forked from ZZWaang/polyphonic-chord-texture-disentanglement
-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathsave_z.py
130 lines (116 loc) · 6.85 KB
/
save_z.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
import os
import numpy as np
import pretty_midi
import torch
from tqdm import tqdm
from inference import compute_voicing_multihot, accompaniment_generation
from models.model import DisentangleVAE, DisentangleVoicingTextureVAE
from utils.utils import melody_split, chord_split, extract_voicing, chord_data2matrix, midi2pr, pr2midi
def inference_stage1(chord_table, acc_ensemble, checkpoint='data/model_master_final.pt', save_z=None, decode_z=None):
if decode_z is not None:
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
model = DisentangleVAE.init_model(device).to(device)
checkpoint = torch.load(checkpoint, map_location=device)
model.load_state_dict(checkpoint)
return model.inference_only_decode(decode_z)
acc_ensemble = melody_split(acc_ensemble, window_size=32, hop_size=32, vector_size=128)
chord_table = chord_split(chord_table, 8, 8)
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
model = DisentangleVAE.init_model(device).to(device)
checkpoint = torch.load(checkpoint, map_location=device)
model.load_state_dict(checkpoint)
pr_matrix = torch.from_numpy(acc_ensemble).float().to(device)
gt_chord = torch.from_numpy(chord_table).float().to(device)
if save_z:
model.inference_save_z(pr_matrix, gt_chord, sample=False, z_path=save_z)
else:
return model.inference_with_loss(pr_matrix, gt_chord, sample=False)
def inference_stage1_chord(chord_table, acc_ensemble, checkpoint='data/model_master_final.pt', decode_z=None):
if decode_z is not None:
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
model = DisentangleVAE.init_model(device).to(device)
checkpoint = torch.load(checkpoint, map_location=device)
model.load_state_dict(checkpoint)
est_x, recon_root, recon_chroma, recon_bass = model.inference_only_decode(decode_z, with_chord=True)
return recon_root, recon_chroma, recon_bass
acc_ensemble = melody_split(acc_ensemble, window_size=32, hop_size=32, vector_size=128)
chord_table = chord_split(chord_table, 8, 8)
device = torch.device('cuda') if torch.cuda.is_available() else torch.device('cpu')
model = DisentangleVAE.init_model(device).to(device)
checkpoint = torch.load(checkpoint, map_location=device)
model.load_state_dict(checkpoint)
pr_matrix = torch.from_numpy(acc_ensemble).float().to(device)
gt_chord = torch.from_numpy(chord_table).float().to(device)
return model.inference_with_loss(pr_matrix, gt_chord, sample=False)
def chroma2midi(chroma, bass):
midi = pretty_midi.PrettyMIDI()
piano = pretty_midi.Instrument(program=0)
for time in range(len(chroma)):
for pitch in range(12):
if chroma[time, pitch][0] < chroma[time, pitch][1]:
note = pretty_midi.Note(
velocity=100, pitch=pitch + 60, start=time * 2, end=(time + 1) * 2)
piano.notes.append(note)
for time in range(len(bass)):
pitch = np.argmax(bass[time])
note = pretty_midi.Note(velocity=100, pitch=pitch + 36, start=time * 2, end=(time + 1) * 2)
piano.notes.append(note)
midi.instruments.append(piano)
return midi
if __name__ == '__main__':
# for file in tqdm(os.listdir('zv_new')):
# try:
# chord_midi = pretty_midi.PrettyMIDI(f'zv_new/{file}')
# voicing_midi = pretty_midi.PrettyMIDI(f'zv_new/{file}')
# chord = chord_data2matrix(chord_midi.instruments[0], chord_midi.get_downbeats(), 'quarter')
# chord = chord[::16, :]
# voicing = midi2pr(voicing_midi, down_sample=4)
# for row in range(len(voicing)):
# for pitch in range(128):
# if voicing[row, pitch] > 0:
# voicing[row, pitch] = 4
# while len(chord) < 8:
# chord = np.row_stack((chord, np.array([0]*36)))
# while len(voicing) < 32:
# voicing = np.row_stack((voicing, np.array([0]*128)))
# name = file.split('.')[0]+'.pt'
# inference_stage1(chord, voicing, checkpoint='result_2023-06-06_122449/models/disvae-nozoth_final.pt', save_z=f'zs/s1/{name}')
# except:
# print(file)
# path = f'zv/4-23.mid_v.mid'
# chord_midi = pretty_midi.PrettyMIDI(path)
# voicing_midi = pretty_midi.PrettyMIDI(path)
# chord = chord_data2matrix(chord_midi.instruments[0], chord_midi.get_downbeats(), 'quarter')
# chord = chord[::16, :]
# voicing = midi2pr(voicing_midi, down_sample=4)
# for row in range(len(voicing)):
# for pitch in range(128):
# if voicing[row, pitch] > 0:
# voicing[row, pitch] = 4
# while len(chord) < 8:
# chord = np.row_stack((chord, np.array([0]*36)))
# while len(voicing) < 32:
# voicing = np.row_stack((voicing, np.array([0]*128)))
# pr2midi(voicing).write('inspect.mid')
# inference_stage1(chord, voicing, checkpoint='data/train_stage1_20220818.pt', save_z='hahaha.pt')
for file in os.listdir(f'infer_output/test_numlayers6_lr0.001_epoch5'):
if file.endswith('.pt') and 'output' in file:
z = torch.load(f'infer_output/test_numlayers6_lr0.001_epoch5/{file}', map_location=torch.device('cuda'))
for i in range(len(z)):
print(file, i)
# recon_root, recon_chroma, recon_bass = inference_stage1_chord(None, None,
# checkpoint='data/train_stage1_20220818.pt',
# decode_z=z[i].unsqueeze(0))
# recon_chroma = recon_chroma.squeeze(0).cpu().detach().numpy()
# recon_root = recon_root.squeeze(0).cpu().detach().numpy()
# chroma2midi(recon_chroma, recon_root).write(f'infer_output/ouput_0523/{folder}/{file.split(".")[0]}_{i}_chord.mid')
est_x = inference_stage1(None, None, checkpoint='result_2023-06-06_122449/models/disvae-nozoth_final.pt', save_z=None,
decode_z=z[i].unsqueeze(0))
accompaniment_generation(est_x, 30).write(f'infer_output/test_numlayers6_lr0.001_epoch5/{file.split(".")[0]}_{i}.mid')
# z = torch.load(r'D:\projects\polydis2\zs\s1\487-23.pt', map_location=torch.device('cuda'))
# est_x = inference_stage1(None, None, checkpoint='data/train_stage1_20220818.pt', save_z=None, decode_z=z)
# accompaniment_generation(est_x, 30).write('487-23.mid')
# z = torch.load(r'D:\projects\polydis2\zs\s1\487-31.pt', map_location=torch.device('cuda'))
# est_x = inference_stage1(None, None, checkpoint='data/train_stage1_20220818.pt', save_z=None,
# decode_z=z)
# accompaniment_generation(est_x, 30).write('487-31.mid')