Spaces:
Running
Running
File size: 5,063 Bytes
e86d760 88b2fbf e86d760 88b2fbf e86d760 7f9e32d e86d760 7f9e32d e86d760 88b2fbf e86d760 7f9e32d e86d760 7f9e32d e86d760 88b2fbf e86d760 |
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 |
#!/usr/bin/python3
# -*- coding: utf-8 -*-
import argparse
import json
import os
from pathlib import Path
import random
import sys
pwd = os.path.abspath(os.path.dirname(__file__))
sys.path.append(os.path.join(pwd, "../../"))
from tqdm import tqdm
import librosa
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument("--file_dir", default="./", type=str)
parser.add_argument(
"--noise_dir",
default=r"E:\Users\tianx\HuggingDatasets\nx_noise\data\noise",
type=str
)
parser.add_argument(
"--speech_dir",
default=r"E:\programmer\asr_datasets\aishell\data_aishell\wav\train",
type=str
)
parser.add_argument("--train_dataset", default="train.jsonl", type=str)
parser.add_argument("--valid_dataset", default="valid.jsonl", type=str)
parser.add_argument("--duration", default=2.0, type=float)
parser.add_argument("--min_snr_db", default=-10, type=float)
parser.add_argument("--max_snr_db", default=20, type=float)
parser.add_argument("--target_sample_rate", default=8000, type=int)
parser.add_argument("--max_count", default=10000, type=int)
args = parser.parse_args()
return args
def filename_generator(data_dir: str):
data_dir = Path(data_dir)
for filename in data_dir.glob("**/*.wav"):
yield filename.as_posix()
def target_second_signal_generator(data_dir: str, duration: int = 2, sample_rate: int = 8000, max_epoch: int = 20000):
data_dir = Path(data_dir)
for epoch_idx in range(max_epoch):
for filename in data_dir.glob("**/*.wav"):
signal, _ = librosa.load(filename.as_posix(), sr=sample_rate)
raw_duration = librosa.get_duration(y=signal, sr=sample_rate)
if raw_duration < duration:
# print(f"duration less than {duration} s. skip filename: {filename.as_posix()}")
continue
if signal.ndim != 1:
raise AssertionError(f"expected ndim 1, instead of {signal.ndim}")
signal_length = len(signal)
win_size = int(duration * sample_rate)
for begin in range(0, signal_length - win_size, win_size):
row = {
"epoch_idx": epoch_idx,
"filename": filename.as_posix(),
"raw_duration": round(raw_duration, 4),
"offset": round(begin / sample_rate, 4),
"duration": round(duration, 4),
}
yield row
def main():
args = get_args()
file_dir = Path(args.file_dir)
file_dir.mkdir(exist_ok=True)
noise_dir = Path(args.noise_dir)
speech_dir = Path(args.speech_dir)
noise_generator = target_second_signal_generator(
noise_dir.as_posix(),
duration=args.duration,
sample_rate=args.target_sample_rate,
max_epoch=100000,
)
speech_generator = target_second_signal_generator(
speech_dir.as_posix(),
duration=args.duration,
sample_rate=args.target_sample_rate,
max_epoch=1,
)
dataset = list()
count = 0
process_bar = tqdm(desc="build dataset excel")
with open(args.train_dataset, "w", encoding="utf-8") as ftrain, open(args.valid_dataset, "w", encoding="utf-8") as fvalid:
for noise, speech in zip(noise_generator, speech_generator):
if count >= args.max_count:
break
noise_filename = noise["filename"]
noise_raw_duration = noise["raw_duration"]
noise_offset = noise["offset"]
noise_duration = noise["duration"]
speech_filename = speech["filename"]
speech_raw_duration = speech["raw_duration"]
speech_offset = speech["offset"]
speech_duration = speech["duration"]
random1 = random.random()
random2 = random.random()
row = {
"noise_filename": noise_filename,
"noise_raw_duration": noise_raw_duration,
"noise_offset": noise_offset,
"noise_duration": noise_duration,
"speech_filename": speech_filename,
"speech_raw_duration": speech_raw_duration,
"speech_offset": speech_offset,
"speech_duration": speech_duration,
"snr_db": random.uniform(args.min_snr_db, args.max_snr_db),
"random1": random1,
}
row = json.dumps(row, ensure_ascii=False)
if random2 < 0.8:
ftrain.write(f"{row}\n")
else:
fvalid.write(f"{row}\n")
count += 1
duration_seconds = count * args.duration
duration_hours = duration_seconds / 3600
process_bar.update(n=1)
process_bar.set_postfix({
# "duration_seconds": round(duration_seconds, 4),
"duration_hours": round(duration_hours, 4),
})
return
if __name__ == "__main__":
main()
|