out_train_folder = osp.join(base_folder, 'audio', 'train', 'synthetic')
if args.save_jams:
out_folder_jams_train = out_train_folder
else:
out_folder_jams_train = None
out_train_tsv = osp.join(base_folder, 'metadata', 'train', 'soundscapes',
'synthetic.tsv')
list_jams_train = glob.glob(osp.join(train_folder, "*.jams"))
fg_path_train = osp.join(base_folder, "audio", "train", "soundbank",
"foreground")
bg_path_train = osp.join(base_folder, 'audio', "train", "soundbank",
"background")
generate_files_from_jams(list_jams_train,
out_train_folder,
fg_path=fg_path_train,
bg_path=bg_path_train,
out_folder_jams=out_folder_jams_train)
generate_tsv_from_jams(list_jams_train, out_train_tsv)
# ########
# Eval
# ########
# In the evaluation part, there multiple subsets which allows to check robustness of systems
eval_folder = osp.join(base_folder, 'metadata', 'eval', "soundscapes")
list_folders = [
osp.join(eval_folder, dI) for dI in os.listdir(eval_folder)
if osp.isdir(osp.join(eval_folder, dI))
]
fg_path_eval = osp.join(base_folder, "audio", "eval", "soundbank",
"foreground")
def test_generate_files_from_jams():
generate_files_from_jams(
[os.path.join(absolute_dir_path, "material", "5.jams")],
os.path.join(absolute_dir_path, "generated", "generated_from_jams"))
post_process_txt_labels(out_folder_30,
output_folder=out_folder_30,
output_tsv=out_tsv)
# ###
# Generate variation of the dataset by varying the FBSNR
# ###
# We create the same dataset with different background SNR
# The foreground background snr ratio is between 0dB and 24dB (see utils snr_min (6dB) and snr_max (30dB))
out_folder_24 = osp.join(out_folder, "24dB")
create_folder(out_folder_24)
jams_to_modify = glob.glob(osp.join(out_folder_30, "*.jams"))
# Put FBSNR to [0;24] range so reducing of 6 compared to [6;30]
minus_6_snr = functools.partial(change_snr, db_change=-6)
new_jams24 = modify_jams(jams_to_modify, minus_6_snr, out_folder_24)
generate_files_from_jams(new_jams24, out_folder_24, out_folder_24)
# Same for 15
out_folder_15 = osp.join(out_folder, "15dB")
create_folder(out_folder_15)
minus_15_snr = functools.partial(change_snr, db_change=-15)
new_jams15 = modify_jams(jams_to_modify, minus_15_snr, out_folder_15)
generate_files_from_jams(new_jams15, out_folder_15, out_folder_15)
out_folder_0 = osp.join(out_folder, "0dB")
create_folder(out_folder_0)
minus_30_snr = functools.partial(change_snr, db_change=-30)
new_jams0 = modify_jams(jams_to_modify, minus_30_snr, out_folder_0)
generate_files_from_jams(new_jams0, out_folder_0, out_folder_0)
"one_event_train")
meta_eval_folder = osp.join(dataset_root, "metadata", "eval")
audio_train_folder = osp.join(dataset_root, "audio", "train",
"one_event_train")
audio_eval_folder = osp.join(dataset_root, "audio", "eval")
list_jams = glob.glob(osp.join(meta_train_folder, "*.jams"))[:10]
fg_path_train = osp.join(soundbank_root, "audio", "train", "soundbank",
"foreground")
bg_path_train = osp.join(soundbank_root, "audio", "train", "soundbank",
"background")
generate_files_from_jams(list_jams,
audio_train_folder,
fg_path=fg_path_train,
bg_path=bg_path_train,
overwrite_jams=args.overwrite_jams)
# Eval
# In the evaluation part, there multiple subsets which allows to check robustness of systems
list_folders = [
osp.join(meta_eval_folder, dI) for dI in os.listdir(meta_eval_folder)
if osp.isdir(osp.join(meta_eval_folder, dI))
]
fg_path_eval = osp.join(soundbank_root, "audio", "eval", "soundbank",
"foreground")
bg_path_eval = osp.join(soundbank_root, "audio", "eval", "soundbank",
"background")
for folder in list_folders:
soundbank_dir = args.soundbank
fg_path_train = osp.join(soundbank_dir, "foreground")
bg_path_train = osp.join(soundbank_dir, "background")
out_audio_dir = args.out_audio_dir
if out_audio_dir is None:
out_audio_dir = jams_folder
if args.save_jams:
out_folder_jams = out_audio_dir
else:
out_folder_jams = None
out_tsv = args.out_tsv
if out_tsv is None:
out_tsv = out_audio_dir.replace("audio", "metadata") + ".tsv"
save_isolated = args.save_isolated
# ########
# Generate
# #######
generate_files_from_jams(list_jams,
out_audio_dir,
out_folder_jams=out_folder_jams,
fg_path=fg_path_train,
bg_path=bg_path_train,
save_isolated_events=save_isolated)
if out_tsv:
generate_tsv_from_jams(list_jams, out_tsv)
# Replace duration line, needed to do the del add because of the type of data
new_obs = jams_obj["annotations"][0].data[cnt]._replace(
duration=time_trunc)
del jams_obj["annotations"][0].data[cnt]
jams_obj["annotations"][0].data.add(new_obs)
return jams_obj
if __name__ == '__main__':
base_soundbank = osp.join("..", "..", "synthetic")
for subset in ["train", "eval"]:
soundbank_path = osp.join(base_soundbank, "audio", subset, "soundbank")
fg_folder = osp.join(soundbank_path, "foreground/")
bg_folder = osp.join(soundbank_path, "background")
dataset_path = osp.join("..", "..", "dataset")
jams_folder = osp.join(dataset_path, "audio", subset,
f"one_event_generated")
outfolder = osp.join(dataset_path, "audio", subset,
f"one_event_generated_200ms")
outtsv = osp.join(dataset_path, "metadata", subset,
f"one_event_generated_200ms.tsv")
list_jams = glob.glob(os.path.join(jams_folder, "*.jams"))
new_jams = modify_jams(list_jams, get_part_audio, outfolder)
generate_files_from_jams(new_jams, outfolder)
generate_tsv_from_jams(new_jams, outtsv)