def main_():
ap = argparse.ArgumentParser()
ap.add_argument("--hop_size", type=int, default=110)
ap.add_argument("--fft_size", type=int, default=2048)
ap.add_argument("wav_list", type=str)
ap.add_argument("ali_rspecifier", type=str)
args = ap.parse_args()
wavlist_fp = open(args.wav_list,
mode='rt') if args.wav_list != "-" else sys.stdin
with kaldi_io.open_or_fd(args.ali_rspecifier, mode="wb") as ali_writer:
for line in wavlist_fp:
s = line.strip().replace("\t", " ").split(" ")
assert len(s) == 2
utt, wav_file = s
rate, wav = read(wav_file)
wav = wav.astype("float")
voiced = np.zeros_like(wav, dtype=np.int32)
ans = split(wav,
frame_length=args.fft_size,
hop_length=args.hop_size)
for a in ans:
sidx, eidx = a
voiced[sidx:eidx] = 1
kaldi_io.write_vec_int(ali_writer, voiced, key=utt)
wavlist_fp.close()
def main_():
ap = ArgumentParser()
ap.add_argument("feat_rspecifier", type=str)
ap.add_argument("ali_wspecifier", type=str)
args = ap.parse_args()
with kaldi_io.open_or_fd(args.ali_wspecifier, mode="wb") as ali_writer:
for utt, feat in kaldi_io.read_mat_ark(args.feat_rspecifier):
T, dim = feat.shape
assert (dim == 1)
ali = feat.squeeze(1).astype(np.int)
kaldi_io.write_vec_int(ali_writer, ali, key=utt)
def main_():
ap = argparse.ArgumentParser()
ap.add_argument("npy_list", type=str)
ap.add_argument("ali_wspecifier", type=str)
args = ap.parse_args()
npy_fp = open(args.npy_list, mode='rt') if args.npy_list != "-" else sys.stdin
with kaldi_io.open_or_fd(args.ali_wspecifier, "wb") as ali_writer:
for line in npy_fp:
s = line.strip().replace("\t", " ").split(" ")
assert len(s) == 2
utt, npy_file = s
data = np.load(npy_file)
assert len(data.shape) == 1
kaldi_io.write_vec_int(ali_writer, data, key=utt)
npy_fp.close()
def main_():
ap = argparse.ArgumentParser(usage="")
# positional args
ap.add_argument("--write_int", type=str_to_bool, default=True)
ap.add_argument("--thresh", type=float, default=0.5)
ap.add_argument("feature_rspecifier", help="input feat")
ap.add_argument("feature_wspecifier", help="output feat")
args = ap.parse_args()
with kaldi_io.open_or_fd(args.feature_wspecifier, "wb") as feature_writer:
for key, feats in kaldi_io.read_mat_ark(args.feature_rspecifier):
new_feats = np.where(feats>=args.thresh, 1, feats)
new_feats = np.where(feats<args.thresh, 0, new_feats)
if args.write_int:
assert (new_feats.shape[1] == 1)
kaldi_io.write_vec_int(feature_writer, new_feats.astype("int32").reshape(new_feats.size), key=key)
else:
kaldi_io.write_mat(feature_writer, new_feats, key=key)
def main_():
ap = argparse.ArgumentParser()
ap.add_argument("wav_list", type=str)
ap.add_argument("ali_rspecifier", type=str)
args = ap.parse_args()
wavlist_fp = open(args.wav_list,
mode='rt') if args.wav_list != "-" else sys.stdin
with kaldi_io.open_or_fd(args.ali_rspecifier, mode="wb") as ali_writer:
for line in wavlist_fp:
s = line.strip().replace("\t", " ").split(" ")
assert len(s) == 2
utt, wav_file = s
with wave.open(wav_file, 'r') as wr:
data = wr.readframes(wr.getnframes())
wav = np.frombuffer(data, dtype=np.int16)
kaldi_io.write_vec_int(ali_writer, wav, key=utt)
wavlist_fp.close()
def write_(writer, mat, utt, *, is_int=False):
if is_int:
kaldi_io.write_vec_int(writer, mat, key=utt)
else:
kaldi_io.write_mat(writer, mat, key=utt)
with open(test_list, 'r') as f:
for line in f.readlines():
key = line.strip()
file_name = key + '.wav'
mix_file = os.path.join(wav_dir, 'mix', file_name)
s1_file = os.path.join(wav_dir, 's1', file_name)
s2_file = os.path.join(wav_dir, 's2', file_name)
mix_wav = audioread(mix_file, samp_rate=8000)
s1_wav = audioread(s1_file, samp_rate=8000)
s2_wav = audioread(s2_file, samp_rate=8000)
s1_gender = gender_dict[key.split('_')[0][0:3]]
s2_gender = gender_dict[key.split('_')[2][0:3]]
gender = np.array([s1_gender, s2_gender]).astype(np.int32)
kaldi_io.write_vec_int(gender_f, gender, key=key)
mix_stft = stft(mix_wav, size=256, shift=64)
mix_abs = np.abs(mix_stft)
mix_angle = np.angle(mix_stft)
s1_stft = stft(s1_wav, size=256, shift=64)
s1_abs = np.abs(s1_stft)
s1_angle = np.angle(s1_stft)
s2_stft = stft(s2_wav, size=256, shift=64)
s2_abs = np.abs(s2_stft)
s2_angle = np.angle(s2_stft)
mix_data = np.concatenate((mix_abs, mix_angle), axis=1)
s1_data = np.concatenate((s1_abs, s1_angle), axis=1)
s2_data = np.concatenate((s2_abs, s2_angle), axis=1)
feats = np.concatenate((mix_data, s1_data, s2_data),
mix_file = os.path.join(wav_dir, i_type, 'mix', key + '.wav')
s1_file = os.path.join(wav_dir, i_type, 's1', key + '.wav')
s2_file = os.path.join(wav_dir, i_type, 's2', key + '.wav')
s3_file = os.path.join(wav_dir, i_type, 's3', key + '.wav')
mix_wav = audioread(mix_file, samp_rate=fs8k)
s1_wav = audioread(s1_file, samp_rate=fs8k)
s2_wav = audioread(s2_file, samp_rate=fs8k)
s3_wav = audioread(s3_file, samp_rate=fs8k)
s1_gender = gender_dict[key.split('_')[0][0:3]]
s2_gender = gender_dict[key.split('_')[2][0:3]]
s3_gender = gender_dict[key.split('_')[4][0:3]]
gender = np.array([s1_gender, s2_gender,
s3_gender]).astype(np.int32)
write_vec_int(write_gf, gender, key=key)
mix_stft = stft(mix_wav, size=size,
shift=shift).astype(np.complex64)
mix_abs = np.abs(mix_stft)
mix_angle = np.angle(mix_stft)
s1_stft = stft(s1_wav, size=size, shift=shift).astype(np.complex64)
s1_abs = np.abs(s1_stft)
s1_angle = np.angle(s1_stft)
s2_stft = stft(s2_wav, size=size, shift=shift).astype(np.complex64)
s2_abs = np.abs(s2_stft)
s2_angle = np.angle(s2_stft)
s3_stft = stft(s3_wav, size=size, shift=shift).astype(np.complex64)
s3_abs = np.abs(s3_stft)
s3_angle = np.angle(s3_stft)