Upload augment.py with huggingface_hub
Browse files- augment.py +78 -0
augment.py
ADDED
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import os
|
3 |
+
import subprocess
|
4 |
+
|
5 |
+
import librosa
|
6 |
+
import numpy as np
|
7 |
+
import soundfile as sf
|
8 |
+
from tqdm import tqdm
|
9 |
+
|
10 |
+
from lib import dataset
|
11 |
+
from lib import spec_utils
|
12 |
+
|
13 |
+
|
14 |
+
if __name__ == '__main__':
|
15 |
+
p = argparse.ArgumentParser()
|
16 |
+
p.add_argument('--sr', '-r', type=int, default=44100)
|
17 |
+
p.add_argument('--hop_length', '-l', type=int, default=1024)
|
18 |
+
p.add_argument('--n_fft', '-f', type=int, default=2048)
|
19 |
+
p.add_argument('--pitch', '-p', type=int, default=-1)
|
20 |
+
p.add_argument('--mixtures', '-m', required=True)
|
21 |
+
p.add_argument('--instruments', '-i', required=True)
|
22 |
+
args = p.parse_args()
|
23 |
+
|
24 |
+
input_i = 'input_i_{}.wav'.format(args.pitch)
|
25 |
+
input_v = 'input_v_{}.wav'.format(args.pitch)
|
26 |
+
output_i = 'output_i_{}.wav'.format(args.pitch)
|
27 |
+
output_v = 'output_v_{}.wav'.format(args.pitch)
|
28 |
+
cmd_i = 'soundstretch {} {} -pitch={}'.format(input_i, output_i, args.pitch)
|
29 |
+
cmd_v = 'soundstretch {} {} -pitch={}'.format(input_v, output_v, args.pitch)
|
30 |
+
cache_suffix = '_pitch{}.npy'.format(args.pitch)
|
31 |
+
|
32 |
+
cache_dir = 'sr{}_hl{}_nf{}'.format(args.sr, args. hop_length, args.n_fft)
|
33 |
+
mix_cache_dir = os.path.join(args.mixtures, cache_dir)
|
34 |
+
inst_cache_dir = os.path.join(args.instruments, cache_dir)
|
35 |
+
os.makedirs(mix_cache_dir, exist_ok=True)
|
36 |
+
os.makedirs(inst_cache_dir, exist_ok=True)
|
37 |
+
|
38 |
+
filelist = dataset.make_pair(args.mixtures, args.instruments)
|
39 |
+
for mix_path, inst_path in tqdm(filelist):
|
40 |
+
mix_basename = os.path.splitext(os.path.basename(mix_path))[0]
|
41 |
+
mix_cache_path = os.path.join(mix_cache_dir, mix_basename + cache_suffix)
|
42 |
+
|
43 |
+
inst_basename = os.path.splitext(os.path.basename(inst_path))[0]
|
44 |
+
inst_cache_path = os.path.join(inst_cache_dir, inst_basename + cache_suffix)
|
45 |
+
|
46 |
+
if os.path.exists(mix_cache_path) and os.path.exists(inst_cache_path):
|
47 |
+
continue
|
48 |
+
|
49 |
+
X, _ = librosa.load(
|
50 |
+
mix_path, sr=args.sr, mono=False, dtype=np.float32, res_type='kaiser_fast')
|
51 |
+
y, _ = librosa.load(
|
52 |
+
inst_path, sr=args.sr, mono=False, dtype=np.float32, res_type='kaiser_fast')
|
53 |
+
|
54 |
+
X, y = spec_utils.align_wave_head_and_tail(X, y, args.sr)
|
55 |
+
v = X - y
|
56 |
+
|
57 |
+
sf.write(input_i, y.T, args.sr)
|
58 |
+
sf.write(input_v, v.T, args.sr)
|
59 |
+
subprocess.call(cmd_i, stderr=subprocess.DEVNULL)
|
60 |
+
subprocess.call(cmd_v, stderr=subprocess.DEVNULL)
|
61 |
+
|
62 |
+
y, _ = librosa.load(
|
63 |
+
output_i, sr=args.sr, mono=False, dtype=np.float32, res_type='kaiser_fast')
|
64 |
+
v, _ = librosa.load(
|
65 |
+
output_v, sr=args.sr, mono=False, dtype=np.float32, res_type='kaiser_fast')
|
66 |
+
|
67 |
+
X = y + v
|
68 |
+
|
69 |
+
spec = spec_utils.wave_to_spectrogram(X, args.hop_length, args.n_fft)
|
70 |
+
np.save(mix_cache_path, spec)
|
71 |
+
|
72 |
+
spec = spec_utils.wave_to_spectrogram(y, args.hop_length, args.n_fft)
|
73 |
+
np.save(inst_cache_path, spec)
|
74 |
+
|
75 |
+
os.remove(input_i)
|
76 |
+
os.remove(input_v)
|
77 |
+
os.remove(output_i)
|
78 |
+
os.remove(output_v)
|