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)
arr_zero = np.zeros((1, 200), dtype=np.int16) #一个全是0的行向量
#import matplotlib.pyplot as plt
#plt.subplot(111)
#plt.imshow(data_input, cmap=plt.get_cmap('gray'))
#plt.show()
#while(len(data_input)<1600): #长度不够时补全到1600
# data_input = np.row_stack((data_input,arr_zero))
#print(len(data_input))
list_symbol = data.list_symbol # 获取拼音列表
labels = [list_symbol[0]]
#while(len(labels) < 64):
# labels.append('')
labels_num = []
for i in labels:
labels_num.append(data.SymbolToNum(i))
data_input = np.array(data_input, dtype=np.int16)
data_input = data_input.reshape(data_input.shape[0],
data_input.shape[1])
labels_num = np.array(labels_num, dtype=np.int16)
labels_num = labels_num.reshape(labels_num.shape[0])
input_length = np.array([data_input.shape[0] // 4 - 3], dtype=np.int16)
input_length = np.array(input_length)
input_length = input_length.reshape(input_length.shape[0])
label_length = np.array([labels_num.shape[0]], dtype=np.int16)
label_length = np.array(label_length)
label_length = label_length.reshape(label_length.shape[0])
x = [data_input, labels_num, input_length, label_length]
#x = next(data.data_genetator(1, self.AUDIO_LENGTH))
#x = kr.utils.np_utils.to_categorical(x)
print(x)
x = np.array(x)
pred = self._model.predict(x=x)
#pred = self._model.predict_on_batch([data_input, labels_num, input_length, label_length])
return [labels, pred]
pass
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)
input_length = len(data_input)
input_length = input_length // 4
data_input = np.array(data_input, dtype=np.float)
in_len = np.zeros((1), dtype=np.int32)
print(in_len.shape)
in_len[0] = input_length
batch_size = 1
x_in = np.zeros((batch_size, 1600, 200), dtype=np.float)
for i in range(batch_size):
x_in[i, 0:len(data_input)] = data_input
base_pred = self.base_model.predict(x=x_in)
print('base_pred:\n', base_pred)
#input_length = tf.squeeze(input_length)
#decode_pred = self.model_decode(x=[x_in, in_len])
#print(decode_pred)
base_pred = base_pred[:, 2:, :]
r = K.ctc_decode(base_pred,
in_len,
greedy=True,
beam_width=64,
top_paths=1)
print('r', r)
#r = K.cast(r[0][0], dtype='float32')
#print('r1', r)
#print('解码完成')
r1 = K.get_value(r[0][0])
print('r1', r1)
print('r0', r[1])
r2 = K.get_value(r[1])
print(r2)
print('解码完成')
list_symbol_dic = data.list_symbol # 获取拼音列表
#arr_zero = np.zeros((1, 200), dtype=np.int16) #一个全是0的行向量
#import matplotlib.pyplot as plt
#plt.subplot(111)
#plt.imshow(data_input, cmap=plt.get_cmap('gray'))
#plt.show()
#while(len(data_input)<1600): #长度不够时补全到1600
# data_input = np.row_stack((data_input,arr_zero))
#print(len(data_input))
#list_symbol = data.list_symbol # 获取拼音列表
#labels = [ list_symbol[0] ]
#while(len(labels) < 64):
# labels.append('')
#labels_num = []
#for i in labels:
# labels_num.append(data.SymbolToNum(i))
#data_input = np.array(data_input, dtype=np.int16)
#data_input = data_input.reshape(data_input.shape[0],data_input.shape[1])
#labels_num = np.array(labels_num, dtype=np.int16)
#labels_num = labels_num.reshape(labels_num.shape[0])
#input_length = np.array([data_input.shape[0] // 4 - 3], dtype=np.int16)
#input_length = np.array(input_length)
#input_length = input_length.reshape(input_length.shape[0])
#label_length = np.array([labels_num.shape[0]], dtype=np.int16)
#label_length = np.array(label_length)
#label_length = label_length.reshape(label_length.shape[0])
#x = [data_input, labels_num, input_length, label_length]
#x = next(data.data_genetator(1, self.AUDIO_LENGTH))
#x = kr.utils.np_utils.to_categorical(x)
#print(x)
#x=np.array(x)
#pred = self._model.predict(x=x)
#pred = self._model.predict_on_batch([data_input, labels_num, input_length, label_length])
#return [labels,pred]
return r1
pass
