def prep_whole_dividable(path_load, path_save, firstdim, secdim):
import string
import os
import numpy as np
import scipy.io
import math
from tqdm import tqdm
# --------------------------------------------------------------
# Get files from path
path_save_one = path_save
for p in range(0,3):
if p == 0:
path_load_real = path_load + '/Train'
path_save_real = path_save_one + '/Train'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
elif p == 1:
path_load_real = path_load + '/Validation'
path_save_real = path_save_one + '/Validation'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
elif p == 2:
path_load_real = path_load + '/Test'
path_save_real = path_save_one + '/Test'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
files_all = os.listdir(path_load_real)
files_all = sorted(files_all,key=lambda x: ((int(x.split("_")[2])),(int(x.split("_")[3])),(int(x.split("_")[4])),(x.split("_")[5])))
print(files_all)
for file in tqdm(files_all):
# for files in range(0,len(files_all),skipcount):
i = files_all.index(file)
# print(i)
fourierspec2 = []
labelpsd1 = []
labelpsd2 = []
# for file in tqdm(files_all[files:files+skipcount]):
path_loadcontents = path_load_real + '/' + file
contents = np.load(path_loadcontents)
fourierspec = []
labelpsd = []
fourier = contents['dbmixed']
fourier = fourier[:firstdim,:secdim]
fourierspec2.append(fourier)
label1 = contents['dbspeech']
labelpsd1.append(label1[:firstdim,:secdim])
label2 = contents['dbmusic']
labelpsd2.append(label2[:firstdim,:secdim])
# if option == 'music':
# label = contents['dbmusic']
# elif option == 'speech':
# label = contents['dbspeech']
# else:
# print('option not right!')
#
# label = label[:firstdim,:secdim]
#
# labelpsd2.append(label)
train_data = np.array(fourierspec2)
train_data = np.reshape(train_data, (train_data.shape[0], train_data.shape[1], train_data.shape[2], 1))
# print(train_data.shape)
train_labels_s = np.array(labelpsd1)
train_labels_s = np.reshape(train_labels_s, (train_labels_s.shape[0], train_labels_s.shape[1], train_labels_s.shape[2], 1))
# print(train_labels_s.shape)
train_labels_m = np.array(labelpsd2)
train_labels_m = np.reshape(train_labels_m, (train_labels_m.shape[0], train_labels_m.shape[1], train_labels_m.shape[2], 1))
# print(train_labels_m.shape)
path_savecontents = path_save_real + '/Processed_' + str(i) + '_whole_dividable.npz'
# print(path_savecontents)
np.savez(path_savecontents, x=train_data, y=train_labels_s, z=train_labels_m)
return path_save_one
def prep_flatten_1weiter(path_load, path_save, skipcount, framelength, option):
import os
import numpy as np
import scipy.io
import math
from tqdm import tqdm
# -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
path_save_one = path_save
for p in range(0,3):
if p == 0:
path_load_real = path_load + '/Train'
path_save_real = path_save_one + '/Train'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
elif p == 1:
path_load_real = path_load + '/Validation'
path_save_real = path_save_one + '/Validation'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
elif p == 2:
path_load_real = path_load + '/Test'
path_save_real = path_save_one + '/Test'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
# Get files from path
files_all = os.listdir(path_load_real)
files_all = sorted(files_all)
for files in range(0,len(files_all),skipcount):
fourierspec2 = []
labelpsd2 = []
for file in tqdm(files_all[files:files+skipcount]):
path_loadcontents = path_load_real + '/' + file
contents = np.load(path_loadcontents)
fourierspec = []
labelpsd = []
fourier = contents['dbmixed']
for n in list(range(0, (math.floor(fourier.shape[1]/framelength)-1)*framelength+1)):
fourierlil = fourier[:,n:n+framelength]
fourierlil = np.transpose(fourierlil)
fourierlil = np.reshape(fourierlil, (fourier.shape[0]*framelength))
fourierspec.append(fourierlil)
fourierspec2.extend(fourierspec)
if option == 'music':
label = contents['dbmusic']
elif option == 'speech':
label = contents['dbspeech']
else:
print('option not right!')
for n in list(range(0, (math.floor(fourier.shape[1]/framelength)-1)*framelength+1)):
labellil = label[:,n+framelength-1]
labelpsd.append(labellil)
labelpsd2.extend(labelpsd)
train_data = np.array(fourierspec2)
train_labels = np.array(labelpsd2)
path_savecontents = path_save_real + '/Processed_' + str(files) + '_' + str(train_data.shape[0]) + '_flatten_1weiter.npz'
np.savez(path_savecontents, x=train_data, y=train_labels)
return path_save_one
def prep_flatten_4weiter(path_load, path_save, skipcount, framelength, option):
import os
import numpy as np
import scipy.io
import math
from tqdm import tqdm
# -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
path_save_one = path_save
for p in range(0,3):
if p == 0:
path_load_real = path_load + '/Train'
path_save_real = path_save_one + '/Train'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
elif p == 1:
path_load_real = path_load + '/Validation'
path_save_real = path_save_one + '/Validation'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
elif p == 2:
path_load_real = path_load + '/Test'
path_save_real = path_save_one + '/Test'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
# Get files from path
files_all = os.listdir(path_load_real)
files_all = sorted(files_all)
for files in tqdm(range(0,len(files_all),skipcount)):
fourierspec2 = []
labelpsd2 = []
for file in tqdm(files_all[files:files+skipcount]):
path_loadcontents = path_load_real + '/' + file
contents = np.load(path_loadcontents)
fourierspec = []
labelpsd = []
fourier = contents['dbmixed']
print(fourier.shape)
# if p == 0 & files == 0:
# diffcount = math.floor(fourier.shape[1]/framelength)
# return diffcount
for n in list(range(0, math.floor(fourier.shape[1]/framelength))):
fourierlil = fourier[:,framelength*n:framelength*n+framelength]
fourierlil = np.transpose(fourierlil)
fourierlil = np.reshape(fourierlil, (fourier.shape[0]*framelength))
print(np.size(fourierlil))
fourierspec.append(fourierlil)
print(np.size(fourierspec))
fourierspec2.extend(fourierspec)
if option == 'music':
label = contents['dbmusic']
elif option == 'speech':
label = contents['dbspeech']
else:
print('option not right!')
for n in list(range(0, math.floor(label.shape[1]/framelength))):
labellil = label[:,framelength*n+framelength-1]
print(np.size(labellil))
labelpsd.append(labellil)
print(np.size(labelpsd))
labelpsd2.extend(labelpsd)
train_data = np.array(fourierspec2)
train_labels = np.array(labelpsd2)
path_savecontents = path_save_real + '/Processed_' + str(files) + '_' + str(train_data.shape[0]) + '_' + str(framelength) + '_flatten.npz'
np.savez(path_savecontents, x=train_data, y=train_labels)
return path_save_one
def prep_framelength(path_load, path_save, framelength):
import os
import numpy as np
import scipy.io
import math
from tqdm import tqdm
# -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
path_save_one = path_save
for p in range(0,3):
if p == 0:
path_load_real = path_load + '/Train'
path_save_real = path_save_one + '/Train'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
elif p == 1:
path_load_real = path_load + '/Validation'
path_save_real = path_save_one + '/Validation'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
elif p == 2:
path_load_real = path_load + '/Test'
path_save_real = path_save_one + '/Test'
if not os.path.exists(path_save_real):
os.makedirs(path_save_real)
# Get files from path
files_all = os.listdir(path_load_real)
files_all = sorted(files_all,key=lambda x: ((int(x.split("_")[2])),(int(x.split("_")[3])),(int(x.split("_")[4])),(x.split("_")[5])))
# files_all = sorted(files_all,key=lambda x: int(x.split("_",2)[2]))
print(files_all)
for file in tqdm(files_all):
i = files_all.index(file)
# print(i)
# fourierspec2 = []
# labelpsd2 = []
path_loadcontents = path_load_real + '/' + file
contents = np.load(path_loadcontents)
fourierspec = []
labelpsd1 = []
labelpsd2 = []
fourier = contents['dbmixed']
for n in list(range(fourier.shape[1]-(framelength-1))):
fourierlil = fourier[:,n:n+framelength]
fourierspec.append(fourierlil)
# print(fourierlil)
# print(np.array(fourierspec).shape)
# if option == 'music':
# label = contents['dbmusic']
# elif option == 'speech':
# label = contents['dbspeech']
# else:
# print('option not right!')
label_s = contents['dbspeech']
for n in list(range(fourier.shape[1]-(framelength-1))):
labellil1 = label_s[:,n+framelength-1]
labelpsd1.append(labellil1)
label_m = contents['dbmusic']
for n in list(range(fourier.shape[1]-(framelength-1))):
labellil2 = label_m[:,n+framelength-1]
labelpsd2.append(labellil2)
# print(labellil)
# print(np.array(labelpsd).shape)
train_data = np.array(fourierspec)
train_data = np.reshape(train_data, (train_data.shape[0], train_data.shape[1], train_data.shape[2], 1))
# print(train_data.shape)
train_labels_s = np.array(labelpsd1)
train_labels_s = np.reshape(train_labels_s, (train_labels_s.shape[0], train_labels_s.shape[1], 1, 1))
# print(train_labels_s.shape)
train_labels_m = np.array(labelpsd2)
train_labels_m = np.reshape(train_labels_m, (train_labels_m.shape[0], train_labels_m.shape[1], 1, 1))
# print(train_labels_m.shape)
path_savecontents = path_save_real + '/Processed_' + str(i) + '_' + str(framelength) + '_frames.npz'
# print(path_savecontents)
np.savez(path_savecontents, x=train_data, y=train_labels_s, z=train_labels_m)
return path_save_one