File size: 9,598 Bytes
c961996
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
import sys
sys.path.append('..')

from torch.utils.data import Dataset
import pickle
import random
from . import LyricsCommentData

class LyricsCommentsDataset(Dataset):

    def __init__(self, random=False):
        super(LyricsCommentsDataset, self).__init__()
        self.random = random
        with open("dataset.pkl", "rb") as f:
            self.data = pickle.load(f)

    def __len__(self):
        return len(self.data)

    def __getitem__(self, item):
        lyrics = self.data[item].lyrics
        # if random:
        #     comment = random.choice(self.data[item].comments)
        # else:
        comment = self.data[item].comments[0]
        # the longest?
        for i, (tmp_item, _) in enumerate(self.data[item].comments):
            if len(tmp_item) > len(comment[0]):
                comment = self.data[item].comments[i]

        comment = comment[0] # keep comments w/o rating

        return [lyrics, comment]


class LyricsCommentsDatasetClean(Dataset):

    def __init__(self, random=False):
        super(LyricsCommentsDatasetClean, self).__init__()
        self.random = random
        with open("cleaned_dataset.pkl", "rb") as f:
            self.data = pickle.load(f)

    def __len__(self):
        return len(self.data)

    def __getitem__(self, item):
        lyrics = self.data[item].lyrics
        comment = self.data[item].comment

        return [lyrics, comment]


class LyricsCommentsDatasetPsuedo(Dataset):

    def __init__(self, dataset_path, random=False):
        super(LyricsCommentsDatasetPsuedo, self).__init__()
        self.random = random
        with open(dataset_path, "rb") as f:
            self.data = pickle.load(f)

    def __len__(self):
        return len(self.data)

    def __getitem__(self, item):
        lyrics = self.data[item].lyrics.replace('\n', ';')
        comment = self.data[item].comment

        return [lyrics, comment]


class LyricsCommentsDatasetPsuedo_fusion(Dataset):

    def __init__(self, dataset_path):
        super(LyricsCommentsDatasetPsuedo_fusion, self).__init__()
        with open(dataset_path, "rb") as f:
            self.data = pickle.load(f)

    def __len__(self):
        return len(self.data)


    def __getitem__(self, item):
        lyrics = self.data[item].lyrics.replace('\n', ';')
        comment = self.data[item].comment
        music_id = self.data[item].music4all_id

        return [lyrics, comment, music_id]


from torch.utils.data import Dataset, DataLoader
import torch
from MusicData import MusicData
import csv
import os
from pydub import AudioSegment
import matplotlib.pyplot as plt
from scipy.io import wavfile
from tempfile import mktemp
from scipy import signal
import numpy as np
import torchaudio
import transformers
import nltk


class Music4AllDataset(Dataset):
    def __init__(self,
                 mel_bins,
                 audio_length,
                 pad_length,
                 tag_file_path=r"Music4All/music4all/id_genres.csv",
                 augment=True):
        self.tag_file_path = tag_file_path
        self.allow_cache = True
        self.mel_bins = mel_bins
        self.audio_length = audio_length
        self.pad_length = pad_length
        self.augment = augment
        # read all tags
        tags_file = open(tag_file_path, 'r', encoding='utf-8')
        self.tags_reader = list(csv.reader(tags_file, delimiter='\t'))[1:]
        tags_file.close()
        if self.augment:
            self.data_augmentation()

    def data_augmentation(self):
        pass

    def __len__(self):
        return len(self.tags_reader)

    def __getitem__(self, item):
        """

        :param item: index
        :return: tags and mel-spectrogram.
        """
        id = self.tags_reader[item][0]
        tags = self.tags_reader[item][1] #.split(',')

        # pad tags
        # if len(tags) >= self.pad_length:
        #     tags = tags[:self.pad_length]
        # else:
        #     for i in range(self.pad_length - len(tags)):
        #         tags.append("[PAD]")

        spec_path = os.path.join("Music4All/temp_data/specs/data_cache/", id + ".npy")
        exist_cache = os.path.isfile(spec_path)
        # search cache
        # if exist cache, load
        if self.allow_cache and exist_cache:
            spectrogram = torch.Tensor(np.load(spec_path))
        # if does not exist, calculate and save
        else:
            audio_path = os.path.join("Music4All/music4all/audios",
                                      id + '.mp3'
                                      )
            (data, sample_rate) = torchaudio.backend.sox_io_backend.load(audio_path)
            spectrogram = torchaudio.transforms.MelSpectrogram(n_mels=self.mel_bins,
                                                               n_fft=512,
                                                               sample_rate=sample_rate,
                                                               f_max=8000.0,
                                                               f_min=0.0,
                                                               )(torch.Tensor(data))
            # TODO: There is a huge bug!
            # cut length
            if self.audio_length is not None:
                spectrogram = spectrogram[:, :, :self.audio_length]
            # to mono
            spectrogram = spectrogram[0, :, :].unsqueeze(0)

            if self.allow_cache:
                np.save(spec_path, spectrogram.numpy())

        return tags, spectrogram


class MusCapsDataset(Dataset):
    def __init__(self,
                 mel_bins,
                 audio_length,
                 pad_length,
                 tag_file_path=r"Music4All/music4all/id_genres.csv",
                 augment=True):
        self.tag_file_path = tag_file_path
        self.allow_cache = True
        self.mel_bins = mel_bins
        self.audio_length = audio_length
        self.pad_length = pad_length
        self.augment = augment
        # read all tags
        tags_file = open(tag_file_path, 'r', encoding='utf-8')
        self.tags_reader = list(csv.reader(tags_file, delimiter='\t'))[1:]
        tags_file.close()
        if self.augment:
            self.data_augmentation()

    def data_augmentation(self):
        pass

    def __len__(self):
        return len(self.tags_reader)

    def __getitem__(self, item):
        """

        :param item: index
        :return: tags and mel-spectrogram.
        """
        id = self.tags_reader[item][0]
        tags = self.tags_reader[item][1] #.split(',')

        # pad tags
        # if len(tags) >= self.pad_length:
        #     tags = tags[:self.pad_length]
        # else:
        #     for i in range(self.pad_length - len(tags)):
        #         tags.append("[PAD]")

        spec_path = os.path.join("Music4All/temp_data/specs/data_cache/", id + ".npy")
        exist_cache = os.path.isfile(spec_path)
        # search cache
        # if exist cache, load
        if self.allow_cache and exist_cache:
            spectrogram = torch.Tensor(np.load(spec_path))
        # if does not exist, calculate and save
        else:
            audio_path = os.path.join("Music4All/music4all/audios",
                                      id + '.mp3'
                                      )
            (data, sample_rate) = torchaudio.backend.sox_io_backend.load(audio_path)
            spectrogram = torchaudio.transforms.MelSpectrogram(n_mels=self.mel_bins,
                                                               n_fft=512,
                                                               sample_rate=sample_rate,
                                                               f_max=8000.0,
                                                               f_min=0.0,
                                                               )(torch.Tensor(data))
            # cut length
            if self.audio_length is not None:
                spectrogram = spectrogram[:, :, :self.audio_length]
            # to mono
            spectrogram = spectrogram[0, :, :].unsqueeze(0)
            np.save(spec_path, spectrogram.numpy())

        return tags, spectrogram

class GTZANDataset(Dataset):
    def __init__(self, raw_dataset, is_augment=True, window=1366):
        self.raw = raw_dataset
        self.data = list()
        self.mel_bins = 96
        self.gtzan_genres = [
            "blues",
            "classical",
            "country",
            "disco",
            "hiphop",
            "jazz",
            "metal",
            "pop",
            "reggae",
            "rock",
        ]
        self.is_augment = is_augment
        self.window = window
        self.init()

    def init(self):
        for i, (waveform, sample_rate, label) in enumerate(self.raw):
            spectrogram = torchaudio.transforms.MelSpectrogram(n_mels=self.mel_bins)(torch.Tensor(waveform))
            if self.is_augment:
                self.augment(spectrogram, label)
            else:
                self.data.append((spectrogram[:,:,:self.window], label))

    def augment(self, spectrogram, label):
        length = spectrogram.shape[-1] # length
        # augment audio with sliding window
        hop_length = 250
        slices = (length - self.window) // hop_length
        for i in range(slices):
            self.data.append((spectrogram[:, :, i * hop_length:self.window + i*hop_length], label))



    def __len__(self):
        return len(self.data)

    def __getitem__(self, index):
        spectrogram, label = self.data[index]
        label = self.gtzan_genres.index(label)
        return spectrogram, label