-
Notifications
You must be signed in to change notification settings - Fork 5
/
data.py
203 lines (190 loc) · 8.32 KB
/
data.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
import os
import pathlib
from tqdm import tqdm
import pretty_midi as pm
import numpy as np
import torch
import warnings
import re
import librosa.display
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import torch.utils.data as data
INPUT_LENGTH = 8192
MAX_LENGTH = 32768
pathlist = list(pathlib.Path('Classics').glob('**/*.mid'))
trainlist = pathlist[:-144]
testlist = pathlist[-144:]
def natural_sort_key(s, _nsre=re.compile('(\\d+)')):
return [int(text) if text.isdigit() else text.lower() for text in _nsre.split(s)]
def full_piano_roll(path, receptive_field):
with warnings.catch_warnings():
warnings.simplefilter('ignore')
song = pm.PrettyMIDI(midi_file=str(path))
piano_rolls = [(_.get_piano_roll(fs=song.resolution), _.program) for _ in song.instruments if not _.is_drum]
drum_rolls = [(_.get_piano_roll(fs=song.resolution), _.program) for _ in song.instruments if _.is_drum]
length = np.amax([roll.shape[1] for roll, _ in piano_rolls + drum_rolls])
data = np.zeros(shape=(129 * 129 + 1, length))
for roll, instrument in piano_rolls:
data[instrument * 128: (instrument + 1) * 128] += np.pad(roll, [(0, 0), (0, length - roll.shape[1])], 'constant')
data[128 * 129 + instrument] = 1
for roll, instrument in drum_rolls:
data[128 * 128 : 128 * 129] += np.pad(roll, [(0, 0), (0, length - roll.shape[1])], 'constant')
data[129 * 129 - 1] = 1
if length >= MAX_LENGTH:
num = np.random.randint(0, length - MAX_LENGTH + 1)
data = data[:, num : num + MAX_LENGTH]
data[129 * 129] += 1 - data.sum(axis=0)
data = data > 0
answer = np.transpose(data[:, receptive_field + 1:], (1, 0))
return data.astype(np.float32), answer.astype(np.float32)
def piano_roll(path, receptive_field):
with warnings.catch_warnings():
warnings.simplefilter('ignore')
song = pm.PrettyMIDI(midi_file=str(path))
classes = [0, 3, 5, 7, 8, 9]
limits = [[24, 96], [36, 84], [24, 96], [36, 84], [36, 84], [60, 96]]
piano_rolls = [(_.get_piano_roll(fs=song.resolution), _.program) for _ in song.instruments if not _.is_drum and _.program // 8 in classes]
length = np.amax([roll.shape[1] for roll, _ in piano_rolls])
data_full = np.zeros(shape=(331, length))
for roll, instrument in piano_rolls:
i = classes.index(instrument // 8)
sliced_roll = roll[limits[i][0]:limits[i][1]]
data_full[limits[i][0]:limits[i][1]] += np.pad(sliced_roll, [(0, 0), (0, length - sliced_roll.shape[1])], 'constant')
data_full[325 + i] = 1
if length < INPUT_LENGTH:
data = np.pad(data_full, [(0, 0), (INPUT_LENGTH - length, 0)], 'constant')
else:
num = np.random.randint(0, length - INPUT_LENGTH + 1)
data = data_full[:, num : INPUT_LENGTH + num]
data[324] += 1 - data[:324].sum(axis = 0)
data = data > 0
answer = np.transpose(data[:325, receptive_field + 1:], (1, 0))
return data.astype(np.float32), answer.astype(np.float32)
def clean(x):
return x[:-1]
def save_roll(x, step):
fig = plt.figure(figsize=(72, 24))
librosa.display.specshow(x, x_axis='time', hop_length=1, sr=96, fmin=pm.note_number_to_hz(12))
plt.title('{}'.format(step))
fig.savefig('Samples/{}.png'.format(step))
plt.close(fig)
def piano_rolls_to_midi(x, fs=96):
channels = [72, 48, 72, 48, 48, 36]
for i in range(1, 6):
channels[i] += channels[i - 1]
x = np.split(x * 100, channels)
midi = pm.PrettyMIDI()
limits = [[24, 96], [36, 84], [24, 96], [36, 84], [36, 84], [60, 96]]
instruments = [0, 24, 40, 56, 64, 72]
for roll, instrument, limit in zip(x, instruments, limits):
current_inst = pm.Instrument(instrument)
current_roll = np.pad(roll, [(limit[0], 128 - limit[1]), (1, 1)], 'constant')
notes = current_roll.shape[0]
velocity_changes = np.nonzero(np.diff(current_roll).T)
prev_velocities = np.zeros(notes, dtype=int)
note_on_time = np.zeros(notes)
for time, note in zip(*velocity_changes):
velocity = current_roll[note, time + 1]
time /= fs
if velocity > 0:
if prev_velocities[note] == 0:
note_on_time[note] = time
prev_velocities[note] = velocity
else:
if time > note_on_time[note] + 1 / fs:
pm_note = pm.Note(
velocity=prev_velocities[note],
pitch=note,
start=note_on_time[note],
end=time
)
current_inst.notes.append(pm_note)
prev_velocities[note] = 0
midi.instruments.append(current_inst)
return midi
class Dataset(data.Dataset):
def __init__(self, train, receptive_field):
super(Dataset, self).__init__()
if train:
self.pathlist = trainlist
else:
self.pathlist = testlist
self.receptive_field = receptive_field
def __getitem__(self, index):
data = piano_roll(self.pathlist[index], self.receptive_field)
return data
def __len__(self):
return len(self.pathlist)
class DataLoader(data.DataLoader):
def __init__(self, batch_size, receptive_field, shuffle=True, num_workers=16, train=True):
super(DataLoader, self).__init__(Dataset(train, receptive_field), batch_size, shuffle, num_workers=num_workers)
def Test():
pathlist = list(pathlib.Path('Datasets/lmd_matched').glob('**/*.mid'))
# pathlist = list(pathlib.Path('Classics').glob('**/*.mid'))# + list(pathlib.Path('Classics').glob('**/*.MID'))
np.random.shuffle(pathlist)
print(len(pathlist))
# instruments = [0, 3, 5, 7, 8, 9]
# limits = [[24, 96], [36, 84], [24, 96], [36, 84], [36, 84], [60, 96]]
lengthlist = []
resolutionlist = []
namelist = []
programlist = [0] * 128
# ratiolist = []
over_limit = 0
for path in tqdm(pathlist[:1000]):
try:
song = pm.PrettyMIDI(midi_file=str(path))
except:
continue
#resolutionlist.append(song.resolution)
lengthlist.append(len(song.instruments))
# lengthlist.append(song.get_end_time())
# rolls = [(_.get_piano_roll(), _.program) for _ in song.instruments if _.program // 8 in instruments and not _.is_drum]
# rolls = [(_.get_piano_roll(), _.program) for _ in song.instruments if not _.is_drum]
# rolls = [_.program for _ in song.instruments if not _.is_drum]
# drum_rolls = [_ for _ in song.instruments if _.is_drum]
# if len(rolls) > 0:
# program_bool = [0] * 128
# for _, i in rolls:
# program_bool[i] += 1
# for i, j in enumerate(program_bool):
# programlist[i] += j > 0
# lengthlist.append(np.amax([_.shape[1] for _, _1 in rolls]))
# lengthlist.append(song.get_end_time())
# namelist.append((str(path), str(np.amax([_.shape[1] for _, _1 in rolls]))))
# ratiolist.append(np.amax([_.shape[1] for _, _1 in rolls]) / song.get_end_time())
# lengthlist.append((str(path), np.amax([_.shape[1] for _, _1 in rolls] + [INPUT_LENGTH]) - INPUT_LENGTH + 1))
# if np.amax([_.shape[1] for _, _1 in rolls]) >= 8192:
# over_limit += 1
# if song.get_end_time() >= 81:
# over_limit += 1
# if len(drum_rolls) > 0:
# program_bool = [0] * 128
# for i in drum_rolls:
# for note in i.notes:
# program_bool[note.pitch] += 1
# for i, j in enumerate(program_bool):
# programlist[i] += j > 0
print(programlist)
print(over_limit)
print(np.sum(lengthlist))
# file_length = open('lmd_length.txt', 'w')
# for path, length in namelist:
# file_length.write(path + ' ' + str(length) + '\n')
# file_length.close()
# lengthlist /= np.sum(lengthlist)
plt.hist(lengthlist)
# plt.hist(resolutionlist, bins=100)
# plt.axis([0, 1e5, 0, 200])
plt.grid()
plt.show()
# plt.savefig('Images/Amounts.png')
plt.close()
# plt.hist(ratiolist, bins=100)
# plt.grid()
# plt.show()
# plt.close()
if __name__ == '__main__':
Test()