File size: 8,093 Bytes
41bc8a8 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 | import argparse
import glob
import math
import multiprocessing as mp
import os
import statistics
import sys
from multiprocessing import Pool
import numpy as np
import pandas as pd
import soundfile as sf
import torchaudio
from pydub import AudioSegment
from tqdm import tqdm
def check_sample_rate(file_path):
try:
info = sf.info(file_path)
if info.samplerate != 24000:
return file_path
except Exception as e:
return None # In case of error, return None
def process_files(file_list):
with Pool() as pool:
result = list(
tqdm(pool.imap(check_sample_rate, file_list), total=len(file_list))
)
return [file for file in result if file is not None]
def read_paths_from_file(filename):
with open(filename, "r", encoding="utf8") as file:
path_list = []
for i in tqdm(file):
path = i.split("|")[1]
path_list.append(path.strip("\n"))
return path_list[:]
def gather_paths_from_glob():
return glob.glob("./**/*.wav", recursive=True)
def detect_leading_silence(sound, silence_threshold=-50, chunk_size=64):
trim_ms = 0
assert chunk_size > 0
while sound[
trim_ms : trim_ms + chunk_size
].dBFS < silence_threshold and trim_ms < len(sound):
trim_ms += chunk_size
return trim_ms
def preprocess_audio(path, target_dBFS, frame_rate):
durations = []
dbfs = []
audio = AudioSegment.from_file(path)
dbfs.append(audio.dBFS)
audio = audio.set_channels(1)
audio = audio.set_frame_rate(frame_rate).set_sample_width(2)
start_trim = detect_leading_silence(audio)
end_trim = detect_leading_silence(audio.reverse())
duration = len(audio)
audio = audio[start_trim : duration - end_trim]
audio = (
AudioSegment.silent(duration=256, frame_rate=22050)
+ audio
+ AudioSegment.silent(duration=256, frame_rate=22050)
)
if path[-4:] == ".wav":
audio.export(path[:-4] + ".wav", format="wav")
elif path[-5:] == ".flac":
audio.export(path[:-5] + ".flac", format="flac")
else:
audio.export(path[:-4] + ".wav", format="wav")
durations.append(audio.duration_seconds)
return dbfs, durations
def preprocess_audio_chunk(args):
path_list_chunk, target_dBFS, frame_rate, n = args
dbfs = []
durations = []
for i in tqdm(path_list_chunk, desc="preprocess " + str(n)):
try:
audio = AudioSegment.from_file(i)
dbfs_i, durations_i = preprocess_audio(i, target_dBFS, frame_rate)
dbfs.extend(dbfs_i)
durations.extend(durations_i)
except Exception as e:
print(n, i, e)
return dbfs, durations
def preprocess_audio_paths(path_list, target_dBFS, frame_rate, num_workers):
chunk_size = len(path_list) // num_workers
path_chunks = [
path_list[i : i + chunk_size] for i in range(0, len(path_list), chunk_size)
]
with Pool(num_workers) as pool:
results = pool.map(
preprocess_audio_chunk,
[
(chunk, target_dBFS, frame_rate, n)
for n, chunk in enumerate(path_chunks)
],
)
dbfs = []
durations = []
for dbfs_i, durations_i in results:
dbfs.extend(dbfs_i)
durations.extend(durations_i)
return dbfs, durations
def gather_metadata_chunk(args):
path_list_chunk = args
dbfs = []
durations = []
files = []
for i in tqdm(path_list_chunk):
try:
path = i.split("|")[0]
audio = AudioSegment.from_file(path)
if audio.dBFS == -math.inf:
print("=====================")
print(path)
print("=====================")
continue
dbfs.append(audio.dBFS)
durations.append(audio.duration_seconds)
files.append((audio.duration_seconds, i))
if audio.duration_seconds == 0:
print(i)
except Exception as e:
print(e, i)
return dbfs, durations, files
def gather_metadata(path_list, num_workers):
chunk_size = len(path_list) // num_workers
path_chunks = [
path_list[i : i + chunk_size] for i in range(0, len(path_list), chunk_size)
]
with Pool(num_workers) as pool:
results = pool.map(gather_metadata_chunk, [chunk for chunk in path_chunks])
dbfs = []
durations = []
files = []
for dbfs_i, durations_i, files_i in results:
dbfs.extend(dbfs_i)
durations.extend(durations_i)
files.extend(files_i)
files = sorted(files, key=lambda x: x[0])
with open(os.path.join(data_dir, "files_duration.txt"), "w") as file:
file.write("\n".join([i[1] + "|" + str(i[0]) for i in files]))
with open(os.path.join(data_dir, "files.txt"), "w") as file:
file.write("\n".join([i[1] for i in files if 2.0 < float(i[0]) < 15.0]))
return dbfs, durations
def process_audio_data(input_file, mode, num_workers, data_dir):
if input_file:
path_list = read_paths_from_file(input_file)
else:
path_list = gather_paths_from_glob()
speakers = []
for n, i in enumerate(path_list):
try:
speakers.append(i.split("/")[-2])
except:
print(n, i)
print("total audio files:", len(path_list))
if mode == "preprocess":
print("Preprocessing!")
target_dBFS = -24.196741 # not using
frame_rate = 24000
dbfs, durations = preprocess_audio_paths(
path_list[:], target_dBFS, frame_rate, num_workers
)
print("min duration : ", min(durations))
print("max duration : ", max(durations))
print("avg duration : ", sum(durations) / len(durations))
print("Standard Deviation of durations % s" % (statistics.stdev(durations)))
print("total duration : ", sum(durations))
print("DONE")
if mode == "metadata":
print("Gathering metadata")
dbfs, durations = gather_metadata(path_list[:], num_workers)
print("min duration : ", min(durations))
print("max duration : ", max(durations))
print("avg duration : ", sum(durations) / len(durations))
print("total duration : ", sum(durations))
print("Standard Deviation of sample is % s" % (statistics.stdev(durations)))
print("DONE")
# pd.DataFrame({'dBFS': dbfs, 'duration': durations, 'files': [i[1] for i in files]}).to_csv("meta.csv", index=False)
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description="Audio processing script for preprocessing and metadata gathering",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
# Preprocess audio files from a text file
python convert_factorize.py --input paths.txt --mode preprocess
# Gather metadata from audio files in current directory
python convert_factorize.py --mode metadata
# Preprocess audio files found recursively in current directory
python convert_factorize.py --mode preprocess
"""
)
parser.add_argument(
"--input", "-i",
type=str,
help="Path to text file containing audio file paths in 'language|abspath|text' format"
)
parser.add_argument(
"--mode", "-m",
type=str,
choices=["preprocess", "metadata"],
required=True,
help="Processing mode: 'preprocess' to process audio files, 'metadata' to gather statistics"
)
parser.add_argument(
"--workers", "-w",
type=int,
default=4,
help="Number of worker processes for parallel processing (default: 4)"
)
args = parser.parse_args()
data_dir = "/".join(args.input.split("/")[:-1])
print(f"Data directory: {data_dir}")
print(f"Input file: {args.input}")
print(f"Mode: {args.mode}")
print(f"Number of workers: {args.workers}")
process_audio_data(args.input, args.mode, args.workers, data_dir)
|