def CopyFileIfExists(utt_suffix, filename, input_dir, output_dir):
if os.path.isfile(input_dir + "/" + filename):
dict = ParseFileToDict(input_dir + "/" + filename,
value_processor=lambda x: " ".join(x))
if len(utt_suffix) > 0:
new_dict = {}
for key in dict.keys():
new_dict[key + "-" + utt_suffix] = dict[key]
dict = new_dict
WriteDictToFile(dict, output_dir + "/" + filename)
def CreateCorruptedUtt2uniq(input_dir, output_dir, prefix):
corrupted_utt2uniq = {}
# Parse the utt2spk to get the utterance id
utt2spk = ParseFileToDict(input_dir + "/utt2spk", value_processor = lambda x: " ".join(x))
keys = sorted(utt2spk.keys())
for utt_id in keys:
new_utt_id = GetNewId(utt_id, prefix)
corrupted_utt2uniq[new_utt_id] = utt_id
WriteDictToFile(corrupted_utt2uniq, output_dir + "/utt2uniq")
def main():
args = GetArgs()
fg_snrs = map(int, args.fg_snr_str.split(":"))
bg_snrs = map(int, args.bg_snr_str.split(":"))
input_dir = args.input_dir
output_dir = args.output_dir
num_bg_noises = map(int, args.num_bg_noises.split(":"))
reco2dur = ParseFileToDict(input_dir + "/reco2dur",
value_processor=lambda x: float(x[0]))
wav_scp_file = open(input_dir + "/wav.scp", 'r').readlines()
noise_wavs = {}
noise_reco2dur = {}
bg_noise_utts = []
fg_noise_utts = []
# Load background noises
if args.bg_noise_dir:
bg_noise_wav_filename = args.bg_noise_dir + "/wav.scp"
bg_noise_utts, bg_noise_wavs = GetNoiseList(bg_noise_wav_filename)
bg_noise_reco2dur = ParseFileToDict(
args.bg_noise_dir + "/reco2dur",
value_processor=lambda x: float(x[0]))
noise_wavs.update(bg_noise_wavs)
noise_reco2dur.update(bg_noise_reco2dur)
# Load background noises
if args.fg_noise_dir:
fg_noise_wav_filename = args.fg_noise_dir + "/wav.scp"
fg_noise_reco2dur_filename = args.fg_noise_dir + "/reco2dur"
fg_noise_utts, fg_noise_wavs = GetNoiseList(fg_noise_wav_filename)
fg_noise_reco2dur = ParseFileToDict(
args.fg_noise_dir + "/reco2dur",
value_processor=lambda x: float(x[0]))
noise_wavs.update(fg_noise_wavs)
noise_reco2dur.update(fg_noise_reco2dur)
random.seed(args.random_seed)
new_utt2wav = {}
new_utt2spk = {}
# Augment each line in the wav file
for line in wav_scp_file:
toks = line.rstrip().split(" ")
utt = toks[0]
wav = " ".join(toks[1:])
dur = reco2dur[utt]
new_wav = AugmentWav(utt, wav, dur, fg_snrs, bg_snrs, fg_noise_utts,
bg_noise_utts, noise_wavs, noise_reco2dur,
args.fg_interval, num_bg_noises)
new_utt = utt + "-" + args.utt_suffix
new_utt2wav[new_utt] = new_wav
if not os.path.exists(output_dir):
os.makedirs(output_dir)
WriteDictToFile(new_utt2wav, output_dir + "/wav.scp")
CopyFileIfExists(args.utt_suffix, "utt2spk", input_dir, output_dir)
CopyFileIfExists(args.utt_suffix, "utt2lang", input_dir, output_dir)
CopyFileIfExists(args.utt_suffix, "text", input_dir, output_dir)
CopyFileIfExists(args.utt_suffix, "utt2spk", input_dir, output_dir)
CopyFileIfExists(args.utt_suffix, "vad.scp", input_dir, output_dir)
CopyFileIfExists("", "spk2gender", input_dir, output_dir)
data_lib.RunKaldiCommand(
"utils/fix_data_dir.sh {output_dir}".format(output_dir=output_dir))
def main():
args = GetArgs()
fg_snrs = list(map(int, args.fg_snr_str.split(":")))
bg_snrs = list(map(int, args.bg_snr_str.split(":")))
input_dir = args.input_dir
output_dir = args.output_dir
num_bg_noises = list(map(int, args.num_bg_noises.split(":")))
reco2dur = ParseFileToDict(input_dir + "/reco2dur",
value_processor = lambda x: float(x[0]))
wav_scp_file = open(input_dir + "/wav.scp", 'r').readlines()
noise_wavs = {}
noise_reco2dur = {}
bg_noise_utts = []
fg_noise_utts = []
# Load background noises
if args.bg_noise_dir:
bg_noise_wav_filename = args.bg_noise_dir + "/wav.scp"
bg_noise_utts, bg_noise_wavs = GetNoiseList(bg_noise_wav_filename)
bg_noise_reco2dur = ParseFileToDict(args.bg_noise_dir + "/reco2dur",
value_processor = lambda x: float(x[0]))
noise_wavs.update(bg_noise_wavs)
noise_reco2dur.update(bg_noise_reco2dur)
# Load background noises
if args.fg_noise_dir:
fg_noise_wav_filename = args.fg_noise_dir + "/wav.scp"
fg_noise_reco2dur_filename = args.fg_noise_dir + "/reco2dur"
fg_noise_utts, fg_noise_wavs = GetNoiseList(fg_noise_wav_filename)
fg_noise_reco2dur = ParseFileToDict(args.fg_noise_dir + "/reco2dur",
value_processor = lambda x: float(x[0]))
noise_wavs.update(fg_noise_wavs)
noise_reco2dur.update(fg_noise_reco2dur)
random.seed(args.random_seed)
new_utt2wav = {}
new_utt2spk = {}
# Augment each line in the wav file
for line in wav_scp_file:
toks = line.rstrip().split(" ")
utt = toks[0]
wav = " ".join(toks[1:])
dur = reco2dur[utt]
new_wav = AugmentWav(utt, wav, dur, fg_snrs, bg_snrs, fg_noise_utts,
bg_noise_utts, noise_wavs, noise_reco2dur, args.fg_interval,
num_bg_noises)
new_utt = GetNewId(utt, args.utt_prefix)
new_utt2wav[new_utt] = new_wav
if not os.path.exists(output_dir):
os.makedirs(output_dir)
WriteDictToFile(new_utt2wav, output_dir + "/wav.scp")
AddPrefixToFields(input_dir + "/utt2spk", output_dir + "/utt2spk", args.utt_prefix, field = [0, 1])
data_lib.RunKaldiCommand("utils/utt2spk_to_spk2utt.pl <{output_dir}/utt2spk >{output_dir}/spk2utt"
.format(output_dir = output_dir))
if os.path.isfile(input_dir + "/utt2uniq"):
AddPrefixToFields(input_dir + "/utt2uniq", output_dir + "/utt2uniq", args.utt_prefix, field =[0])
else:
# Create the utt2uniq file
CreateCorruptedUtt2uniq(input_dir, output_dir, args.utt_prefix)
if os.path.isfile(input_dir + "/text"):
AddPrefixToFields(input_dir + "/text", output_dir + "/text", args.utt_prefix, field =[0])
if os.path.isfile(input_dir + "/segments"):
AddPrefixToFields(input_dir + "/segments", output_dir + "/segments", args.utt_prefix, field = [0,1])
if os.path.isfile(input_dir + "/reco2file_and_channel"):
AddPrefixToFields(input_dir + "/reco2file_and_channel", output_dir + "/reco2file_and_channel", args.utt_prefix, field = [0,1])
data_lib.RunKaldiCommand("utils/fix_data_dir.sh {output_dir}"
.format(output_dir = output_dir))
data_lib.RunKaldiCommand("utils/validate_data_dir.sh --no-feats {output_dir}"
.format(output_dir = output_dir))
sys.exit()