def vectorize(self, inputs):
"""
Vectorize inputs.
Parameters
----------
inputs: List[np.array]
List[np.array] or List[malaya_speech.model.frame.Frame].
Returns
-------
result: np.array
returned [B, D].
"""
inputs = [
input.array if isinstance(input, Frame) else input
for input in inputs
]
inputs = [self._vectorizer(input) for input in inputs]
inputs, lengths = padding_sequence_nd(inputs, dim=0, return_len=True)
r = self._execute(
inputs=[inputs, lengths],
input_labels=['Placeholder', 'Placeholder_1'],
output_labels=['logits'],
)
return r['logits']
def predict(self, inputs):
"""
Enhance inputs, will return melspectrogram.
Parameters
----------
inputs: List[np.array]
Returns
-------
result: List
"""
inputs = [
input.array if isinstance(input, Frame) else input
for input in inputs
]
mels = [featurization.scale_mel(s).T for s in inputs]
x, lens = padding_sequence_nd(mels, maxlen=256, dim=0, return_len=True)
r = self._execute(
inputs=[x],
input_labels=['Placeholder'],
output_labels=['logits'],
)
l = r['logits']
results = []
for index in range(len(x)):
results.append(
featurization.unscale_mel(x[index, :lens[index]].T +
l[index, :lens[index], :, 0].T))
return results
def predict_proba(self, inputs):
"""
Predict inputs, will return probability.
Parameters
----------
inputs: List[np.array]
List[np.array] or List[malaya_speech.model.frame.Frame].
Returns
-------
result: np.array
returned [B, D].
"""
inputs = [
input.array if isinstance(input, Frame) else input
for input in inputs
]
inputs = [self._vectorizer(input, **self._extra) for input in inputs]
if self.__model__ == 'deep-speaker':
dim = 0
else:
dim = 1
inputs = padding_sequence_nd(inputs, dim=dim)
inputs = np.expand_dims(inputs, -1)
r = self._execute(
inputs=[inputs],
input_labels=['Placeholder'],
output_labels=['logits'],
)
return softmax(r['logits'], axis=-1)
def vectorize(self, inputs):
"""
Vectorize inputs.
Parameters
----------
inputs: List[np.array]
List[np.array] or List[malaya_speech.model.frame.Frame].
Returns
-------
result: np.array
returned [B, D].
"""
inputs = [
input.array if isinstance(input, Frame) else input
for input in inputs
]
inputs = [self._vectorizer(input) for input in inputs]
inputs, lengths = padding_sequence_nd(inputs, dim=0, return_len=True)
if self._output_nodes:
r = self._sess.run(
self._output_nodes['logits'],
feed_dict={
self._input_nodes['Placeholder']: inputs,
self._input_nodes['Placeholder_1']: lengths,
},
)
else:
inputs = to_tf(self._eager_g.inputs, [inputs, lengths])
r = self._eager_g(**inputs)[0].numpy()
return r
def predict(self, inputs):
"""
Enhance inputs, will return melspectrogram.
Parameters
----------
inputs: List[np.array]
Returns
-------
result: List
"""
inputs = [
input.array if isinstance(input, Frame) else input
for input in inputs
]
mels = [featurization.scale_mel(s).T for s in inputs]
x, lens = padding_sequence_nd(mels, maxlen=256, dim=0, return_len=True)
if self._output_nodes:
l = self._sess.run(
self._output_nodes['logits'],
feed_dict={self._input_nodes['Placeholder']: x},
)
else:
inputs = to_tf(self._eager_g.inputs, [x])
l = self._eager_g(**inputs)[0].numpy()
results = []
for index in range(len(x)):
results.append(
featurization.unscale_mel(x[index, :lens[index]].T +
l[index, :lens[index], :, 0].T))
return results
def predict_proba(self, inputs):
"""
Predict inputs, will return probability.
Parameters
----------
inputs: List[np.array]
List[np.array] or List[malaya_speech.model.frame.Frame].
Returns
-------
result: np.array
returned [B, D].
"""
inputs = [
input.array if isinstance(input, Frame) else input
for input in inputs
]
inputs = [self._vectorizer(input, **self._extra) for input in inputs]
if self.__model__ == 'deep-speaker':
dim = 0
else:
dim = 1
inputs = padding_sequence_nd(inputs, dim=dim)
inputs = np.expand_dims(inputs, -1)
if self._output_nodes:
r = self._sess.run(
self._output_nodes['logits'],
feed_dict={self._input_nodes['Placeholder']: inputs},
)
else:
inputs = to_tf(self._eager_g.inputs, [inputs])
r = self._eager_g(**inputs)[0].numpy()
return r
