def generate_cmvn(self, filelist=None, dry_run=False): # pylint: disable=unused-argument
''' Generate mean and vars of features. '''
sums, square, count = utils.create_cmvn_statis(
self.taskconf['audio']['feature_size'],
self.taskconf['audio']['add_delta_deltas'])
self.sampler.chunk_size = 100000
self.sampler.pad_chunks = False
num_done = 0
for inputs, _, _, _, _ in \
self.generate_data():
# update stats
if inputs.ndim == 3:
inputs = np.expand_dims(inputs, axis=0)
sums, square, count = utils.update_cmvn_statis(
inputs, sums, square, count, axis=(0, 1))
num_done += 1
if num_done % 100 == 0:
logging.info('Done %d samples.' % (num_done))
if num_done > self.cmvn_max_samples:
break
# compute cmvn
mean, var = utils.compute_cmvn(sums, square, count)
if dry_run:
logging.info('save cmvn:{}'.format(self._cmvn_path))
else:
np.save(self._cmvn_path, (mean, var))
logging.info('generate cmvn done')
logging.info(mean)
logging.info(var)
def testCreateCmvnStatis(self): #pylint: disable=invalid-name
''' test creat_cmvn_statics '''
feat_size = 40
delta_deltas = True
sums, square, count = utils.create_cmvn_statis(feat_size, delta_deltas)
self.assertAllEqual(sums.shape, [1, feat_size, 3])
self.assertAllEqual(square.shape, [1, feat_size, 3])
self.assertAllEqual(count, 0)
delta_deltas = False
sums, square, count = utils.create_cmvn_statis(feat_size, delta_deltas)
self.assertAllEqual(sums.shape, [1, feat_size, 1])
self.assertAllEqual(square.shape, [1, feat_size, 1])
self.assertAllEqual(count, 0)
def generate_cmvn(self, filelist=None, dry_run=False):
del filelist
assert self._stride == 1.0
batch_size = self.config['solver']['optimizer']['batch_size']
features, labels = self.input_fn(
utils.INFER, batch_size,
num_epoch=1)().make_one_shot_iterator().get_next()
del labels
suffix = self.taskconf['suffix']
if suffix == '.npy':
logging.info('generate cmvn from numpy')
feature = features['inputs']
else:
logging.info('genearte cmvn from wav')
# tf extractor graph
params = feat_lib.speech_ops.speech_params(
sr=self.taskconf['audio']['sr'],
bins=self.taskconf['audio']['feature_size'],
add_delta_deltas=self.taskconf['audio']['add_delta_deltas'],
audio_frame_length=self.taskconf['audio']['winlen'],
audio_frame_step=self.taskconf['audio']['winstep'])
#[batch, Time] -> [batch, time, audio_channel]
waveforms = tf.expand_dims(features['inputs'], axis=-1)
#[batch, Time, feat_size, channles]
feature = feat_lib.speech_ops.batch_extract_feature(
waveforms, params)
# create stats vars
sums, square, count = utils.create_cmvn_statis(
self.taskconf['audio']['feature_size'],
self.taskconf['audio']['add_delta_deltas'])
try:
with tf.Session() as sess:
while True:
feat_np = sess.run(feature)
# update stats
sums, square, count = utils.update_cmvn_statis(feat_np,
sums,
square,
count,
axis=(0, 1))
except tf.errors.OutOfRangeError:
pass
# compute cmvn
mean, var = utils.compute_cmvn(sums, square, count)
logging.info('mean:{}'.format(mean))
logging.info('var:{}'.format(var))
if not dry_run:
np.save(self._cmvn_path, (mean, var))
logging.info('save cmvn:{}'.format(self._cmvn_path))
logging.info('generate cmvn done')
def testComputeCmvn(self): #pylint: disable=invalid-name
''' test compute cmvn '''
np.random.seed(12)
feat_size = 40
delta_deltas = True
shape = [2, 10, feat_size, 3 if delta_deltas else 1]
sums, square, count = utils.create_cmvn_statis(feat_size, delta_deltas)
feat = np.random.randn(*shape)
feat = feat.astype(np.float32)
sums, square, count = utils.update_cmvn_statis(feat, sums, square,
count)
mean, var = utils.compute_cmvn(sums, square, count)
mean_true, var_true = np.mean(feat, axis=(0, 1)), np.var(feat,
axis=(0, 1))
self.assertAllEqual(mean.shape, [1] + shape[2:])
self.assertAllClose(np.squeeze(mean, axis=0), mean_true)
self.assertAllClose(np.squeeze(var, axis=0), var_true)
def testUpdateCmvnStatis(self): #pylint: disable=invalid-name
''' test update cmvn statics '''
np.random.seed(12)
feat_size = 40
delta_deltas = True
shape = [2, 10, feat_size, 3 if delta_deltas else 1]
sums, square, count = utils.create_cmvn_statis(feat_size, delta_deltas)
feat = np.random.randn(*shape)
sums_true = np.expand_dims(np.sum(feat, axis=(0, 1)), axis=0)
square_true = np.expand_dims(np.sum(np.square(feat), axis=(0, 1)),
axis=0)
count_true = np.prod(shape[:2])
sums, square, count = utils.update_cmvn_statis(feat, sums, square,
count)
self.assertAllEqual(sums, sums_true)
self.assertAllEqual(square, square_true)
self.assertAllEqual(count, count_true)
def generate_cmvn(self, filelist=None, dry_run=False):
''' Generate mean and vars of features. '''
sums, square, count = utils.create_cmvn_statis(
self.taskconf['audio']['feature_size'],
self.taskconf['audio']['add_delta_deltas'])
self.sampler.set_chunk_size(100000)
self.sampler.set_pad_chunks(False)
for inputs, _, _, _, _, _ in \
self.generate_data():
# update stats
if inputs.ndim == 3:
inputs = np.expand_dims(inputs, axis=0)
sums, square, count = utils.update_cmvn_statis(
inputs, sums, square, count, axis=(0, 1))
# compute cmvn
mean, var = utils.compute_cmvn(sums, square, count)
if dry_run:
logging.info('save cmvn:{}'.format(self._cmvn_path))
else:
np.save(self._cmvn_path, (mean, var))
logging.info('generate cmvn done')
logging.info(mean)
logging.info(var)
