def startSimulationSingleChannel(context, CNNkernel, sizekernelCNN, stridesCNN,
DenseNeuron, nameplace, dir_training, ax, ay,
numContext, dirAudio2, fi, mic_log_mel, hsc,
fittizio, simulatedcheck, realcheck,
numberdir_fittizio, pathdir_fittizio):
pathsave = 'Single_Channel' + dir_training + '/'
pathsave = pathsave + 'context' + str(context) + '/'
pathsave = pathsave + nameplace + '/'
pathsave = pathsave + 'CNNkernel' + str(int(CNNkernel[0])) + '/'
pathsave = pathsave + 'sizeKernel' + str(int(sizekernelCNN)) + '/'
pathsave = pathsave + 'strides' + str(int(stridesCNN)) + '/'
pathsave = pathsave + 'Densenumber' + str(int(len(DenseNeuron))) + '/'
pathsave = pathsave + 'Denseneuron' + str(int(DenseNeuron[0])) + '/'
if not os.path.exists(pathsave):
os.makedirs(pathsave)
VAD_tr_GCC1 = []
SLOC_tr_GCC_X = []
from getDatasetSingleChannel import getHSCMADevDataset
if hsc:
from getDatasetSingleChannel import getHSCMADevDataset
getHSCMADevDataset(VAD_tr_GCC1, SLOC_tr_GCC_X, ax, ay, context,
numContext, dirAudio2, fi, nameplace, mic_log_mel)
if simulatedcheck:
from getDatasetSingleChannel import getEVALITADataset
getEVALITADataset(VAD_tr_GCC1, SLOC_tr_GCC_X, ax, ay, realcheck,
context, numContext, dirAudio2, fi, nameplace,
mic_log_mel)
if fittizio:
from getDatasetSingleChannel import getDLSDataset
getDLSDataset(VAD_tr_GCC1, SLOC_tr_GCC_X, ax, ay, numberdir_fittizio,
pathdir_fittizio, context, numContext, dirAudio2, fi,
nameplace, mic_log_mel)
VAD_tr_GCC1 = np.asarray(VAD_tr_GCC1)
SLOC_tr_GCC_X = np.asarray(SLOC_tr_GCC_X)
indexes = np.random.permutation(VAD_tr_GCC1.shape[0])
VAD_tr_GCC = VAD_tr_GCC1[indexes]
print VAD_tr_GCC.shape
SLOC_tr_GCC = SLOC_tr_GCC_X[indexes]
del indexes, SLOC_tr_GCC_X, VAD_tr_GCC1
from getModel import getModelSingleChannel
model = getModelSingleChannel(CNNkernel, sizekernelCNN, DenseNeuron,
stridesCNN, fi, context, VAD_tr_GCC,
SLOC_tr_GCC, pathsave, nameplace)
from constants import *
import scipy.io.wavfile as wav
import scipy
from csp2dgpp import getCSP, organizeGCC
import cPickle
dev_test_array = ['Dev', 'Test']
dir_dt_array = ['HSCMA_DIRHA_dev', 'HSCMA_DIRHA_test']
dir_real_sim_array = ['real', 'sim']
dirrslower = dir_real_sim_array[1]
for dirdt in dev_test_array:
print dirdt
if dirdt == 'Dev':
dirdtupper = dev_test_array[0]
dirdt_hs = dir_dt_array[0]
devortest = 'dev'
else:
dirdtupper = dev_test_array[1]
dirdt_hs = dir_dt_array[1]
devortest = 'test'
realorsim = dir_real_sim_array[1]
dirrs = real_sim_array[1]
lan_array = ['IT', 'GK', 'GE', 'PT']
numberdir_hscma = 10
numberdirtraining_hscma = np.linspace(1, numberdir_hscma,
numberdir_hscma)
for lan in lan_array:
print lan
dirAudio = pathdir_hscma + dirdt_hs + '/' + dirrs + '/' + dirdtupper + '/' + lan + '/'
for numerocartelle in numberdirtraining_hscma:
print 'training number: ' + str(int(numerocartelle))
dira = dirAudio + dirrslower + str(
int(numerocartelle)) + dirAudio2
from funzioniSupplementari import getMic
AUDIO = getMic(nameplace, dira)
del getMic
AUDIO16 = []
for i in range(0, mic_log_mel):
AUDIO16.append(scipy.signal.decimate(AUDIO[i], 48 / 16))
l = len(AUDIO16[0]) / hopframe
if realorsim == 'sim' or (realorsim == 'real' and
(nameplace == 'Livingroom'
or nameplace == 'Kitchen')):
filesource = dirAudio + dirrslower + str(
int(numerocartelle)
) + '/' + dirSource2 + '/' + nameplace + formatref
from funzioniSupplementari import positionSource
POSSOURCE = positionSource(filesource)
del positionSource
CSPREAL = []
POSTEMP = []
for i in range(len(POSSOURCE)):
POSTEMP.append(POSSOURCE[i])
POSSOURCE = np.asarray(POSTEMP)
del POSTEMP
getCSP(l, lun, N, fi, AUDIO16, window, CSPREAL)
CSPREAL = np.asarray(CSPREAL)
CSPTEMP = organizeGCC(CSPREAL, fi, kmax, l)
del CSPREAL
VAD_tr_GCC1 = []
for i in range(l):
MatrixInput1 = np.zeros(shape=(len(fi), context, kmax))
n = 0
while n < context:
t = 0
if (i - numContext + n <
0) or (i - numContext + n) >= l:
t = i
else:
t = i - numContext + n
for j in range(len(fi)):
MatrixInput1[j, n] = CSPTEMP[t, j]
n = n + 1
VAD_tr_GCC1.append(MatrixInput1)
VAD_tr_GCC1 = np.asarray(VAD_tr_GCC1)
SLOC_testx = model.predict([VAD_tr_GCC1])
POSSLOC = []
for i in range(len(SLOC_testx)):
Px = SLOC_testx[i][0] * float(ax)
Py = SLOC_testx[i][1] * float(ay)
POSSLOC.append([Px, Py, 1500])
POSSLOC = np.asarray(POSSLOC)
s = pathsave + realorsim + '_' + devortest + '_' + lan + '_' + str(
int(numerocartelle)
) + '_' + nameplace + '_' + 'oracle' + '_' + 'singlechannel'
locfile = s + '.loc'
loc = open(locfile, 'w')
cPickle.dump(POSSLOC, loc)
loc.close()
del model
def startSimulationJoint(context, CNNkernel, sizekernelCNN, stridesCNN,
DenseNeuron, DropoutLayer, nameplace, dir_training,
ax, ay, numContext, dirAudio2, fi, mic_log_mel, hsc,
fittizio, simulatedcheck, realcheck,
numberdir_fittizio, pathdir_fittizio):
pathsave = 'Joint' + dir_training + '/'
pathsave = pathsave + 'context' + str(context) + '/'
pathsave = pathsave + nameplace + '/'
pathsave = pathsave + 'CNNkernel' + str(int(CNNkernel[0])) + '/'
pathsave = pathsave + 'sizeKernel' + str(int(sizekernelCNN)) + '/'
pathsave = pathsave + 'strides' + str(int(stridesCNN)) + '/'
pathsave = pathsave + 'Densenumber' + str(int(len(DenseNeuron))) + '/'
pathsave = pathsave + 'Denseneuron' + str(int(DenseNeuron[0])) + '/'
pathsave = pathsave + 'Dropout' + str(int(len(DropoutLayer))) + '/'
if not os.path.exists(pathsave):
os.makedirs(pathsave)
VAD_tr_GCC1 = []
SLOC_tr_GCC_X = []
VAD_tr_LOG1 = []
SLOC_tr_LOG_X = []
from getDatasetJoint import getHSCMADevDataset
if hsc:
from getDatasetJoint import getHSCMADevDataset
getHSCMADevDataset(VAD_tr_LOG1, VAD_tr_GCC1, SLOC_tr_LOG_X,
SLOC_tr_GCC_X, ax, ay, context, numContext,
dirAudio2, fi, nameplace, mic_log_mel)
if simulatedcheck:
from getDatasetJoint import getEVALITADataset
getEVALITADataset(VAD_tr_LOG1, VAD_tr_GCC1, SLOC_tr_LOG_X,
SLOC_tr_GCC_X, ax, ay, realcheck, context,
numContext, dirAudio2, fi, nameplace, mic_log_mel)
if fittizio:
from getDatasetJoint import getDLSDataset
getDLSDataset(VAD_tr_LOG1, VAD_tr_GCC1, SLOC_tr_LOG_X, SLOC_tr_GCC_X,
ax, ay, numberdir_fittizio, pathdir_fittizio, context,
numContext, dirAudio2, fi, nameplace, mic_log_mel)
VAD_tr_GCC1 = np.asarray(VAD_tr_GCC1)
SLOC_tr_GCC_X = np.asarray(SLOC_tr_GCC_X)
VAD_tr_LOG1 = np.asarray(VAD_tr_LOG1)
SLOC_tr_LOG_X = np.asarray(SLOC_tr_LOG_X)
indexes = np.random.permutation(VAD_tr_GCC1.shape[0])
VAD_tr_GCC = VAD_tr_GCC1[indexes]
del VAD_tr_GCC1
print VAD_tr_GCC.shape
SLOC_tr_GCC = SLOC_tr_GCC_X[indexes]
del SLOC_tr_GCC_X
VAD_tr_LOG = VAD_tr_LOG1[indexes]
del VAD_tr_LOG1
SLOC_tr_LOG = SLOC_tr_LOG_X[indexes]
del SLOC_tr_LOG_X, indexes
from getModel import getModelJoint
model = getModelJoint(CNNkernel, sizekernelCNN, DenseNeuron, DropoutLayer,
stridesCNN, fi, context, mic_log_mel, VAD_tr_LOG,
VAD_tr_GCC, SLOC_tr_LOG, SLOC_tr_GCC, pathsave,
nameplace)
from constants import *
import scipy.io.wavfile as wav
import scipy
from csp2dgpp import getCSP, organizeGCC
import cPickle
from python_speech_features import logfbank
import sklearn
from sklearn import mixture
dev_test_array = ['Dev', 'Test']
dir_dt_array = ['HSCMA_DIRHA_dev', 'HSCMA_DIRHA_test']
dirrslower = dir_real_sim_array[1]
for dirdt in dev_test_array:
print dirdt
if dirdt == 'Dev':
dirdtupper = dev_test_array[0]
dirdt_hs = dir_dt_array[0]
devortest = 'dev'
else:
dirdtupper = dev_test_array[1]
dirdt_hs = dir_dt_array[1]
devortest = 'test'
realorsim = dir_real_sim_array[1]
dirrs = real_sim_array[1]
lan_array = ['IT', 'GK', 'GE', 'PT']
numberdir_hscma = 10
numberdirtraining_hscma = np.linspace(1, numberdir_hscma,
numberdir_hscma)
for lan in lan_array:
print lan
dirAudio = pathdir_hscma + dirdt_hs + '/' + dirrs + '/' + dirdtupper + '/' + lan + '/'
for numerocartelle in numberdirtraining_hscma:
print 'testing number: ' + str(int(numerocartelle))
dira = dirAudio + dirrslower + str(
int(numerocartelle)) + dirAudio2
from funzioniSupplementari import getMic
AUDIO = getMic(nameplace, dira)
del getMic
AUDIO16 = []
for i in range(0, mic_log_mel):
AUDIO16.append(scipy.signal.decimate(AUDIO[i], 48 / 16))
l = len(AUDIO16[0]) / hopframe
if realorsim == 'sim' or (realorsim == 'real' and
(nameplace == 'Livingroom'
or nameplace == 'Kitchen')):
filesource = dirAudio + dirrslower + str(
int(numerocartelle)
) + '/' + dirSource2 + '/' + nameplace + formatref
from funzioniSupplementari import positionSource
POSSOURCE = positionSource(filesource)
del positionSource
CSPREAL = []
POSTEMP = []
for i in range(len(POSSOURCE)):
POSTEMP.append(POSSOURCE[i])
POSSOURCE = np.asarray(POSTEMP)
del POSTEMP
getCSP(l, lun, N, fi, AUDIO16, window, CSPREAL)
CSPREAL = np.asarray(CSPREAL)
CSPTEMP = organizeGCC(CSPREAL, fi, kmax, l)
del CSPREAL
LOGMEL = []
for i in range(0, mic_log_mel):
vector = logfbank(AUDIO16[i],
rate,
nfilt=numfilter,
winlen=(float(lenframe) / float(rate)),
winstep=(float(hopframe) / float(rate)),
nfft=lenframe)
vector = sklearn.preprocessing.normalize(vector)
LOGMEL.append(vector)
VAD_tr_LOG1 = []
VAD_tr_GCC1 = []
for i in range(l):
MatrixInputLOG = np.zeros(shape=(mic_log_mel, context,
numfilter))
MatrixInputGCC = np.zeros(shape=(len(fi), context, kmax))
n = 0
while n < context:
t = 0
if (i - numContext + n <
0) or (i - numContext + n) >= l:
t = i
else:
t = i - numContext + n
for j in range(mic_log_mel):
MatrixInputLOG[j, n] = LOGMEL[j][t]
for j in range(len(fi)):
MatrixInputGCC[j, n] = CSPTEMP[t, j]
n = n + 1
VAD_tr_LOG1.append(MatrixInputLOG)
VAD_tr_GCC1.append(MatrixInputGCC)
VAD_tr_GCC1 = np.asarray(VAD_tr_GCC1)
VAD_tr_LOG1 = np.asarray(VAD_tr_LOG1)
[VAD_output,
SLOC_OUTPUT] = model.predict([VAD_tr_LOG1, VAD_tr_GCC1])
POSSLOC = []
for i in range(len(SLOC_OUTPUT)):
Px = SLOC_OUTPUT[i][0]
Py = SLOC_OUTPUT[i][1]
POSSLOC.append([Px, Py, 1500])
POSSLOC = np.asarray(POSSLOC)
VAD_output = np.asarray(VAD_output)
VAD_output = VAD_output[:, 1]
s = pathsave + realorsim + '_' + devortest + '_' + lan + '_' + str(
int(numerocartelle)
) + '_' + nameplace + '_' + 'oracle' + '_' + 'slocjoint'
locfile = s + '.loc'
loc = open(locfile, 'w')
cPickle.dump(POSSLOC, loc)
loc.close()
s = pathsave + realorsim + '_' + devortest + '_' + lan + '_' + str(
int(numerocartelle)) + '_' + nameplace + '_' + 'vadjoint'
vadfile = s + '.spk'
vad = open(vadfile, 'w')
cPickle.dump(VAD_output, vad)
vad.close()
del model
def getHSCMADevDataset(VAD_tr_GCC1, SLOC_tr_GCC_X, ax, ay, context, numContext,
dirAudio2, fi, nameplace, mic_log_mel):
dev_test_array = ['Dev']
dir_dt_array = ['HSCMA_DIRHA_dev', 'HSCMA_DIRHA_test']
dirrslower = dir_real_sim_array[1]
for dirdt in dev_test_array:
print dirdt
if dirdt == 'Dev':
dirdtupper = dev_test_array[0]
dirdt_hs = dir_dt_array[0]
else:
dirdtupper = dev_test_array[1]
dirdt_hs = dir_dt_array[1]
realorsim = dir_real_sim_array[1]
dirrs = real_sim_array[1]
lan_array = ['IT', 'GK', 'GE', 'PT']
numberdir_hscma = 10
numberdirtraining_hscma = np.linspace(1, numberdir_hscma,
numberdir_hscma)
for lan in lan_array:
print lan
dirAudio = pathdir_hscma + dirdt_hs + '/' + dirrs + '/' + dirdtupper + '/' + lan + '/'
for numerocartelle in numberdirtraining_hscma:
print 'training number: ' + str(int(numerocartelle))
dira = dirAudio + dirrslower + str(
int(numerocartelle)) + dirAudio2
from funzioniSupplementari import getMic
AUDIO = getMic(nameplace, dira)
del getMic
AUDIO16 = []
for i in range(0, mic_log_mel):
AUDIO16.append(scipy.signal.decimate(AUDIO[i], 48 / 16))
l = len(AUDIO16[0]) / hopframe
if realorsim == 'sim' or (realorsim == 'real' and
(nameplace == 'Livingroom'
or nameplace == 'Kitchen')):
filesource = dirAudio + dirrslower + str(
int(numerocartelle)
) + '/' + dirSource2 + '/' + nameplace + formatref
from funzioniSupplementari import positionSource
POSSOURCE = positionSource(filesource)
del positionSource
CSPREAL = []
POSTEMP = []
for i in range(len(POSSOURCE)):
POSTEMP.append(POSSOURCE[i])
POSSOURCE = np.asarray(POSTEMP)
del POSTEMP
getCSP(l, lun, N, fi, AUDIO16, window, CSPREAL)
CSPREAL = np.asarray(CSPREAL)
CSPTEMP = organizeGCC(CSPREAL, fi, kmax, l)
del CSPREAL
for i in range(l):
if POSSOURCE[i][0] != 0 or POSSOURCE[i][
1] != 0 or POSSOURCE[i][2] != 0:
MatrixInput1 = np.zeros(shape=(len(fi), context, kmax))
n = 0
while n < context:
t = 0
if (i - numContext + n <
0) or (i - numContext + n) >= l:
t = i
else:
t = i - numContext + n
for j in range(len(fi)):
MatrixInput1[j, n] = CSPTEMP[t, j]
n = n + 1
for j in range(len(fi)):
VAD_tr_GCC1[j].append(MatrixInput1[j])
SLOC_tr_GCC_X.append([
float(POSSOURCE[i][0]) / ax,
float(POSSOURCE[i][1]) / ay
])
def getEVALITADataset(VAD_tr_GCC1, SLOC_tr_GCC_X, ax, ay, realcheck, context,
numContext, dirAudio2, fi, nameplace, mic_log_mel):
for realorsim in dir_real_sim_array:
if (realorsim == 'real' and realcheck == True) or realorsim == 'sim':
print realorsim
dirrslower = realorsim
if realorsim == 'real':
dirrs = real_sim_array[0]
numberdir = 22
numberdirtraining = np.linspace(1, numberdir, numberdir)
elif realorsim == 'sim':
dirrs = real_sim_array[1]
numberdir = 80
numberdirtraining = np.linspace(1, numberdir, numberdir)
for numerocartelle in numberdirtraining:
#for numerocartelle in range(1,5):
print 'training folder number: ' + str(numerocartelle)
dira = pathdir + dirrs + '/' + dirrslower + str(
int(numerocartelle)) + dirAudio2
from funzioniSupplementari import getMic
AUDIO = getMic(nameplace, dira)
del getMic
AUDIO16 = []
for i in range(0, mic_log_mel):
AUDIO16.append(scipy.signal.decimate(AUDIO[i], 48 / 16))
l = len(AUDIO16[0]) / hopframe
if realorsim == 'sim' or (realorsim == 'real' and
(nameplace == 'Livingroom'
or nameplace == 'Kitchen')):
filesource = pathdir + dirrs + '/' + dirrslower + str(
int(numerocartelle)
) + '/' + dirSource2 + '/' + nameplace + formatref
from funzioniSupplementari import positionSource
POSSOURCE = positionSource(filesource)
del positionSource
CSPREAL = []
POSTEMP = []
for i in range(len(POSSOURCE)):
POSTEMP.append(POSSOURCE[i])
POSSOURCE = np.asarray(POSTEMP)
del POSTEMP
getCSP(l, lun, N, fi, AUDIO16, window, CSPREAL)
CSPREAL = np.asarray(CSPREAL)
CSPTEMP = organizeGCC(CSPREAL, fi, kmax, l)
del CSPREAL
for i in range(l):
if POSSOURCE[i][0] != 0 or POSSOURCE[i][
1] != 0 or POSSOURCE[i][2] != 0:
MatrixInput1 = np.zeros(shape=(len(fi), context, kmax))
n = 0
while n < context:
t = 0
if (i - numContext + n <
0) or (i - numContext + n) >= l:
t = i
else:
t = i - numContext + n
for j in range(len(fi)):
MatrixInput1[j, n] = CSPTEMP[t, j]
n = n + 1
for j in range(len(fi)):
VAD_tr_GCC1[j].append(MatrixInput1[j])
SLOC_tr_GCC_X.append([
float(POSSOURCE[i][0]) / ax,
float(POSSOURCE[i][1]) / ay
])
def getEVALITADataset(VAD_tr_LOG1,VAD_tr_GCC1,SLOC_tr_LOG_X,SLOC_tr_GCC_X,ax,ay,realcheck,context,numContext,dirAudio2,fi,nameplace,mic_log_mel):
for realorsim in dir_real_sim_array:
if (realorsim=='real' and realcheck==True) or realorsim=='sim':
print realorsim
dirrslower=realorsim
if realorsim=='real':
dirrs=real_sim_array[0]
numberdir=22
numbertraining=numberdir
numberdirtraining=np.linspace(1,numberdir,numberdir)
elif realorsim=='sim':
dirrs=real_sim_array[1]
numberdir=80
numberdirtraining=np.linspace(1,numberdir,numberdir)
for numerocartelle in numberdirtraining:
#for numerocartelle in range(1,15):
print 'training folder number: '+str(numerocartelle)
dira=pathdir+dirrs+'/'+dirrslower+str(int(numerocartelle))+dirAudio2
from funzioniSupplementari import getMic
AUDIO=getMic(nameplace,dira)
del getMic
AUDIO16=[]
for i in range(0,mic_log_mel):
AUDIO16.append(scipy.signal.decimate(AUDIO[i],48/16))
l=len(AUDIO16[0])/hopframe
if realorsim=='sim' or (realorsim=='real' and (nameplace=='Livingroom' or nameplace=='Kitchen')):
filesource=pathdir+dirrs+'/'+dirrslower+str(int(numerocartelle))+'/'+dirSource2+'/'+nameplace+formatref
from funzioniSupplementari import positionSource
POSSOURCE=positionSource(filesource)
del positionSource
CSPREAL=[]
POSTEMP=[]
for i in range(len(POSSOURCE)):
POSTEMP.append(POSSOURCE[i])
POSSOURCE=np.asarray(POSTEMP)
del POSTEMP
getCSP(l,lun,N,fi,AUDIO16,window,CSPREAL)
CSPREAL=np.asarray(CSPREAL)
CSPTEMP=organizeGCC(CSPREAL,fi,kmax,l)
del CSPREAL
LOGMEL=[]
for i in range(0,mic_log_mel):
vector=logfbank(AUDIO16[i],rate,nfilt=numfilter,winlen=(float(lenframe)/float(rate)),winstep=(float(hopframe)/float(rate)),nfft=lenframe)
vector=sklearn.preprocessing.normalize(vector)
LOGMEL.append(vector)
for i in range(l):
MatrixInputLOG=np.zeros(shape=(mic_log_mel,context,numfilter))
MatrixInputGCC=np.zeros(shape=(len(fi),context,kmax))
n=0
while n<context:
t=0
if (i-numContext+n<0) or (i-numContext+n)>=l:
t=i
else:
t=i-numContext+n
for j in range(mic_log_mel):
MatrixInputLOG[j,n]=LOGMEL[j][t]
for j in range(len(fi)):
MatrixInputGCC[j,n]=CSPTEMP[t,j]
n=n+1
VAD_tr_LOG1.append(MatrixInputLOG)
VAD_tr_GCC1.append(MatrixInputGCC)
if POSSOURCE[i][0]!=0 or POSSOURCE[i][1]!=0 or POSSOURCE[i][2]!=0:
SLOC_tr_GCC_X.append([float(POSSOURCE[i][0])/ax,float(POSSOURCE[i][1])/ay])
SLOC_tr_LOG_X.append([0,1])
else:
SLOC_tr_LOG_X.append([1,0])
SLOC_tr_GCC_X.append([-1,-1])