def export2kaldi(databyid):
kaldidata = databyid
filter_field = ('Click1', 'Hush', 'Click2', 'Click3') #, 'LongPress')
#exclude_field = ('LongPress')
test_size = 0.2
trainset = []
testset = []
filtered_kaldidata = {}
for k, v in kaldidata.items():
label = k.split('_')[0]
if label in filter_field:
#if label not in exclude_field:
if label not in filtered_kaldidata:
filtered_kaldidata[label] = []
filtered_kaldidata[label].append((k, v))
for label in filtered_kaldidata:
length = len(filtered_kaldidata[label])
random.shuffle(filtered_kaldidata[label])
testset.extend(filtered_kaldidata[label][:int(length * test_size)])
trainset.extend(filtered_kaldidata[label][int(length * test_size):])
with kaldi_io.open_or_fd('feats_train.ark', 'wb') as f:
for k, m in trainset:
kaldi_io.write_mat(f, m, k)
with kaldi_io.open_or_fd('feats_test.ark', 'wb') as f:
for k, m in testset:
kaldi_io.write_mat(f, m, k)
def make_feature(wav_path_list,
noise_wav_list,
feat_dir,
thread_num,
argument=False,
repeat_num=1):
mag_ark_scp_output = 'ark:| copy-feats --compress=true ark:- ark,scp:{0}/feats{1}.ark,{0}/feats{1}.scp'.format(
feat_dir, thread_num)
ang_ark_scp_output = 'ark:| copy-feats --compress=true ark:- ark,scp:{0}/angles{1}.ark,{0}/angles{1}.scp'.format(
feat_dir, thread_num)
if argument:
fwrite = open(os.path.join(feat_dir, 'db' + str(thread_num)), 'a')
f_mag = kaldi_io.open_or_fd(mag_ark_scp_output, 'wb')
f_ang = kaldi_io.open_or_fd(ang_ark_scp_output, 'wb')
print("进入num循环")
for num in range(repeat_num):
for tmp in wav_path_list:
uttid, wav_path = tmp
clean = load_audio(wav_path)
y = None
print("argument = ", argument)
while y is None:
if argument:
print("argument=True")
noise_path = choice(noise_wav_list)
n = load_audio(noise_path[0])
db = np.random.uniform(low=0, high=20)
y = MakeMixture(clean, n, db)
uttid_new = uttid + '__mix{}'.format(num)
print(uttid_new + ' ' + str(db) + '\n')
fwrite.write(uttid_new + ' ' + str(db) + '\n')
else:
y = clean
uttid_new = uttid
# STFT
print("y = ", y)
if y is not None:
D = librosa.stft(y,
n_fft=512,
hop_length=256,
win_length=512,
window=scipy.signal.hamming)
spect = np.abs(D)
angle = np.angle(D)
print("创建STFT")
##feat = np.concatenate((spect, angle), axis=1)
##feat = feat.transpose((1, 0))
kaldi_io.write_mat(f_mag,
spect.transpose((1, 0)),
key=uttid_new)
kaldi_io.write_mat(f_ang,
angle.transpose((1, 0)),
key=uttid_new)
else:
print(noise_path, tmp, 'error')
if argument:
fwrite.close()
def testFloatVectorReadWrite(self):
"""
Test read/write for float vectors.
"""
# read,
flt_vec = {
k: v
for k, v in kaldi_io.read_vec_flt_scp('tests/data/conf.scp')
} # scp,
return
flt_vec2 = {
k: v
for k, v in kaldi_io.read_vec_flt_ark('tests/data/conf.ark')
} # binary-ark,
flt_vec3 = {
k: v
for k, v in kaldi_io.read_vec_flt_ark('tests/data/conf_ascii.ark')
} # ascii-ark,
# store,
with kaldi_io.open_or_fd('tests/data_re-saved/conf.ark', 'wb') as f:
for k, v in flt_vec.items():
kaldi_io.write_vec_flt(f, v, k)
# read and compare,
for k, v in kaldi_io.read_vec_flt_ark('tests/data_re-saved/conf.ark'):
self.assertTrue(np.array_equal(v, flt_vec[k]),
msg="flt. vector same after re-saving")
def scp2dict(ipath2scp):
fd = kaldi_io.open_or_fd(ipath2scp) # ipath2scp can be a pipeline
id2path = {}
for line in fd:
(id, path) = line.decode("utf-8").rstrip().split(' ', 1)
id2path[id] = path
return id2path
def testMatrixReadWrite(self):
"""
Test read/write for float matrices.
"""
# read,
flt_mat = {
k: m
for k, m in kaldi_io.read_mat_scp('tests/data/feats_ascii.scp')
} # ascii-scp,
flt_mat2 = {
k: m
for k, m in kaldi_io.read_mat_ark('tests/data/feats_ascii.ark')
} # ascii-ark,
flt_mat3 = {
k: m
for k, m in kaldi_io.read_mat_ark('tests/data/feats.ark')
} # ascii-ark,
# store,
with kaldi_io.open_or_fd('tests/data_re-saved/mat.ark', 'wb') as f:
for k, m in flt_mat3.items():
kaldi_io.write_mat(f, m, k)
# read and compare,
for k, m in kaldi_io.read_mat_ark('tests/data_re-saved/mat.ark'):
self.assertTrue(np.array_equal(m, flt_mat3[k]),
msg="flt. matrix same after re-saving")
def read_plda(file_or_fd):
""" Loads PLDA from a file in kaldi format (binary or text).
Input:
file_or_fd - file name or file handle with kaldi PLDA model.
Output:
Tuple (mu, tr, psi) defining a PLDA model using the kaldi parametrization:
mu - mean vector
tr - transform whitening within- and diagonalizing across-class covariance matrix
psi - diagonal of the across-class covariance in the transformed space
"""
fd = open_or_fd(file_or_fd)
try:
binary = fd.read(2)
if binary == b'\x00B':
assert (fd.read(7) == b'<Plda> ')
plda_mean = _read_vec_binary(fd)
plda_trans = _read_mat_binary(fd)
plda_psi = _read_vec_binary(fd)
else:
assert (binary + fd.read(5) == b'<Plda> ')
plda_mean = np.array(fd.readline().strip(' \n[]').split(),
dtype=float)
assert (fd.read(2) == b' [')
plda_trans = _read_mat_ascii(fd)
plda_psi = np.array(fd.readline().strip(' \n[]').split(),
dtype=float)
assert (fd.read(8) == b'</Plda> ')
finally:
if fd is not file_or_fd:
fd.close()
return plda_mean, plda_trans, plda_psi
def main():
print("start time:",
time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(time.time())))
model = load_model()
ext = ExtractEmbedding(model, "layer4")
dir = "/home/work_nfs4_ssd/hzhao/feature/voxceleb1/test/feats.scp"
dataset = getTestData(dir)
batch_size = 256
test_loader = DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=False)
for i, trainset in enumerate(test_loader):
(input_x, target_var) = trainset
input_x = input_x.cuda()
#print(input_x.size(),type(input_x))
output = ext(input_x)
output = output.squeeze()
output = output.cpu()
#tensor to numpy
output = output.detach().numpy()
#print(":::",len(target_var))
target_var = np.squeeze(target_var)
#tmp="/home/work_nfs/lizhang/node6/kaldi02/kaldi02/egs/voxceleb/v2/tmp"
filename = "./test/99_test/enroll/xvector." + str(i) + ".ark"
f = kaldi_io.open_or_fd(filename, "wb")
for i, uttid in enumerate(target_var):
kaldi_io.write_vec_flt(f, output[i], key=uttid)
print("end time:",
time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(time.time())))
def fetch_audio_segment(wavid, ipath2wav_scp, seg=(0.0, math.inf), fs=16000):
"""
given wavid, return an audio segment from ipath2wav_scp
args: wavid -- string, id of a audio file
ipath2wav_scp -- the path to wav.scp
(wav.scp has the same format as kaldi's:
each row of the file is <wavid> <wavpath>)
seg -- a tuple of float, (start_time, end_time)
fs -- sampling frequency
return: y -- audio samples with numpy format
fs -- sampling frequency
"""
fd = kaldi_io.open_or_fd(ipath2wav_scp)
for line in fd:
(wid, path) = line.decode("utf-8").rstrip().split(' ', 1)
if wavid == wid:
y, fs = kaldi_read_wav(path)
start_t, end_t = seg
end_t = min(end_t,
y.shape[0] / fs) # the second term is float by default
assert start_t < end_t and start_t >= 0.0, "InputArg: seg {0} invalid".format(
str(seg))
# return the segment and fs
return y[int(start_t * fs):int(end_t * fs)], fs
# wavid not found
raise Exception("wavid: {0} not found in file {1}".format(
wavid, ipath2wav_scp))
def forward_dct(args,
cpc_model,
device,
data_loader,
output_ark,
output_scp,
dct_dim=24):
''' forward with dct '''
logger.info("Starting Forward Passing")
cpc_model.eval() # not training cdc model
ark_scp_output = 'ark:| copy-feats --compress=true ark:- ark,scp:' + output_ark + ',' + output_scp
with torch.no_grad():
with ko.open_or_fd(ark_scp_output, 'wb') as f:
for [utt_id, data] in data_loader:
data = data.float().unsqueeze(1).to(
device) # add channel dimension
data = data.contiguous()
hidden = cpc_model.init_hidden(len(data))
output, hidden = cpc_model.predict(data, hidden)
mat = output.squeeze(
0).cpu().numpy() # kaldi io does not accept torch tensor
dct_mat = fft.dct(mat, type=2, n=dct_dim) # apply dct
ko.write_mat(f, dct_mat, key=utt_id[0])
def fetch_llkprob_segment(wavid,
ipath2prob_scp,
seg=(0.0, math.inf),
win_len=0.025,
hop_len=0.010):
"""
given wavid, return an loglikehood probability segment from ipath2prob_scp
args: wavid -- string, id of a audio file
ipath2prob_scp -- the path to llk_prob.scp
each wavid corresponds to a float vector of llk_prob
llk_prob: the prob of a specific GMM generating a frame
seg -- a tuple of (start_time, end_time)
win_len -- window length in second
hop_len -- window shift in second
return: vec -- llk_prob curve with numpy format
"""
fd = kaldi_io.open_or_fd(ipath2prob_scp)
for line in fd:
(wid, path) = line.decode("utf-8").rstrip().split(' ', 1)
if wavid == wid:
vec = kaldi_io.read_vec_flt(path) # np.array
start_t, end_t = seg
end_t = min(end_t, vec.shape[0] *
hop_len) # the second term is float by default
assert start_t < end_t and start_t >= 0.0, "InputArg: seg {0} invalid".format(
str(seg))
start_f = int(start_t / hop_len)
end_f = int(end_t / hop_len)
return vec[start_f:end_f]
def dvec_compute(generator,
ds_eval,
device,
num_jobs=20,
outfolder='./exp/example_dvecs'):
# naively compute the embeddings for each window
# ds_len = len(ds_feats)
all_utts = ds_eval.all_utts
ds_len = len(all_utts)
indices = np.arange(ds_len)
job_split = np.array_split(indices, num_jobs)
generator.eval().to(device)
for job_num, job in enumerate(tqdm(job_split)):
print('Starting job {}'.format(job_num))
ark_scp_output = 'ark:| copy-vector ark:- ark,scp:{0}/xvector.{1}.ark,{0}/xvector.{1}.scp'.format(
outfolder, job_num + 1)
job_utts = all_utts[job]
job_feats = ds_eval.get_batches(job_utts)
job_feats = mtd(job_feats, device)
with torch.no_grad():
job_embeds = torch.cat([
generator(x.unsqueeze(0)) for x in tqdm(job_feats)
]).cpu().numpy()
with kaldi_io.open_or_fd(ark_scp_output, 'wb') as f:
for xvec, key in zip(job_embeds, job_utts):
kaldi_io.write_vec_flt(f, xvec, key=key)
def main():
logging.basicConfig(format='[%(filename)s:%(lineno)d] %(message)s',
level=logging.WARN)
parser = argparse.ArgumentParser(
description="load .rec rawfile from first argument SCPPATH, \
and convert all rec to scpfile-contained-utterence-id-indexed ark file to ARKPATH"
)
parser.add_argument("SCPPATH", help="scp file path")
parser.add_argument("ARKPATH", help="ark file path")
parser.add_argument("OUTPUTTYPE", help="raw | diff | baseline")
args = parser.parse_args()
logging.debug(args)
SCP_FILEPATH = args.SCPPATH
ARK_FILEPATH = args.ARKPATH
OUTPUT_TYPE = args.OUTPUTTYPE
if not ARK_FILEPATH.endswith('.ark') or not SCP_FILEPATH.endswith('.scp'):
logging.error('extension error')
exit()
logging.debug(SCP_FILEPATH)
uttid2recpath = io_helper.loadscp(SCP_FILEPATH)
logging.debug(uttid2recpath)
dataset = loaddata(uttid2recpath, OUTPUT_TYPE)
with kaldi_io.open_or_fd(ARK_FILEPATH, 'wb') as f:
for k, m in dataset.items():
kaldi_io.write_mat(f, m, k)
def compute_plda_score(enr_embeds,
test_embeds,
plda_dir,
all_pair=True,
mean=False):
os.environ["KALDI_ROOT"] = "/host/projects/kaldi"
if mean:
enr_keys = ['enr']
n_uttrs = len(enr_embeds)
enr_embeds = enr_embeds.mean(0, keepdims=True)
else:
enr_keys = ['enr_{}'.format(i) for i in range(len(enr_embeds))]
test_keys = ['test_{}'.format(i) for i in range(len(test_embeds))]
keys = enr_keys + test_keys
embeds = np.concatenate([enr_embeds, test_embeds])
# write trials
score_dir = TemporaryDirectory()
with open("{}/num_utts.ark".format(score_dir.name), "w") as f:
if mean:
f.write("enr {}\n".format(n_uttrs))
else:
for key in enr_keys:
f.write("{} {}\n".format(key, 1))
with open("{}/kaldi_trial".format(score_dir.name), "w") as f:
if all_pair:
trial_pairs = itertools.product(enr_keys, test_keys)
else:
trial_pairs = zip(enr_keys, test_keys)
for pair in trial_pairs:
f.write(" ".join(pair))
f.write("\n")
# write feat
ark_scp_output='ark:| copy-vector ark:- ' +\
'ark,scp:{output}/feats.ark,{output}/feats.scp'.format(output=score_dir.name)
with kaldi_io.open_or_fd(ark_scp_output, "wb") as f:
for key, vec in zip(keys, embeds):
kaldi_io.write_vec_flt(f, vec.squeeze(), key=str(key))
# call scoring
ret = subprocess.call(["./plda_score.sh", plda_dir, score_dir.name])
if ret != 0:
print("plda scoring fails")
raise ValueError
# read plda scores
plda_scores = pd.read_csv("{}/plda_scores".format(score_dir.name),
delimiter=" ",
names=["enroll", "test", "score"]).score
plda_scores = np.array(plda_scores)
plda_scores = plda_scores.reshape(len(enr_keys), -1)
if all_pair:
assert plda_scores.shape[1] == len(test_embeds)
score_dir.cleanup()
return plda_scores
def scp2dict(ipath2scp):
fd = kaldi_io.open_or_fd(ipath2scp) # iapth2scp can be a pipeline
id2path = {}
for line in fd:
items = line.decode("utf-8").rstrip().split(' ')
id = items[0]
id2path[id] = items[1:]
return id2path
def read_wav_scp(ipath2wav_scp):
fd = kaldi_io.open_or_fd(ipath2wav_scp)
try:
for line in fd:
uttid, path2wav = line.rstrip().decode().split(' ', 1)
yield uttid, path2wav
finally:
if fd is not ipath2wav_scp: fd.close()
def kaldi_get_datas(file_path: str, targets: List[Tuple[str]],
dep: bool = False) -> Tuple[Callable, Callable, Callable, int]:
def get_src_data_factory(raw_src_datas):
def func(utt_ids):
utt_datas = []
for utt_id in utt_ids:
rx_file = raw_src_datas[utt_id]
utt_datas.append(kaldi_io.read_mat(rx_file))
source_data = np.concatenate((utt_datas), axis=0)
return source_data
return func
raw_src_datas: Dict[str, str] = {}
for line in kaldi_io.open_or_fd(file_path):
key, rxfile = line.decode().split(' ')
raw_src_datas[key] = rxfile
get_src_data = get_src_data_factory(raw_src_datas)
targets = list(targets)
instances_before_filtering = len(targets)
targets = list(filter(lambda fields: fields[2].strip(), targets))
logger.warning(f"number of empty targets: {instances_before_filtering - len(targets)}")
orders = [i for i, fields in enumerate(targets) if
all(utt_id in raw_src_datas for utt_id in fields[0])]
ignored = [fields[0] for i, fields in enumerate(targets) if
not all(utt_id in raw_src_datas for utt_id in fields[0])]
print(ignored)
dropped_instances = len(targets) - len(orders)
if not dropped_instances:
logger.info("No instances dropped from {}.".format(file_path))
else:
logger.warning("Dropped {} instances from {}.".format(dropped_instances,
file_path))
# for utt_ids, src_trns, *tgt_trns in tqdm(targets):
# utt_datas = []
# source_data = get_src_data(utt_ids)
# source_datas.append(source_data)
# target_datas.append(tgt_trns[0])
# src_lens = [src.shape[0] for src in source_datas]
# source_orders = np.argsort(src_lens)
src_data_func = data_func_factory(lambda idx: get_src_data(targets[idx][0]), orders)
tgt_data_func = data_func_factory(lambda idx: targets[idx][2], orders)
annotations = None
if dep:
annotations = read_dependencies(os.path.dirname(file_path))
anno_data_func = data_func_factory(lambda idx: annotations[idx] if annotations else None,
orders)
return src_data_func, tgt_data_func, anno_data_func, len(orders)
def __execute_command__(self, datain, cmd):
#try:
fin, fout = kio.open_or_fd(cmd, 'wb')
kio.write_wav(fin, datain, self.sr, key='utt')
fin.close() #so its clear nothing new arrives
feats_ark = kio.read_mat_ark(fout)
for _, feats in feats_ark:
fout.close()
return feats.T #there is only one to read
def construct_tensor(orig_feat_scp, ark_scp_output, tuncate_len):
with ko.open_or_fd(ark_scp_output, 'wb') as f:
for key, mat in ko.read_mat_scp(orig_feat_scp):
tensor = tensor_cnn_utt(mat, truncate_len)
repetition = int(tensor.shape[1] / truncate_len)
for i in range(repetition):
sub_tensor = tensor[:, i * truncate_len:(i + 1) * truncate_len]
new_key = key + '-' + str(i)
ko.write_mat(f, sub_tensor, key=new_key)
def decode(args, dataset, model, priors, device='cpu'):
'''
Produce lattices from the input utterances.
'''
# This is all of the kaldi code we are calling. We are just piping out
# out features to latgen-faster-mapped which does all of the lattice
# generation.
lat_output = '''ark:| copy-feats ark:- ark:- |\
latgen-faster-mapped --min-active={} --max-active={} \
--max-mem={} \
--lattice-beam={} --beam={} \
--acoustic-scale={} --allow-partial=true \
--word-symbol-table={} \
{} {} ark:- ark:- | lattice-scale --acoustic-scale={} ark:- ark:- |\
gzip -c > {}/lat.{}.gz'''.format(args.min_active, args.max_active,
args.max_mem, args.lattice_beam,
args.beam, args.acoustic_scale,
args.words_file, args.trans_mdl,
args.hclg, args.post_decode_acwt,
args.dumpdir, args.job)
# Do the decoding (dumping senone posteriors)
model.eval()
with torch.no_grad():
with kaldi_io.open_or_fd(lat_output, 'wb') as f:
utt_mat = []
prev_key = b''
generator = evaluation_batches(dataset)
# Each minibatch is guaranteed to have at most 1 utterance. We need
# to append the output of subsequent minibatches corresponding to
# the same utterances. These are stored in ``utt_mat'', which is
# just a buffer to accumulate the posterior outputs of minibatches
# corresponding to the same utterance. The posterior state
# probabilities are normalized (subtraction in log space), by the
# log priors in order to produce pseudo-likelihoods useable for
# for lattice generation with latgen-faster-mapped
for key, mat in decode_dataset(args,
generator,
model,
device='cpu',
output_idx=args.output_idx):
if len(utt_mat) > 0 and key != prev_key:
kaldi_io.write_mat(f,
np.concatenate(utt_mat,
axis=0)[:utt_length, :],
key=prev_key.decode('utf-8'))
utt_mat = []
utt_mat.append(mat - args.prior_scale * priors)
prev_key = key
utt_length = dataset.utt_lengths[key] // dataset.subsample
# Flush utt_mat buffer at the end
if len(utt_mat) > 0:
kaldi_io.write_mat(f,
np.concatenate(utt_mat,
axis=0)[:utt_length, :],
key=prev_key.decode('utf-8'))
def write_kaldi(orig_feat_scp, ark_scp_output, max_len):
"""Write the slice feature matrix to ark_scp_output
"""
with ko.open_or_fd(ark_scp_output,'wb') as f:
for key,mat in ko.read_mat_scp(orig_feat_scp):
tensor = tensor_cnn_utt(mat, max_len)
if tensor.shape[1] != max_len:
print(tensor.shape)
ko.write_mat(f, tensor, key=key)
def plda_write(self, plda):
with kaldi_io.open_or_fd(plda, 'wb') as f:
kaldi_io.write_vec_flt(f, self.mean, key='mean')
kaldi_io.write_vec_flt(f,
self.within_var.reshape(-1, 1),
key='within_var')
kaldi_io.write_vec_flt(f,
self.between_var.reshape(-1, 1),
key='between_var')
def extract_file(wav_lines, wfilename, winstep, winlen, mode):
ark_scp_output = 'ark:| copy-feats ark:- ark,scp:%s.ark,%s.scp' %(wfilename, wfilename)
with kaldi_io.open_or_fd(ark_scp_output, 'wb') as wf:
for line in wav_lines:
items = line.split()
key = items[0]
wav_ = items[5]
mat = extract(wav_, winstep, winlen, mode)
kaldi_io.write_mat(wf, mat, key=key)
def outputfile_mat2ark(mat, filepath):
logging.debug(filepath)
if not filepath.endswith('.ark'):
logging.error('extension error')
return
with kaldi_io.open_or_fd(filepath, 'wb') as f:
for k, m in mat.items():
if 0 in m.shape:
continue
kaldi_io.write_mat(f, m, k)
def eval_dnn(args):
""" The main function for doing evaluation on a trained network.
Args:
args: a Namespace object with the required parameters
obtained from the function process_args()
"""
model_dir = args.model_dir
use_gpu = args.use_gpu == 'yes'
min_chunk_size = args.min_chunk_size
chunk_size = args.chunk_size
# First change the output files temp ones and at the end rename them
wspecifier, ark, scp = process_wspecifier(args.vector_wspecifier)
if ark is not None and os.path.exists(
ark) and scp is not None and os.path.exists(scp):
logger.info(
'Both output ark and scp files exist. Return from this call.')
return
model = Model()
print(args.feature_rspecifier)
with kaldi_io.open_or_fd(args.feature_rspecifier) as input_fid:
with kaldi_io.open_or_fd(wspecifier) as output_fid:
model.make_embedding(input_fid, output_fid, model_dir,
min_chunk_size, chunk_size, use_gpu, logger)
# rename output files
if ark is not None:
os.rename(ark + '.tmp.ark', ark)
# first load scp and correct them to point to renamed ark file.
if scp is not None:
with open(scp + '.tmp.scp', 'rt') as fid_in:
with open(scp + '.tmp', 'wr') as fid_out:
text = fid_in.read()
text = text.replace('ark.tmp.ark', 'ark')
# Sometimes there is no \n at the end of file ank cause a Kaldi error.
# For preventing this error juts check the last char and append \n if not exist
if text[-1] != '\n':
text += '\n'
fid_out.write(text)
os.rename(scp + '.tmp', scp)
def main():
args = get_args()
inp_feats_scp = args.inp_feats_scp
out_feats_ark = args.out_feats_ark
ark_scp_output = 'ark:| copy-feats --compress=true ark:- ark,scp:{p}.ark,{p}.scp'.format(
p=out_feats_ark)
with kaldi_io.open_or_fd(ark_scp_output, 'wb') as f:
for utt, feats in kaldi_io.read_mat_scp(inp_feats_scp):
mfcc = convert_mfcc_to_fbank(feats)
np.save('ark_check4/{u}.npy'.format(u=utt), mfcc)
kaldi_io.write_mat(f, mfcc, key=utt)
def read_mat_key(file, target_key):
"""read the matrix of the target key/utterance from a kaldi scp file
"""
fd = ko.open_or_fd(file)
try:
for line in fd:
(key, rxfile) = line.decode().split(' ')
if key == target_key:
return ko.read_mat(rxfile)
finally:
if fd is not file: fd.close()
def plda_read(self,plda):
with kaldi_io.open_or_fd(plda,'rb') as f:
for key,vec in kaldi_io.read_vec_flt_ark(f):
if key == 'mean':
self.mean = vec.reshape(-1,1)
self.dim = self.mean.shape[0]
elif key == 'within_var':
self.within_var = vec.reshape(self.dim, self.dim)
else:
self.between_var = vec.reshape(self.dim, self.dim)
def read_all_key(file):
"""return all keys/utterances of a kaldi scp file
"""
key_list = []
fd = ko.open_or_fd(file)
try:
for line in fd:
(key, _) = line.decode().split(' ')
key_list.append(key)
finally:
if fd is not file: fd.close()
return key_list
def write_vectors(utt_list, csvector_list, data_path):
size = len(utt_list)
csvector_dict = {}
for i in range(size):
sub_list = utt_list[i]
sub_vector = csvector_list[i]
for j in range(len(sub_list)):
csvector_dict[sub_list[j]] = sub_vector[j]
with kaldi_io.open_or_fd(data_path,'wb') as f:
for k,v in csvector_dict.items():
kaldi_io.write_vec_flt(f, v, k)
def testInt32VectorReadWrite(self):
"""
Test read/write for int32 vectors.
"""
# read,
i32_vec = { k:v for k,v in kaldi_io.read_vec_int_ark('tests/data/ali.ark') } # binary,
i32_vec2 = { k:v for k,v in kaldi_io.read_vec_int_ark('tests/data/ali_ascii.ark') } # ascii,
# re-save the data,
with kaldi_io.open_or_fd('tests/data_re-saved/ali.ark','wb') as f:
for k,v in i32_vec.items(): kaldi_io.write_vec_int(f, v, k)
# read and make sure it is the same,
for k,v in kaldi_io.read_vec_int_ark('tests/data_re-saved/ali.ark'):
self.assertTrue(np.array_equal(v,i32_vec[k]), msg="int32 vector same after re-saving")
def testWriteReadPosteriors(self):
data = [[(0, 0.0), (1, 0.1), (2, 0.2)],
[(0, 0.00), (1, 0.11), (2, 0.22)],
[(0, 0.000), (1, 0.111), (3, 0.333)]]
key = 'posterior_test1'
with kaldi_io.open_or_fd('tests/data_re-saved/posterior_tests.ark',
'wb') as w:
kaldi_io.write_post(w, data, key=key)
with kaldi_io.open_or_fd('tests/data_re-saved/posterior_tests.ark',
'rb') as r:
posts = [(k, posteriors)
for k, posteriors in kaldi_io.read_post_ark(r)]
self.assertEqual(len(posts), 1)
self.assertEqual(posts[0][0], key)
rdata = posts[0][1]
self.assertEqual(len(rdata), len(data))
for a1, a2 in zip(rdata, data):
self.assertEqual(len(a1), len(a2))
for ((idx1, p1), (idx, p)) in zip(a1, a2):
self.assertEqual(idx1, idx)
self.assertAlmostEqual(p1, p)
def testMatrixReadWrite(self):
"""
Test read/write for float matrices.
"""
# read,
flt_mat = { k:m for k,m in kaldi_io.read_mat_scp('tests/data/feats_ascii.scp') } # ascii-scp,
flt_mat2 = { k:m for k,m in kaldi_io.read_mat_ark('tests/data/feats_ascii.ark') } # ascii-ark,
flt_mat3 = { k:m for k,m in kaldi_io.read_mat_ark('tests/data/feats.ark') } # ascii-ark,
# store,
with kaldi_io.open_or_fd('tests/data_re-saved/mat.ark','wb') as f:
for k,m in flt_mat3.items(): kaldi_io.write_mat(f, m, k)
# read and compare,
for k,m in kaldi_io.read_mat_ark('tests/data_re-saved/mat.ark'):
self.assertTrue(np.array_equal(m, flt_mat3[k]), msg="flt. matrix same after re-saving")
def testFloatVectorReadWrite(self):
"""
Test read/write for float vectors.
"""
# read,
flt_vec = { k:v for k,v in kaldi_io.read_vec_flt_scp('tests/data/conf.scp') } # scp,
return
flt_vec2 = { k:v for k,v in kaldi_io.read_vec_flt_ark('tests/data/conf.ark') } # binary-ark,
flt_vec3 = { k:v for k,v in kaldi_io.read_vec_flt_ark('tests/data/conf_ascii.ark') } # ascii-ark,
# store,
with kaldi_io.open_or_fd('tests/data_re-saved/conf.ark','wb') as f:
for k,v in flt_vec.items(): kaldi_io.write_vec_flt(f, v, k)
# read and compare,
for k,v in kaldi_io.read_vec_flt_ark('tests/data_re-saved/conf.ark'):
self.assertTrue(np.array_equal(v,flt_vec[k]), msg="flt. vector same after re-saving")
def testPipeReadWrite(self):
"""
Test read/write for pipes.
Note: make sure the "os.environ['KALDI_ROOT']" in "kaldi_io/kaldi_io.py" is correct.
"""
# the following line disables 'stderr' forwarding, comment it for DEBUG,
with open("/dev/null","w") as sys.stderr:
# read,
flt_mat4 = { k:m for k,m in kaldi_io.read_mat_ark('ark:copy-feats ark:tests/data/feats.ark ark:- |') }
# write to pipe,
with kaldi_io.open_or_fd('ark:| copy-feats ark:- ark:tests/data_re-saved/mat_pipe.ark','wb') as f:
for k,m in flt_mat4.items(): kaldi_io.write_mat(f, m, k)
# read it again and compare,
for k,m in kaldi_io.read_mat_ark('tests/data_re-saved/mat_pipe.ark'):
self.assertTrue(np.array_equal(m, flt_mat4[k]),"flt. matrix same after read/write via pipe")
# read some other formats from pipe,
i32_vec3 = { k:v for k,v in kaldi_io.read_vec_int_ark('ark:copy-int-vector ark:tests/data/ali.ark ark:- |') }
flt_vec4 = { k:v for k,v in kaldi_io.read_vec_flt_ark('ark:copy-vector ark:tests/data/conf.ark ark:- |') }
if do_training: net.train() test_flag=0 N_batches=int(N_snt/batch_size) if rnn==0: N_ex_tr=data_set.shape[0] N_batches=int(N_ex_tr/batch_size) if do_eval: N_batches=N_snt net.eval() test_flag=1 batch_size=1 if do_forward: post_file=kaldi_io.open_or_fd(out_file,'wb') counts = load_counts(count_file) beg_batch=0 end_batch=batch_size snt_index=0 beg_snt=0 loss_sum=0 err_sum=0 for i in range(N_batches):
