def __init__(self, k,
scaled_lap=scaled_laplacian,
activation=None,
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
bias_constraint=None,
**kwargs):
if 'input_shape' not in kwargs and 'input_dim' in kwargs:
kwargs['input_shape'] = (kwargs.pop('input_dim'),)
super(LocallyConnectedGC, self).__init__(**kwargs)
self.k = k
self.scaled_lap = scaled_lap
self.activation = activations.get(activation)
self.use_bias = use_bias
self.kernel_initializer = initializers.get(kernel_initializer)
self.bias_initializer = initializers.get(bias_initializer)
self.kernel_regularizer = regularizers.get(kernel_regularizer)
self.bias_regularizer = regularizers.get(bias_regularizer)
self.activity_regularizer = regularizers.get(activity_regularizer)
self.kernel_constraint = constraints.get(kernel_constraint)
self.bias_constraint = constraints.get(bias_constraint)
self.supports_masking = True
self.supports = []
# k-hop adjacency matrix
S = K.constant(K.to_dense(calculate_adjacency_k(self.scaled_lap, self.k)))
self.supports.append(S)
def K_eval(x, backend=K):
K = backend
try:
return K.get_value(K.to_dense(x))
except Exception as e:
try:
eval_fn = K.function([], [x])
return eval_fn([])[0]
except Exception as e:
return K.eager(K.eval)(x)
def __init__(
self,
k,
units,
normalization=False,
attn_heads=1,
attn_heads_reduction='concat', # {'concat', 'average'}
scaled_lap=scaled_laplacian,
activation=None,
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
bias_constraint=None,
**kwargs):
if attn_heads_reduction not in {'concat', 'average'}:
raise ValueError('Possbile reduction methods: concat, average')
self.k = k
self.units = units
self.normalization = normalization
self.attn_heads = attn_heads # Number of attention heads (K in the paper)
self.attn_heads_reduction = attn_heads_reduction # Eq. 5 and 6 in the paper
self.scaled_lap = scaled_lap
self.activation = activations.get(activation) # Eq. 4 in the paper
self.use_bias = use_bias
self.kernel_initializer = initializers.get(kernel_initializer)
self.bias_initializer = initializers.get(bias_initializer)
self.kernel_regularizer = regularizers.get(kernel_regularizer)
self.bias_regularizer = regularizers.get(bias_regularizer)
self.activity_regularizer = regularizers.get(activity_regularizer)
self.kernel_constraint = constraints.get(kernel_constraint)
self.bias_constraint = constraints.get(bias_constraint)
self.supports_masking = False
# Populated by build()
self.kernels = [] # Layer kernels for attention heads
self.parameters = []
self.fc = []
self.biases = [] # Layer biases for attention heads
self.supports = []
s = K.constant(
K.to_dense(calculate_adjacency_k(self.scaled_lap, self.k)))
self.supports.append(s)
super(DynamicGC, self).__init__(**kwargs)
def __init__(self,
k,
units,
scaled_lap=scaled_laplacian,
inner_act=None,
activation='sigmoid',
**kwargs):
if 'input_shape' not in kwargs and 'input_dim' in kwargs:
kwargs['input_shape'] = (kwargs.pop('input_dim'), )
super(RecurrentDGC, self).__init__(**kwargs)
self.k = k
self.units = units
self.state_size = dim
self.output_size = dim
self.scaled_lap = scaled_lap
self.inner_act = inner_act
self.activation = activations.get(activation)
self.supports_masking = True
self.supports = []
s = K.constant(
K.to_dense(calculate_adjacency_k(self.scaled_lap, self.k)))
self.supports.append(s)
def batch(X, batch_size, seed=0, iterator=True):
""""
Partitions a dataset into batches, returning a batch dataset or an iterator.
:param X: The dataset to batch
:param batch_size: The size of each batch
:param seed: The shuffle seed
:retrun: A tensor batch dataset, or as a numpy iterator.
"""
# buffer_size = int(1e6)
buffer_size = X.shape[0] # For perfect shuffle, buff is the size of X
if K.is_sparse(X): # If a sparse tensor
X = K.to_dense(X)
elif sp.sparse.issparse(X): # If a sparse matrix
X = X.todense()
batches = Dataset.from_tensor_slices(X). \
shuffle(buffer_size=buffer_size, seed=seed). \
batch(batch_size, drop_remainder=True)
if iterator:
batches = batches.as_numpy_iterator()
return batches
def test_to_dense_matrix(self):
test_sparse_matrix = self.generate_test_sparse_matrix()
assert_allclose(K.to_dense(test_sparse_matrix), test_sparse_matrix.toarray())
def test_to_dense(self):
test_sparse_matrix = self.generate_test_sparse_matrix()
test_var = K.variable(test_sparse_matrix)
assert_allclose(K.to_dense(test_var), test_sparse_matrix.toarray())
def RecognizeSpeech(self, wavsignal, fs):
'''
最终做语音识别用的函数，识别一个wav序列的语音
不过这里现在还有bug
'''
#data = self.data
data = DataSpeech('E:\\语音数据集')
data.LoadDataList('dev')
# 获取输入特征
#data_input = data.GetMfccFeature(wavsignal, fs)
data_input = data.GetFrequencyFeature(wavsignal, fs)
list_symbol_dic = data.list_symbol # 获取拼音列表
labels = [
'dong1', 'bei3', 'jun1', 'de5', 'yi4', 'xie1', 'ai4', 'guo2',
'jiang4', 'shi4', 'ma3', 'zhan4', 'shan1', 'li3', 'du4', 'tang2',
'ju4', 'wu3', 'su1', 'bing3', 'ai4', 'deng4', 'tie3', 'mei2',
'deng3', 'ye3', 'fen4', 'qi3', 'kang4', 'zhan4'
]
#labels = [ list_symbol_dic[-1] ]
#labels = [ list_symbol_dic[-1] ]
#while(len(labels) < 32):
# labels.append(list_symbol_dic[-1])
feat_out = []
#print("数据编号",n_start,filename)
for i in labels:
if ('' != i):
n = data.SymbolToNum(i)
feat_out.append(n)
print(feat_out)
labels = feat_out
x = next(
self.data_gen(data_input=np.array(data_input),
data_labels=np.array(labels),
input_length=len(data_input),
labels_length=len(labels),
batch_size=2))
[test_input_data, y, test_input_length, label_length], labels = x
xx = [test_input_data, y, test_input_length, label_length]
pred = self._model.predict(x=xx)
print(pred)
shape = pred[:, :].shape
print(shape)
#print(test_input_data)
y_p = self.test_func([test_input_data])
print(type(y_p))
print('y_p:', y_p)
for j in range(0, 200):
mean = sum(y_p[0][0][j]) / len(y_p[0][0][j])
print('max y_p:', max(y_p[0][0][j]), 'min y_p:', min(y_p[0][0][j]),
'mean y_p:', mean, 'mid y_p:', y_p[0][0][j][100])
print('argmin:', np.argmin(y_p[0][0][j]), 'argmax:',
np.argmax(y_p[0][0][j]))
count = 0
for i in y_p[0][0][j]:
if (i < mean):
count += 1
print('count:', count)
print(K.is_sparse(y_p))
y_p = K.to_dense(y_p)
print(K.is_sparse(y_p))
#y_p = tf.sparse_to_dense(y_p,(2,397),1417,0)
print(test_input_length.T)
test_input_length = test_input_length.reshape(2, 1)
func_in_len = self.test_func_input_length([test_input_length])
print(type(func_in_len))
#in_len = np.ones(shape[0]) * shape[1]
ctc_decoded = K.ctc_decode(y_p, input_length=func_in_len)
print(ctc_decoded)
#ctc_decoded = ctc_decoded[0][0]
#out = K.get_value(ctc_decoded)[:,:64]
#pred = self._model.predict_on_batch([data_input, labels_num, input_length, label_length])
return pred[0][0]
pass