File size: 4,102 Bytes
c22b544 | 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 | """
Zero out the speech channel (channel 0) from all WAV files in audio_mixtures_old_both
and strip speech from all JSON command variant targets.
Usage:
python scripts/zero_out_speech.py \
--src data/audio_mixtures_old_both \
--dst data/audio_mixtures_no_speech \
--splits train test test_700
"""
import argparse
import json
import os
import glob
from pathlib import Path
import torch
import torchaudio
def process_json(metadata):
"""Strip speech from all command variant targets. Returns count of empty-target variants."""
empty_count = 0
# List format (train/test)
if "command_variants" in metadata:
for variant in metadata["command_variants"]:
variant["target_sources"] = [s for s in variant["target_sources"] if s != "speech"]
variant["target_channels"] = [c for c in variant["target_channels"] if c != 0]
if len(variant["target_sources"]) == 0:
empty_count += 1
# Singular dict format (test_700 pre-computed)
if "command_variant" in metadata and isinstance(metadata["command_variant"], dict):
cv = metadata["command_variant"]
cv["target_sources"] = [s for s in cv["target_sources"] if s != "speech"]
cv["target_channels"] = [c for c in cv["target_channels"] if c != 0]
if len(cv["target_sources"]) == 0:
empty_count += 1
return empty_count
def process_split(src_dir, dst_dir):
"""Process all WAV/JSON pairs in a split directory."""
os.makedirs(dst_dir, exist_ok=True)
wav_files = sorted(glob.glob(os.path.join(src_dir, "*.wav")))
total_files = 0
total_empty_variants = 0
for wav_path in wav_files:
basename = os.path.splitext(os.path.basename(wav_path))[0]
json_path = os.path.join(src_dir, basename + ".json")
dst_wav = os.path.join(dst_dir, basename + ".wav")
dst_json = os.path.join(dst_dir, basename + ".json")
# --- WAV: zero out channel 0 (speech) ---
audio, sr = torchaudio.load(wav_path) # (5, T)
audio[0, :] = 0.0
torchaudio.save(dst_wav, audio, sr)
# --- JSON: strip speech from targets ---
if os.path.exists(json_path):
with open(json_path, "r") as f:
metadata = json.load(f)
empty_count = process_json(metadata)
total_empty_variants += empty_count
with open(dst_json, "w") as f:
json.dump(metadata, f, indent=2)
total_files += 1
if total_files % 500 == 0:
print(f" Processed {total_files}/{len(wav_files)} files...")
return total_files, total_empty_variants
def main():
parser = argparse.ArgumentParser(description="Zero out speech channel from audio_mixtures dataset")
parser.add_argument("--src", required=True, help="Source audio_mixtures directory")
parser.add_argument("--dst", required=True, help="Destination directory")
parser.add_argument("--splits", nargs="+", default=["train", "test", "test_700"],
help="Split directories to process")
args = parser.parse_args()
print(f"Source: {args.src}")
print(f"Destination: {args.dst}")
print(f"Splits: {args.splits}")
print()
os.makedirs(args.dst, exist_ok=True)
grand_total_files = 0
grand_total_empty = 0
for split in args.splits:
src_split = os.path.join(args.src, split)
dst_split = os.path.join(args.dst, split)
if not os.path.isdir(src_split):
print(f"Skipping {split}/ (not found)")
continue
print(f"Processing {split}/...")
n_files, n_empty = process_split(src_split, dst_split)
grand_total_files += n_files
grand_total_empty += n_empty
print(f" Done: {n_files} files, {n_empty} empty-target variants (noise-cancelling)")
print()
print("=" * 50)
print(f"Total files processed: {grand_total_files}")
print(f"Total empty-target variants: {grand_total_empty}")
print(f"Output at: {args.dst}")
if __name__ == "__main__":
main()
|