def evaluate(original_file, test_file):
ori_arr, factors = read_file_test(original_file, 13, "factors")
test_arr = read_file_test(test_file, 13)
ori_arr = ori_arr[:,1:13] * factors #normalise_mfcc(ori_arr) # take 12 coefficients
#print ori_arr
test_arr = test_arr[:,1:13]
#print test_arr
#exit()
subtract = (ori_arr - test_arr) ** 2
rms = sqrt(subtract.sum()) #/(ori_arr.shape[0] * ori_arr.shape[1])
[m,n] = ori_arr.shape
a = ori_arr.reshape(1,m*n)
b = test_arr.reshape(1,m*n)
return np.corrcoef(a,b)[0,1]
return rms
def _testing_original_file(self, test_dir, type_data):
dnn_predict_dir = '/home/danglab/Results/PR_ver2/' + type_data + 'no_EMA/'+'test_' + str(self.test_number) + '/'
if not os.path.exists(dnn_predict_dir):
os.makedirs(dnn_predict_dir)
for afile in self.missing_filename_list:
test_arr, factors = read_file_test(test_dir + afile + '_in.txt', self.feature_dim_in, "factors") #read a missing_feature
#print factors
energy = test_arr[:,0] #ko cho energy vao DNN
test_arr = test_arr[:,1:self.feature_dim_in]
self._write_predict_missing(dnn_predict_dir + afile + '.txt', energy, test_arr, factors) # write result to file
def _testing_aritc_interpolation(self, test_dir):
dnn_predict_dir = '/home/danglab/Results/New_PR/' + 'space/artic_inter/'+'test_' + str(self.test_number) + '/10ms/'
if not os.path.exists(dnn_predict_dir):
os.makedirs(dnn_predict_dir)
for afile in self.missing_filename_list:
test_arr, factors = read_file_test(test_dir + afile + '_10ms_in.txt', self.feature_dim_in, "factors") #read a missing_feature
#print factors
energy = test_arr[:,0] #ko cho energy vao DNN
test_arr = test_arr[:,1:self.feature_dim_in]
test_arr[:,12:self.feature_dim_in] = 0 # remove articulatory data
self._write_predict_2_file(dnn_predict_dir + afile + '.txt', energy, self.predict(test_arr), factors) # write result to file
def evaluate(original_file, test_file):
#print original_file
#print test_file
ori_arr, factors = read_file_test(original_file, 13, "factors")
test_arr = read_file_test(test_file, 13)
[m,n] = ori_arr.shape
#print m,n
ori_arr = ori_arr[:,1:13] * factors #normalise_mfcc(ori_arr) # take 12 coefficients
test_arr = test_arr[:,1:13]
subtract = (ori_arr - test_arr) ** 2
#subtract = (ori_arr[2:m-3,:] - test_arr[2:m-3,:]) ** 2
rms = sqrt(subtract.sum()) #/(ori_arr.shape[0] * ori_arr.shape[1])
#print
#print ori_arr[m-3]
#print test_arr[m-3]
#print subtract[m-3]
#exit()
#a = ori_arr.reshape(1,m*n)
#b = test_arr.reshape(1,m*n)
#return np.corrcoef(a,b)[0,1]
return rms
def _testing_original_file(self, test_dir, type_data):
dnn_predict_dir = '/home/danglab/Results/New_PR/' + type_data + 'full_EMA/'+'test_' + str(self.test_number) + '/'
if not os.path.exists(dnn_predict_dir):
os.makedirs(dnn_predict_dir)
for afile in self.missing_filename_list:
test_arr, factors = read_file_test(test_dir + afile + '_in.txt', self.feature_dim_in, "factors") #read a missing_feature
#print factors
energy = test_arr[:,0] #ko cho energy vao DNN
test_arr = test_arr[:,1:self.feature_dim_in]
if "full_EMA" not in dnn_predict_dir:
print "no_ema"
test_arr[:,12:self.feature_dim_in] = 0 # remove articulatory data
test_arr = self._stack_segment_test(test_arr)
print test_arr.shape
self._write_predict_2_file(dnn_predict_dir + afile + '.txt', energy, self.predict(test_arr), factors) # write result to file
def _testing_space(self, test_dir, type_data):
dnn_predict_dir = '/home/danglab/Results/New_PR/' + type_data + 'full_EMA/'+'test_' + str(self.test_number) + '/'
if not os.path.exists(dnn_predict_dir):
os.makedirs(dnn_predict_dir)
for afile in self.missing_filename_list:
test_arr, factors = read_file_test(test_dir + afile + '_20ms_in.txt', self.feature_dim_in, "factors") #read a missing_feature
#print factors
energy = test_arr[:,0] #ko cho energy vao DNN
test_arr = test_arr[:,1:self.feature_dim_in]
if "full_EMA" not in dnn_predict_dir:
print "no_ema"
test_arr[:,12:self.feature_dim_in] = 0 # remove articulatory data
test_arr = self._stack_segment_test(test_arr)
#position 0,2,4,6,... is missing audio, only test these parts
test_missing_part = test_arr[0:test_arr.shape[0]:2] # lay nhung frame bi mat du lieu
no_missing_part = test_arr[1:test_arr.shape[0]:2]
print test_missing_part.shape, no_missing_part.shape, test_arr.shape
self._write_predict_missing(dnn_predict_dir + afile + '.txt', energy, self.predict(test_missing_part), no_missing_part, factors) # write result to file
def _testing_noise_space(self, test_dir, type_data):
for type_test in sorted(os.listdir(test_dir)):
if (not type_test.endswith('zip')) and 'output' not in type_test:
type_test_dir = test_dir + type_test + '/'
print type_test_dir
dnn_predict_dir = '/home/danglab/Results/New_PR/' + type_data + 'SP_no_EMA/'+'test_' + str(self.test_number) + '/' + type_test + '/'
if not os.path.exists(dnn_predict_dir):
os.makedirs(dnn_predict_dir)
print type_test
duration = type_test.split('_')[1] # 50ms, 100ms
#listtest = sorted(os.listdir(type_test_dir))
#for afile in listtest:
for prefix_file in self.missing_filename_list:
afile = prefix_file + '_' + duration + '_in.txt'
test_arr, factors = read_file_test(type_test_dir + afile, self.feature_dim_in, "factors") #read a missing_feature
find_ = [m.start() for m in re.finditer('_', afile)]
energy = test_arr[:,0] #ko cho energy vao DNN
test_arr = test_arr[:,1:self.feature_dim_in]
#test_arr[:,12:self.n_input_f] = 0 # remove articulatory data
#print factors
self._write_predict_2_file(dnn_predict_dir + afile.replace(afile[find_[4]:len(afile)-4],''), energy, self.predict(test_arr), factors) # write result to file
def get_data(filename):
arr = read_file_test(filename, num_features) # 0:13, enegy+MFCC, 13:49, articulotory position
# position = arr[:,13:13 + 12] # position at i
return arr[:, 0], arr[:, 1:13], arr[:, 13 : 13 + 12]