')
parser.add_argument('--num_class', type=int, default=100, help='Total number of classes in the dataset')
parser.add_argument('--base_class', type=int, default=55, help='number of base class (default: 60)')
parser.add_argument('--way', type=int, default=5, help='class number of per task (default: 5)')
parser.add_argument('--shot', type=int, default=5, help='shot of per class (default: 5)')
# hyper option
parser.add_argument('--session', type=int, default=10, metavar='N',
help='num. of sessions, including one base session and n incremental sessions (default:10)')
return parser.parse_args()
if __name__ == '__main__':
args = parse_args()
args.tasks = args.session - 1
train_dataset = NsynthDatasets(args, phase='train')
val_dataset = NsynthDatasets(args, phase='val')
test_dataset = NsynthDatasets(args, phase='test')
train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=48, shuffle=True,
num_workers=10, pin_memory=True)
test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=48 * 2, shuffle=False,
num_workers=10, pin_memory=True)
val_loader = torch.utils.data.DataLoader(val_dataset, batch_size=48, shuffle=True,
num_workers=32, pin_memory=True)
# loop all the batch
num_epochs = 1
# data_loader = train_loader
# data_loader = val_loader
data_loader = test_loader
for epoch in range(num_epochs):
for batch_idx, batch_data in enumerate(data_loader):
# - unpack data
fea_batch, label_batch = batch_data
# forward backward ,update, etc.
if (batch_idx + 1) % 2 == 0:
print(f'epoch {epoch + 1}/{num_epochs},'
f'features shape : {fea_batch.shape},'
f' label: {label_batch}\n'
)
print('done.\n\n\n\n')
# load data for FCAC
datasets = nsynth_dataset_for_fscil(args)
# session i
i = 1
trainset_i = datasets['train'][i]
valset_0 = datasets['val'] # -
testset_i = datasets['test'][i]