File size: 5,995 Bytes
7b918f7 |
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 |
import numpy as np
import os
import librosa
import tqdm
import pickle
import random
import argparse
import yaml
import pathlib
def get_arg():
parser = argparse.ArgumentParser()
parser.add_argument("--config_path", required=True, type=pathlib.Path)
parser.add_argument("--corpus_type", default=None, type=str)
parser.add_argument("--source_path", default=None, type=pathlib.Path)
parser.add_argument("--source_path_task", default=None, type=pathlib.Path)
parser.add_argument("--aux_path", default=None, type=pathlib.Path)
parser.add_argument("--preprocessed_path", default=None, type=pathlib.Path)
parser.add_argument("--n_train", default=None, type=int)
parser.add_argument("--n_val", default=None, type=int)
parser.add_argument("--n_test", default=None, type=int)
return parser.parse_args()
def preprocess(config):
# configs
preprocessed_dir = pathlib.Path(config["general"]["preprocessed_path"])
n_train = config["preprocess"]["n_train"]
n_val = config["preprocess"]["n_val"]
n_test = config["preprocess"]["n_test"]
SR = config["preprocess"]["sampling_rate"]
os.makedirs(preprocessed_dir, exist_ok=True)
sourcepath = pathlib.Path(config["general"]["source_path"])
if config["general"]["corpus_type"] == "single":
fulllist = list(sourcepath.glob("*.wav"))
random.seed(0)
random.shuffle(fulllist)
train_filelist = fulllist[:n_train]
val_filelist = fulllist[n_train : n_train + n_val]
test_filelist = fulllist[n_train + n_val : n_train + n_val + n_test]
filelist = train_filelist + val_filelist + test_filelist
elif config["general"]["corpus_type"] == "multi-seen":
fulllist = list(sourcepath.glob("*/*.wav"))
random.seed(0)
random.shuffle(fulllist)
train_filelist = fulllist[:n_train]
val_filelist = fulllist[n_train : n_train + n_val]
test_filelist = fulllist[n_train + n_val : n_train + n_val + n_test]
filelist = train_filelist + val_filelist + test_filelist
elif config["general"]["corpus_type"] == "multi-unseen":
spk_list = list(set([x.parent for x in sourcepath.glob("*/*.wav")]))
train_filelist = []
val_filelist = []
test_filelist = []
random.seed(0)
random.shuffle(spk_list)
for i, spk in enumerate(spk_list):
sourcespkpath = sourcepath / spk
if i < n_train:
train_filelist.extend(list(sourcespkpath.glob("*.wav")))
elif i < n_train + n_val:
val_filelist.extend(list(sourcespkpath.glob("*.wav")))
elif i < n_train + n_val + n_test:
test_filelist.extend(list(sourcespkpath.glob("*.wav")))
filelist = train_filelist + val_filelist + test_filelist
else:
raise NotImplementedError(
"corpus_type specified in config.yaml should be {single, multi-seen, multi-unseen}"
)
with open(preprocessed_dir / "train.txt", "w", encoding="utf-8") as f:
for m in train_filelist:
f.write(str(m) + "\n")
with open(preprocessed_dir / "val.txt", "w", encoding="utf-8") as f:
for m in val_filelist:
f.write(str(m) + "\n")
with open(preprocessed_dir / "test.txt", "w", encoding="utf-8") as f:
for m in test_filelist:
f.write(str(m) + "\n")
for wp in tqdm.tqdm(filelist):
if config["general"]["corpus_type"] == "single":
basename = str(wp.stem)
else:
basename = str(wp.parent.name) + "-" + str(wp.stem)
wav, _ = librosa.load(wp, sr=SR)
wavsegs = []
if config["general"]["aux_path"] != None:
auxpath = pathlib.Path(config["general"]["aux_path"])
if config["general"]["corpus_type"] == "single":
wav_aux, _ = librosa.load(auxpath / wp.name, sr=SR)
else:
wav_aux, _ = librosa.load(auxpath / wp.parent.name / wp.name, sr=SR)
wavauxsegs = []
if config["general"]["aux_path"] == None:
wavsegs.append(wav)
else:
min_seq_len = min(len(wav), len(wav_aux))
wav = wav[:min_seq_len]
wav_aux = wav_aux[:min_seq_len]
wavsegs.append(wav)
wavauxsegs.append(wav_aux)
wavsegs = np.asarray(wavsegs).astype(np.float32)
if config["general"]["aux_path"] != None:
wavauxsegs = np.asarray(wavauxsegs).astype(np.float32)
else:
wavauxsegs = None
d_preprocessed = {"wavs": wavsegs, "wavsaux": wavauxsegs}
with open(preprocessed_dir / "{}.pickle".format(basename), "wb") as fw:
pickle.dump(d_preprocessed, fw)
if __name__ == "__main__":
args = get_arg()
config = yaml.load(open(args.config_path, "r"), Loader=yaml.FullLoader)
for key in ["corpus_type", "source_path", "aux_path", "preprocessed_path"]:
if getattr(args, key) != None:
config["general"][key] = str(getattr(args, key))
for key in ["n_train", "n_val", "n_test"]:
if getattr(args, key) != None:
config["preprocess"][key] = getattr(args, key)
print("Performing preprocessing ...")
preprocess(config)
if "dual" in config:
if config["dual"]["enable"]:
task_config = yaml.load(
open(config["dual"]["config_path"], "r"), Loader=yaml.FullLoader
)
task_preprocessed_dir = (
pathlib.Path(config["general"]["preprocessed_path"]).parent
/ pathlib.Path(task_config["general"]["preprocessed_path"]).name
)
task_config["general"]["preprocessed_path"] = task_preprocessed_dir
if args.source_path_task != None:
task_config["general"]["source_path"] = args.source_path_task
print("Performing preprocessing for multi-task learning ...")
preprocess(task_config)
|