def run(self):
while True:
keyboard.wait(hotkey="k + l")
saidtext = get_audio()
if saidtext == None:
notification.show_toast("Error",
"Sorry could not recognize your voice",
threaded=True,
duration=3)
else:
print("What you said: ", saidtext)
vr.split_command(saidtext)
def main():
speaker = "takeru"
emotionList = ["happiness", "anger", "sadness"]
pitch = 100
speed = 100
volume = 100
print("*** AI自動返信開始 ***")
# print("([E]を押して終了)")
while True:
root_vR: ClVoiceRecognition = VoiceRecognition.ClVoiceRecognition(
) # インスタンス生成
result = root_vR.voiceToText()
if bool(result[0]):
root_gR = GetReply.ClGetReply(result[1]) # インスタンス生成
reply = root_gR.getReply()
else:
reply = "すみません。よく分かりません。"
# TestCode 感情(ランダム)
emotion = emotionList[rd.randint(0, 2)]
emotion_level = rd.randint(1, 4)
# TestCode_END
root_sS = SpeechSynthesis.ClSpeechSynthesis("voice", reply, speaker,
emotion, emotion_level,
pitch, speed,
volume) # インスタンス生成
voiceFile = root_sS.getVoice()
root_tS = TextSpeaker.ClTextSpeaker(voiceFile) # インスタンス生成
root_tS.outputWav()
if not bool(result[0]):
sys.exit()
def test():
vr.textToSpeech('What numbers do you need to subtract?')
numbers = vr.speechToText()
numbers.split(' ')
print(numbers)
answer = [int(i) for i in numbers]
i = 1
ans = answer[0]
while i < len(answer):
sub = answer[i]
ans = ans / sub
print(ans)
i += 1
vr.textToSpeech(ans)
def CallParameter(MinidB,RefPcmPath):
#wavfilelist=Get_ref_file(path_ref_file)
#path1 = wavpath + wavfilelist[i]
isExists=os.path.exists("output.pcm")
if isExists:
os.remove("output.pcm")
#print RefPcmPath
command1 = "ffmpeg -i "+ RefPcmPath+" -f s16le -ar 48000 -acodec pcm_s16le output.pcm"
subprocess.call(command1)
#subprocess.call('AECResultAutoAnalysis.exe output.pcm 48000 '+ MinidB)
PCMdBList = GetVoiceList.getdBList("output.pcm", 48000)
(VoiceList,VoiceRefTimePeakdB) = GetVoiceList.GetVoiceList(PCMdBList, 48000, int(MinidB))
commandref="AECResultAutoAnalysis.exe output.pcm 48000 "+ str(MinidB)
var = os.popen(commandref)
str1=var.read(100)
str1 = re.findall(r"\d+\.?\d*",str1)
#print str
RefTime = str1[1]
RefVPTime = VoiceRefTimePeakdB[0]
RefPeakdB = VoiceRefTimePeakdB[2]
#------计算语音识别字数-------#
isExists=os.path.exists("output.wav")
if isExists:
os.remove("output.wav")
command2 = "ffmpeg -i "+ RefPcmPath+" -f s16le -ar 8000 -acodec pcm_s16le output.wav"
subprocess.call(command2)
(ResultStr,ResultStrLen) = VoiceRecognition.VoiceRecognition("output.wav")
ResultStrLen=ResultStrLen
#print ResultStrLenList
return (RefVPTime,RefTime,RefPeakdB,ResultStrLen)
def CallParameter(MinidB):
wavpath=os.getcwd()+"\\voice\\"
wavfilelist=[]
PcmRefTime=[]
PcmRefPeakdB=[]
ResultStrLenList=[]
wavfilelist=Get_ref_file(wavpath)
for i in range(len(wavfilelist)):
path1 = wavpath + wavfilelist[i]
isExists=os.path.exists("output.pcm")
if isExists:
os.remove("output.pcm")
command1 = "ffmpeg -i "+ path1+" -f s16le -ar 48000 -acodec pcm_s16le output.pcm"
print command1
subprocess.call(command1)
#subprocess.call('AECResultAutoAnalysis.exe output.pcm 48000 '+ MinidB)
commandref="AECResultAutoAnalysis.exe output.pcm 48000 "+ str(MinidB)
#subprocess.call(commandref)
var = os.popen(commandref)
str1=var.read(100)
str1 = re.findall(r"\d+\.?\d*",str1)
#print str1
PcmRefTime.append(str1[1])
PcmRefPeakdB.append(str1[2])
#------识别字数-------#
FFMPEG=multiprocessing.Process(target=os.system,args=('.\\ffmpeg.exe -f s16le -ar 48000 -ac 1 -i '+path1+' -acodec pcm_s16le -ar 8000 -ac 1 out.wav -y',))
FFMPEG.daemon=True
FFMPEG.start()
FFMPEG.join()
(ResultStr, ResultStrLen) = VoiceRecognition.VoiceRecognition('out.wav')
#(ResultStr,ResultStrLen) = VoiceRecognition.VoiceRecognition(path1)
ResultStrLenList.append(ResultStrLen)
#print ResultStrLenList
return (PcmRefTime,PcmRefPeakdB,ResultStrLenList)
def addition():
vr.textToSpeech('What numbers do you need to add?')
numbers = vr.speechToText()
numbers.split(' ')
answer = list(map(int, numbers))
vr.textToSpeech('the answer is')
vr.textToSpeech(sum(answer))
def division():
vr.textToSpeech('What numbers do you need to divide?')
numbers = vr.speechToText()
numbers.split(' ')
answer = list(map(int, numbers))
i = 1
ans = answer[0]
while i < len(answer):
sub = answer[i]
ans = ans / sub
print(ans)
i += 1
vr.textToSpeech('the answer is')
vr.textToSpeech(ans)
class MainThread(QThread):
img_signal = pyqtSignal(object) # 输出图像的信号
realsenseCamera = RealsenseCamera()
aligned_frames = 0 # a bunch of images
color_image = 0 # colorful images
calied_img = 0
marked_img = 0
value = 0 # 用来找阈值
isVoiceGet = 0
#Coordinate Transfer
needCali = 0
# Recognize model
model = 0
recognizeResult = 0
recCount = 0
recList = 0
# Workspace Area
pickArea = (11, 10.5, 30, 26)
toolBox = ({
'id': 1,
'name': 'screwdriver',
'pos': (15, -24),
'wide': 0.0
}, {
'id': 2,
'name': 'pincer',
'pos': (25, -11),
'wide': 0.5
}, {
'id': 3,
'name': 'alien',
'pos': (25, 0),
'wide': 4.0
}, {
'id': 4,
'name': 'pen',
'pos': (25, 0),
'wide': 0.0
})
toolInHand = 'BY1'
# robotArm
robotArm = RobotArm('COM3')
#voice recognition
voiceRecognition = VoiceRecognition()
exitFlag = 0
# init: load data
def init(self):
self.aligned_frames = self.realsenseCamera.get_aligned_frames()
self.marked_img = self.calied_img = self.color_image = self.realsenseCamera.get_color_image_from_frames(
self.aligned_frames)
self.model = create_model("mask_rcnn_tools_0030.h5")
self._camCalibrate()
# lack: Transform, voicRec
pass
# stop: Actually there's no way to stop
def stop(self):
self.exitFlag = 1
pass
# start: Start the main Thread
def run(self):
print('running...')
self.init()
print('init done')
while self.exitFlag != 1:
# 1. Motion Detection
last_color_image = self.color_image
color_image = self.realsenseCamera.get_color_image_from_frames(
self.realsenseCamera.get_aligned_frames())
while (self._ifMoving(last_color_image, color_image, 6,
2800) == False and self.isVoiceGet
== 0): # fine tune the kern and Thres to get best result
last_color_image = color_image
color_image = self.realsenseCamera.get_color_image_from_frames(
self.realsenseCamera.get_aligned_frames())
if (self.isVoiceGet == 0):
while (self._ifMoving(last_color_image, color_image, 6,
800) == True):
last_color_image = color_image
a_img = self.realsenseCamera.get_aligned_frames()
color_image = self.realsenseCamera.get_color_image_from_frames(
a_img)
time.sleep(0.5)
self.aligned_frames = self.realsenseCamera.get_aligned_frames()
self.color_image = self.realsenseCamera.get_color_image_from_frames(
self.aligned_frames)
print('Start Detect!')
# 2. Tools Detect
rgb_image = cv2.cvtColor(self.color_image, cv2.COLOR_BGR2RGB)
start_time = time.time()
self.recognizeResult, self.marked_img = ob_evaluate(
model=self.model,
image=rgb_image,
sampleImg=self.calied_img)
self.recCount, self.recList = calculate(self.recognizeResult)
print("time till pick color=", time.time() - start_time)
print('Detect Finish!')
print('Central', self._getCentral())
# 3. get Coordinate
if (self.needCali):
self._camCalibrate()
realXYZ, self.toolInHand = self._getCentral()
if (realXYZ[0] != 0 or realXYZ[1] != 0):
for i in self.toolBox:
if (i['name'] == self.toolInHand):
targetXYZ = i['pos']
w = i['wide']
targetXYZ = (targetXYZ[0], targetXYZ[1], 0.2)
print('Coord Obtained')
print('---------Finish----------')
print('tool is:' + self.toolInHand)
print('Actual Coord:', realXYZ)
realXYZ[2] = 0.2
# 4. Move Arm
self.robotArm.moveObject(fromXYZ=realXYZ,
toXYZ=targetXYZ,
wide=w)
self.robotArm.resetArm()
print('exec Finish')
else:
print('No Tools Detected!..')
#---------------------------------------------------------------------------
else:
# get voice Message
self.aligned_frames = self.realsenseCamera.get_aligned_frames()
self.color_image = self.realsenseCamera.get_color_image_from_frames(
self.aligned_frames)
print('Voice detected! Start!')
rgb_image = cv2.cvtColor(self.color_image, cv2.COLOR_BGR2RGB)
start_time = time.time()
self.recognizeResult, self.marked_img = ob_evaluate(
model=self.model,
image=rgb_image,
sampleImg=self.calied_img)
self.recCount, self.recList = calculate(self.recognizeResult)
print("time till pick color=", time.time() - start_time)
print('Detect Finish!')
# print('Central', self._getCentral())
for i in self.toolBox:
if i['id'] == self.isVoiceGet:
toolName = i['name']
xyData, self.toolInHand = self._getCentral(toolName)
if (xyData[0] != 0 or xyData[1] != 0):
if (self.needCali):
self._camCalibrate()
realXYZ = self.realsenseCamera.coor_trans_pixelxy2worldXYZ(
self.aligned_frames, xyData)
for i in self.toolBox:
if (i['name'] == self.toolInHand):
targetXYZ = i['pos']
w = i['wide']
targetXYZ = (targetXYZ[0], targetXYZ[1], 0.2)
print('Coord Obtained')
print('---------Finish----------')
print('Actual Coord:', realXYZ)
realXYZ[2] = 0.2
# 4. Move Arm
self.robotArm.moveObject(fromXYZ=realXYZ,
toXYZ=targetXYZ,
wide=w)
self.robotArm.resetArm()
print('exec Finish')
else:
# self.robotArm.swing()
print('No Tools Detected!..')
pass
# _audioStart: Start the audio detection thread
def audioT(self):
# press the Buttum
# Run this
self.isVoiceGet = self.voiceRecognition.run()
pass
# camCalibrate: Calibrate the cam
def _camCalibrate(self):
self.realsenseCamera.get_transform_matrix(self.aligned_frames)
while (self.realsenseCamera.if_get_position_flag is not True):
aligned_frames = self.realsenseCamera.get_aligned_frames()
tmpimg = self.realsenseCamera.get_transform_matrix(aligned_frames)
self.marked_img = tmpimg
# cv2.imshow("Positioning", tmpimg)
# cv2.waitKey(1)
# cv2.destroyWindow("Positioning")
self.needCali = 0
print("TF_Matrix Obtained!")
pass
def _ifMoving(self, frame1, frame2, kern, thres):
gray1 = cv2.cvtColor(frame1, cv2.COLOR_BGR2GRAY)
gray2 = cv2.cvtColor(frame2, cv2.COLOR_BGR2GRAY)
if kern % 2 == 0:
kern = kern + 1 # 解决滑动条赋值到高斯滤波器是偶数异常抛出
gray1 = cv2.GaussianBlur(gray1, (kern, kern), 0)
gray2 = cv2.GaussianBlur(gray2, (kern, kern), 0)
frameDelta = cv2.absdiff(gray1, gray2)
thresh = cv2.threshold(frameDelta, 25, 255, cv2.THRESH_BINARY)[1]
thresh = cv2.dilate(thresh, None, iterations=2)
cnts = cv2.findContours(thresh.copy(), cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)[-2]
flat = False
for c in cnts:
if cv2.contourArea(c) > thres: # 对于较小矩形区域，选择忽略
flat = True
return flat
def _getCentral(self, name=None):
cent = (0, 0)
if name == None:
for i in self.recList:
c = i['center']
c = self.realsenseCamera.coor_trans_pixelxy2worldXYZ(
self.aligned_frames, c)
if c is None:
c = (0, 0)
continue
if (c[0] > self.pickArea[0] and c[0] < self.pickArea[2]
and c[1] > self.pickArea[1]
and c[1] < self.pickArea[3]):
cent = c
name = i['name']
else:
for i in self.recList:
if (i['name'] == name):
cent = self.realsenseCamera.coor_trans_pixelxy2worldXYZ(
self.aligned_frames, i['center'])
if cent is None:
cent = (0, 0)
return cent, name
def _exec(self):
pass
import VoiceRecognition as vr
import MathFunc as mf
recognized_text = vr.speechToText()
x = recognized_text.split(' ')
if 'math' and 'problem' in x:
vr.textToSpeech('sure, what kind of problem?')
recognized_text = vr.speechToText()
if recognized_text == 'addition':
mf.addition()
elif recognized_text == 'subtraction':
mf.subtraction()
elif recognized_text == 'multiplication':
mf.multiplication()
elif recognized_text == 'division':
mf.division()
else:
vr.textToSpeech('addition is all I can do')
else:
vr.textToSpeech('I can only do math problems')
ret, frame = cap.read()
# show a frame
#cv2.imshow("capture", frame)
cv2.imwrite("cap.jpg", frame)
#cap.release()
#cv2.destroyAllWindows()
#拍下当前摄像头前面一张图像
live = livecheck.checklive()
print(live)
if live == None:
print("live check return None")
elif float(live) > 0.8:
gethello.opendoor('VoiceTip')
RecordWav2.my_record()
print('Over!')
VoiceReturn = VoiceRecognition.RecoLocalWav()
if VoiceReturn != None:
VoiceText = VoiceReturn[0]
if "开门" in VoiceText:
print('opendoor') #这里应该在写读取’您具有开门资格的语音
gethello.opendoor('Success')
serialComu.arduino_opendoor()
else:
print("Sorry, Voice can't be recognizied!")
gethello.opendoor('VoiceNot')
#serialComu.arduino_opendoor()# 向arduino传递数据进行开门操作
elif float(live) < 0.9:
gethello.opendoor('notreal')
print(
"It seems that the people infront of the door is not a real man"
)
def AECResultAutoAnalysis(PCMPath,SampleRate,MinidB,RefTime,RefPeakdB,RefVLen,AECMod):
start = time.time()
PCMdBList = getdBList(PCMPath, SampleRate)
TestTime = time.strftime('%Y-%m-%d-%H-%M-%S',time.localtime(time.time()))
FFMPEG=multiprocessing.Process(target=os.system,args=('.\\ffmpeg.exe -f s16le -ar '+str(SampleRate)+' -ac 1 -i '+PCMPath+' -acodec pcm_s16le -ar 8000 out.wav -y',))
FFMPEG.daemon=True
FFMPEG.start()
FFMPEG.join()
(VRRestult, VPAllStrLen) = VoiceRecognition.VoiceRecognition('out.wav')
#单讲
if AECMod == 's' or AECMod =='S':
(AllGreaterThanMinidBTime,PeakdB,AVGdB) = PCMGreaterThanMinidBTime(PCMdBList,SampleRate,MinidB)
AECFailPercent = (AllGreaterThanMinidBTime / RefTime) * 100
TestResultFileName = 'SingleTalk.csv'
ResultHead = 'TestTime,Result,Cooked,PeakdB,CookedTime,RefTime,Cooked/RefTime*100%,MaxVPTime,MaxVPTimedB,MiniVPTime,MiniVPTimedB,MaxVPdB,MaxVPdBTime,MiniVPdB,MiniVPdBTime,AVGVPTime,AVGVPdB,VPCount,VRRestult,CookVLen,RefVLen'
if AllGreaterThanMinidBTime > 50 or PeakdB > -20:
AECResult = 'F'
VoicePieceList = GetVoiceList(PCMdBList,SampleRate,MinidB)
VoicePieceCount = len(VoicePieceList)
VoicePieceList.sort()
MaxVPTime = VoicePieceList[VoicePieceCount-1][0] #VP: VoicePiece 获取最长时间片长度
MaxVPTimedB = VoicePieceList[VoicePieceCount-1][1] #最长时间片的最大dB
VoicePieceList.sort(key = lambda x:(x[0],-x[1])) #按照元组的第一个（片时间）从小到大排序,如果第一个相同,则按照元组第2个（片最大dB）从大到小排序，从而可以得到时间最小但声音最大的片
MiniVPTime = VoicePieceList[0][0] #最短时间片长度
MiniVPTimedB = VoicePieceList[0][1] #最短时间片的最大dB
VoicePieceList.sort(key = lambda x:x[1]) #按dB从小到大排序，若dB相同，片时间按从小到大排序
MaxVPdB = VoicePieceList[VoicePieceCount-1][1] #最大dB
MaxVPdBTime = VoicePieceList[VoicePieceCount-1][0] #最大dB所在时间片的长度
VoicePieceList.sort(key = lambda x:(x[1],-x[0])) #按照元组的第2个（片最大dB）从小到大排序,如果第2个相同,则按照元组第1个（片时间）从大到小排序，从而可以得到声音最小但时间最长的片
MiniVPdB = VoicePieceList[0][1] #最小dB
MiniVPdBTime = VoicePieceList[0][0] #最小dB所在时间片的长度
SumVPTime = 0
SumVPdB = 0
for i in range(VoicePieceCount):
SumVPTime = SumVPTime + VoicePieceList[i][0]
SumVPdB = SumVPdB + VoicePieceList[i][1]
AVGVPTime = SumVPTime // VoicePieceCount
AVGVPdB = SumVPdB // VoicePieceCount
Result = ''.join(TestTime+','+AECResult+','+PCMPath+','+str(PeakdB)+','+str(AllGreaterThanMinidBTime)+','+str(RefTime)+','+str(AECFailPercent)+','
+str(MaxVPTime)+','+str(MaxVPTimedB)+','+str(MiniVPTime)+','+str(MiniVPTimedB)+','+str(MaxVPdB)+','+str(MaxVPdBTime)+','+str(MiniVPdB)+
','+str(MiniVPdBTime)+','+str(AVGVPTime)+','+str(AVGVPdB)+','+str(VoicePieceCount)+','+VRRestult.replace(',','|')+','+str(VPAllStrLen)+','+str(RefVLen))
WriteTestResult(TestResultFileName, ResultHead, Result)
else:
AECResult = 'P'
Result = ''.join(TestTime+','+AECResult+','+PCMPath+','+str(PeakdB)+','+str(AllGreaterThanMinidBTime)+','+str(RefTime)+','+str(AECFailPercent)+','*12+
VRRestult.replace(',','|')+','+str(VPAllStrLen)+','+str(RefVLen))
WriteTestResult(TestResultFileName, ResultHead, Result)
#双讲
elif AECMod == 'd' or AECMod =='D':
(AllGreaterThanMinidBTime,PeakdB,AVGdB) = DoubleTalkPCMGreaterThanMinidBTime(PCMdBList,SampleRate,MinidB,RefPeakdB)
AECFailPercent = (AllGreaterThanMinidBTime / RefTime) * 100
TestResultFileName = 'DoubleTalk.csv'
ResultHead = 'TestTime,Result,Cooked,PeakdB,CookedTime,RefTime,Cooked/RefTime*100%,VRRestult,CookVLen,RefVLen'
DValue = VPAllStrLen - RefVLen
if DValue > 0:
AECResult = 'F'
elif DValue < 0:
AECResult = '有压制'
else:
AECResult = '效果良好'
Result = ''.join(TestTime+','+AECResult+','+PCMPath+','+str(PeakdB)+','+str(AllGreaterThanMinidBTime)+','+str(RefTime)+','+str(AECFailPercent)+','+
VRRestult.replace(',','|')+','+str(VPAllStrLen)+','+str(RefVLen))
WriteTestResult(TestResultFileName, ResultHead, Result)
end = time.time()
print end-start,'s'
def run(self):
vr.split_command(self.text)
self.exit()
import sys
sys.path.append("./")
from SpectrumGraph import *
from VoiceRecognition import *
if __name__ == "__main__":
whichEngine = int(
input(
"Select voice processing engine : \n 1 : google (online : more precise : requires internet connection) \n 2 : Built-in (offline) \n"
))
if whichEngine == 1:
whichEngine = "google"
elif whichEngine == 2:
whichEngine = "sphinx"
else:
whichEngine = "sphinx"
VoiceRecognition(whichEngine).recognizeVoice()
time.sleep(5) #noise cancelation waiting time
SpectrumGraph(9, 4, False).start_plot()
def AECResultAutoAnalysis(PCMPath, SampleRate, MinidB, RefVPTime, RefTime,
RefPeakdB, RefVLen, AECMod):
start = time.time()
TestTime = time.strftime('%Y-%m-%d-%H-%M-%S', time.localtime(time.time()))
FFMPEG = multiprocessing.Process(
target=os.system,
args=(r'ffmpeg.exe -f s16le -ar ' + str(SampleRate) + ' -ac 1 -i ' +
PCMPath + ' -acodec pcm_s16le -ar 8000 -ac 1 out.wav -y', ))
FFMPEG.daemon = True
FFMPEG.start()
FFMPEG.join()
(VRRestult, VPAllStrLen) = VoiceRecognition.VoiceRecognition('out.wav')
#VRRestult=VRRestult.encode('gbk')
#print type(VRRestult)
PCMdBList = getdBList(PCMPath, SampleRate)
#单讲
if AECMod == 's' or AECMod == 'S':
(AllGreaterThanMinidBTime, PeakdB,
AVGdB) = PCMGreaterThanMinidBTime(PCMdBList, SampleRate, MinidB)
AECFailPercent = (AllGreaterThanMinidBTime / RefTime) * 100
(VoicePieceList, VoiceRefVPTimePeakdB) = GetVoiceList.GetVoiceList(
PCMdBList, SampleRate, MinidB)
VoicePieceCount = len(VoicePieceList)
CookedVPtime = VoiceRefVPTimePeakdB[0]
AECVPTimePercent = (CookedVPtime / RefVPTime) * 100
TestResultFileName = 'SingleTalk.csv'
ResultHead = 'TestTime,Result,Cooked,PeakdB,CookedVPtime,RefVPTime,CookedTime/RefTime*100%,VPList,CookedTime,RefTime,CookedTime/RefTime*100%,MaxVPTime,MaxVPTimedB,MiniVPTime,MiniVPTimedB,MaxVPdB,MaxVPdBTime,MiniVPdB,MiniVPdBTime,AVGVPTime,AVGVPdB,VPCount,VRRestult,CookVLen,RefVLen,Scorce,average'
VoicePieceList.sort(key=lambda x: x[1]) #
#print VoicePieceList
MaxVPTime = VoicePieceList[VoicePieceCount -
1][1] #VP: VoicePiece 获取最长时间片长度
#print MaxVPTime
MaxVPTimedB = VoicePieceList[VoicePieceCount - 1][2] #最长时间片的最大dB
VoicePieceList.sort(
key=lambda x: (x[1], -x[2])
) #按照元组的第一个（片时间）从小到大排序,如果第一个相同,则按照元组第2个（片最大dB）从大到小排序，从而可以得到时间最小但声音最大的片
MiniVPTime = VoicePieceList[0][1] #最短时间片长度
MiniVPTimedB = VoicePieceList[0][2] #最短时间片的最大dB
VoicePieceList.sort(key=lambda x: x[2]) #按dB从小到大排序，若dB相同，片时间按从小到大排序
MaxVPdB = VoicePieceList[VoicePieceCount - 1][2] #最大dB
MaxVPdBTime = VoicePieceList[VoicePieceCount - 1][1] #最大dB所在时间片的长度
VoicePieceList.sort(
key=lambda x: (x[2], -x[0])
) #按照元组的第2个（片最大dB）从小到大排序,如果第2个相同,则按照元组第1个（片时间）从大到小排序，从而可以得到声音最小但时间最长的片
MiniVPdB = VoicePieceList[0][2] #最小dB
MiniVPdBTime = VoicePieceList[0][1] #最小dB所在时间片的长度
AVGVPTime = (VoiceRefVPTimePeakdB[0]) // VoicePieceCount
AVGVPdB = (VoiceRefVPTimePeakdB[1]) // VoicePieceCount
MaxVPTimedB = VoiceRefVPTimePeakdB[2]
VoicePieceList = str(VoicePieceList).replace(',', '')
if CookedVPtime > 50 or MaxVPdB > -35:
#计算单讲平均值
#Scorce1：语音片段残留时间
Scorce1 = ('%.2f' % (1 - CookedVPtime / RefVPTime))
#Scorce2：平均语音片段长度
Scorce2 = ('%.2f' % (1 - AVGVPTime / RefVPTime))
#Scorce2 = ('%.2f' %(1-VoicePieceCount/12))
#Scorce3:峰值分贝比值 划分区段
Scorce3 = dBCompare(MaxVPdB)
#Scorce41：最大片段残留时间/语音片段残留时间；Scorce42：最大片段所在的声音音量 划分区段
Scorce41 = float(('%.2f' % (1 - MaxVPTime / RefVPTime)))
Scorce42 = float(dBCompare(MaxVPTimedB))
Scorce4 = ('%.2f' % (math.sqrt(Scorce41 * Scorce42)))
print Scorce41, Scorce42, Scorce4
#Scorce5 = ('%.2f' %(abs(MaxVPTimedB)/abs(MinidB)))
#Scorce51：最大音量 划分区段；Scorce52：最大音量所在的语音残留时间
Scorce51 = Scorce3
Scorce52 = float(('%.2f' % (1 - MaxVPdBTime / RefVPTime)))
Scorce5 = float(('%.2f' % (math.sqrt(Scorce51 * Scorce52))))
#语音识别字数
Scorce6 = ('%.2f' % (1 - VPAllStrLen / RefVLen))
#加权平均值
a = (float(Scorce1), float(Scorce2), float(Scorce3),
float(Scorce4), float(Scorce5), float(Scorce6))
average = 5 * (np.average(a, weights=[8, 1, 5, 1, 1, 8]))
ScorceList = str(Scorce1) + ' ' + str(Scorce2) + ' ' + str(
Scorce3) + ' ' + str(Scorce4) + ' ' + str(Scorce5) + ' ' + str(
Scorce6)
AECResult = 'F'
Result = ''.join(TestTime + ',' + AECResult + ',' + PCMPath + ',' +
str(PeakdB) + ',' + str(CookedVPtime) + ',' +
str(RefVPTime) + ',' + str(AECVPTimePercent) +
',' + VoicePieceList + ',' +
str(AllGreaterThanMinidBTime) + ',' +
str(RefTime) + ',' + str(AECFailPercent) + ',' +
str(MaxVPTime) + ',' + str(MaxVPTimedB) + ',' +
str(MiniVPTime) + ',' + str(MiniVPTimedB) + ',' +
str(MaxVPdB) + ',' + str(MaxVPdBTime) + ',' +
str(MiniVPdB) + ',' + str(MiniVPdBTime) + ',' +
str(AVGVPTime) + ',' + str(AVGVPdB) + ',' +
str(VoicePieceCount) + ',' + VRRestult + ',' +
str(VPAllStrLen) + ',' + str(RefVLen) + ',' +
str(ScorceList) + ',' + str(average))
WriteTestResult(TestResultFileName, ResultHead, Result)
else:
AECResult = 'P'
average = 5.00
VoicePieceList = str(VoicePieceList).replace(',', '')
Result = ''.join(TestTime + ',' + AECResult + ',' + PCMPath + ',' +
str(PeakdB) + ',' + str(CookedVPtime) + ',' +
str(RefVPTime) + ',' + str(AECVPTimePercent) +
',' + VoicePieceList + ',' +
str(AllGreaterThanMinidBTime) + ',' +
str(RefTime) + ',' + str(AECFailPercent) +
',' * 12 + VRRestult + ',' + str(VPAllStrLen) +
',' + str(RefVLen)) + ',' * 2 + str(average)
WriteTestResult(TestResultFileName, ResultHead, Result)
#双讲
elif AECMod == 'd' or AECMod == 'D':
reload(sys)
sys.setdefaultencoding('utf-8')
(AllGreaterThanMinidBTime, PeakdB,
AVGdB) = DoubleTalkPCMGreaterThanMinidBTime(PCMdBList, SampleRate,
MinidB, RefPeakdB)
AECFailPercent = (AllGreaterThanMinidBTime / RefTime) * 100
(VoicePieceList, VoiceRefVPTimePeakdB) = GetVoiceList.GetVoiceList(
PCMdBList, SampleRate, MinidB)
VoicePieceCount = len(VoicePieceList)
CookedVPtime = VoiceRefVPTimePeakdB[0]
AECVPTimePercent = (CookedVPtime / RefVPTime) * 100
TestResultFileName = 'DoubleTalk.csv'
ResultHead = 'TestTime,Result,Cooked,PeakdB,CookedVPtime,RefVPTime,CookedTime/RefTime*100%,VPList,CookedTime,RefTime,Cooked/RefTime*100%,VRRestult,CookVLen,RefVLen'
#语音识别得分
DValue = VPAllStrLen - RefVLen
if DValue > 0:
AECResult = 'F'
Scorce1 = 0.0
Scorce2 = float((CookedVPtime - VoicePieceCount * 50) / RefVPTime -
1)
if Scorce2 < 0:
Scorce2 = 0
elif DValue < 0:
AECResult = u'有压制'
Scorce1 = float(('%.2f' % (VPAllStrLen / RefVLen)))
Scorce2 = float(('%.2f' % (CookedVPtime / RefVPTime)))
else:
AECResult = u'效果良好'
Scorce1 = 1.0
Scorce2 = 1.0
#语音残留片段
a = (float(Scorce1), float(Scorce2))
average = 5 * (np.average(a, weights=[8, 2]))
ScorceList = str(Scorce1) + ' ' + str(Scorce2)
#print average,ScorceList
VoicePieceList = str(VoicePieceList).replace(',', '')
Result = ''.join(TestTime + ',' + AECResult + ',' + PCMPath + ',' +
str(PeakdB) + ',' + str(CookedVPtime) + ',' +
str(RefVPTime) + ',' + str(AECVPTimePercent) + ',' +
VoicePieceList + ',' + str(AllGreaterThanMinidBTime) +
',' + str(RefTime) + ',' + str(AECFailPercent) + ',' +
VRRestult + ',' + str(VPAllStrLen) + ',' +
str(RefVLen))
WriteTestResult(TestResultFileName, ResultHead, Result)
end = time.time()
print end - start, 's'