def SpeechPostProcess(resultList, in_len): # 未完成！！！
#print("--------------------")
#print(resultList)
resultList1 = resultList[0]
#print "resultList[0] shape" + str(resultList1.shape) # (200,1,1,1424)
#print resultList1[0][0][0][1423]
# *****************
resultArray = resultList[0]
batchNum = resultArray.shape[0] #200
#print "batchNum: " + str(batchNum)
confidenceNum = resultArray.shape[1] #1
#print "confidenceNum: " + str(confidenceNum)
#print "resultArray: " + str(resultArray)
#print "resultArray shape: " + str(resultArray.shape) # (200,1,1,1424)
confidenceList = resultArray[:, 0, 0, :]
resultArray1 = np.swapaxes(resultArray, 0, 2)
resultArray2 = np.swapaxes(resultArray1, 0, 1)
confidenceList = resultArray2[0]
#print(confidenceList)
#print("+++++++++++++++++++++++++")
#print "confidenceList shape: " + str(confidenceList.shape) # (1,200,1424)
confidenceArray = np.array(confidenceList)
#print "confidenceArray: " + str(confidenceArray)
resultList = confidenceArray
#resultList = resultList[:, :, :]
#print("----------------------")
#print(resultList.shape)
ret = K.ctc_decode(resultList,
in_len,
greedy=True,
beam_width=100,
top_paths=1)
#print(ret[0])
ret1 = K.get_value(ret[0][0])
#print ret1
ret1 = ret1[0]
list_symbol_dic = GetSymbolList()
r_str = []
for i in ret1:
r_str.append(list_symbol_dic[i])
print "拼音序列识别结果：" + str(r_str)
ml = ModelLanguage('language_model')
ml.LoadModel()
#ml = LoadModel()
str_pinyin = r_str
#str_pinyin = ['ni3', 'hao3', 'a1']
#print str_pinyin
r = ml.SpeechToText(str_pinyin)
return r
def SpeechPostProcess(resultLists):
"""
Function description:
Save speech recognition results
Parameter:
resulilist
Return Value:
txt,pinyin
"""
# 将三维矩阵转为二维
# print("AAA")
dets = np.reshape(resultLists, (200, 1424))
# print("BBB")
# 将识别结果转为拼音序列
rr, ret1 = greedy_decode(dets)
# 去除拼音序列中的blank
for i in range(len(ret1)):
if i % 2 == 0:
try:
ret1.remove(1423)
except Except as e:
pass
# print(rr,ret1)
list_symbol_dic = GetSymbolList()
r_str = []
for i in ret1:
r_str.append(list_symbol_dic[i])
# print("拼音序列识别结果：" + str(r_str))
string_pinyin = str(r_str)
current_paths = os.path.dirname(__file__)
ml = ModelLanguage(os.path.join(current_paths + "/language_model"))
ml.LoadModel()
str_pinyin = r_str
r = ml.SpeechToText(str_pinyin)
# 保存语音识别的结果
with open(os.path.join(current_paths, 'results/asr_results.txt'), 'a+b') as f:
data = string_pinyin[1:-1] + '-' + r + '\n'
# print(1111111,data)
data=data.encode()
f.write(data)
f.close()
return r, str_pinyin
def SpeechPostProcess(resultList, in_len):
# 将三维矩阵转为二维
dets = np.reshape(resultList, (200,1424))
# 将识别结果转为拼音序列
rr, ret1 = greedy_decode(dets)
# 去除拼音序列中的blank
for i in range(len(ret1)):
if i % 2 == 0:
try:
ret1.remove(1423)
except:
pass
list_symbol_dic = GetSymbolList()
r_str = []
for i in ret1:
r_str.append(list_symbol_dic[i])
#print "拼音序列识别结果：" + str(r_str)
string_pinyin = str(r_str)
ml = ModelLanguage('language_model')
ml.LoadModel()
str_pinyin = r_str
r = ml.SpeechToText(str_pinyin)
#print(r)
# 保存语音识别的结果
with open('results/asr_results.txt','a+b') as f:
data = string_pinyin[1:-1] + '-' + r + '\n'
#print(data)
data=data.encode()
f.write(data)
f.close()
return r, str_pinyin
def SpeechPostProcess(resultList, in_len):
resultList1 = resultList[0]
resultArray = resultList[0]
batchNum = resultArray.shape[0]
confidenceNum = resultArray.shape[1]
confidenceList = resultArray[:, 0, 0, :]
resultArray1 = np.swapaxes(resultArray, 0, 2)
resultArray2 = np.swapaxes(resultArray1, 0, 1)
confidenceList = resultArray2[0]
confidenceArray = np.array(confidenceList)
resultList = confidenceArray
ret = K.ctc_decode(resultList,
in_len,
greedy=True,
beam_width=100,
top_paths=1)
#print(ret[0])
ret1 = K.get_value(ret[0][0])
#print ret1
ret1 = ret1[0]
list_symbol_dic = GetSymbolList()
r_str = []
for i in ret1:
r_str.append(list_symbol_dic[i])
print "拼音序列识别结果：" + str(r_str)
ml = ModelLanguage('language_model')
ml.LoadModel()
str_pinyin = r_str
r = ml.SpeechToText(str_pinyin)
return r