DiffSingerKR/Pattern_Test.py at main · mizuirodev/DiffSingerKR · GitHub

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
import numpy as np
import matplotlib.pyplot as plt
from tqdm import tqdm
import pickle, hgtk, os
from matplotlib import rc
rc("font", family="Malgun Gothic")

class Note_Predictor:
    def __init__(self):
        notes = np.arange(0, 128)
        # log_f0s = np.log(440 * 2 ** ((notes - 69) / 12))
        log_f0s = np.log(440 * 2 ** ((notes - 69 - 12) / 12))
        log_f0s[0] = 0.0
        self.criterion = np.expand_dims(log_f0s, axis= 0)   # [1, 128]

    def __call__(self, log_f0):
        '''
        log_f0: [F0_t]
        '''
        return np.argmin(
            np.abs(np.expand_dims(log_f0, axis= 1) - self.criterion),
            axis= 1
            )

def Lyric_Compose(lyric):
    composed_lyric = []
    current_string = []
    for letter in lyric + ['Temp_Lyric']:
        if len(current_string) > 0 and current_string[-1] != letter:
            if current_string[-1][-1] != '_' and not current_string[-1] in hgtk.checker.JAMO:
                composed_lyric.append(current_string[0])
                composed_lyric.extend([None] * (len(current_string) - 1))
                current_string = []
            elif current_string[-1][-1] == '_':
                for x in set(current_string):
                    if x in hgtk.letter.CHO:
                        onset = x
                    elif x in hgtk.letter.JOONG:
                        nuclues = x
                    elif len(x) == 2 and x[:1] in hgtk.letter.JONG:
                        coda = x[0]
                    elif x == '_':
                        coda = ''
                    else:
                        raise ValueError(set(current_string))
                composed_lyric.append(hgtk.letter.compose(onset, nuclues, coda))
                composed_lyric.extend([None] * (len(current_string) - 1))
                current_string = []

        current_string.append(letter)

    return composed_lyric

if __name__ == '__main__':
    note_predictor = Note_Predictor()

    pattern_path = 'D:/Datasets/22K.MusicXX/Eval/CSD'
    output_path = 'D:/Datasets/22K.MusicXX/SVS_Pattern_Test2'
    os.makedirs(output_path, exist_ok= True)
    root_and_file_list = [
        (root, file)
        for root, _, files in os.walk(pattern_path)
        for file in files
        ]
    if os.path.exists(os.path.join(output_path, 'Done.txt')):
        with open(os.path.join(output_path, 'Done.txt'), 'r') as f:
            done_list = [line.strip() for line in f.readlines()]
    else:
        done_list = []
    for root, file in tqdm(root_and_file_list, total= len(root_and_file_list)):
        figure_path = os.path.join(output_path, f'{os.path.splitext(file)[0]}.png')
        if figure_path in done_list:
            continue
        elif os.path.exists(figure_path):
            continue
        elif os.path.splitext(file)[1].upper() != '.PICKLE':
            continue

        pattern = pickle.load(open(os.path.join(root, file), 'rb'))
        if not all([(key in pattern.keys()) for key in ['Mel', 'Log_F0', 'Note_Expand', 'Lyric_Expand']]):
            continue

        note_from_log_f0 = note_predictor(np.clip(pattern['Log_F0'], -0.0, np.inf))
        note = pattern['Note_Expand']
        if np.abs(note_from_log_f0 - note).mean() > 10.0:
            lyric = Lyric_Compose(pattern['Lyric_Expand'])

            # plt.figure(figsize= (200, 20))
            plt.figure(figsize= (75, 7.5))
            plt.subplot(211)
            plt.imshow(pattern['Mel'].T, aspect='auto', origin='lower')
            plt.subplot(212)
            plt.plot(note_from_log_f0)
            plt.plot(note)
            plt.xticks(
                range(len(lyric)),
                lyric,
                fontsize = 8
                )
            plt.margins(x= 0)
            plt.tight_layout()
            plt.savefig(figure_path)
            plt.close()

        with open(os.path.join(output_path, 'Done.txt'), 'a') as f:
            f.write(figure_path + '\n')