def cocktail(self):
eps = 0.00000001
rate1, data1 = wavfile.read(self.path1)
rate2, data2 = wavfile.read(self.path2)
if(data1.ndim != 1 or data2.ndim != 1):
self.alarm("Please Chose another file with 1D data")
else:
data1 = data1 - np.mean(data1)
data1 = data1/max(data1)
data2 = data2 - np.mean(data2)
data2 = data2/max(data2)
signals = [data1, data2]
matrix = np.vstack(signals)
whiteMatrix = utl.whitenMatrix(matrix)
X = whiteMatrix
vectors = []
for i in range(0, X.shape[0]):
vector = FA(X, vectors, eps)
vectors.append(vector)
W = np.vstack(vectors)
S = np.dot(W, whiteMatrix)
utl.plotSounds([S[0], S[1]], ["source_1", "source_2"], rate1, "plot_results/cocktail_party/song_separation_plot.png")
wavfile.write("sound_results/cocktail_party/source1.wav" ,rate1, 5000*S[0].astype(np.int16))
wavfile.write("sound_results/cocktail_party/source2.wav" , rate1, 5000*S[1].astype(np.int16))
img = pg.QtGui.QGraphicsPixmapItem(pg.QtGui.QPixmap('plot_results/cocktail_party/song_separation_plot.png'))
self.ui2.widget_party.addItem(img)
self.ui2.widget_party.invertY(True)
self.alarm("Check Plot & Sound Results Files")
# Read the mixed signals
rate1, data1 = wavfile.read('./mixed' + name[0] + '.wav')
rate2, data2 = wavfile.read('./mixed' + name[1] + '.wav')
# Centering the mixed signals and scaling the values as well
data1 = data1 - np.mean(data1)
data1 = data1 / 32768
data2 = data2 - np.mean(data2)
data2 = data2 / 32768
# Creating a matrix out of the signals
signals = [data1, data2]
matrix = np.vstack(signals)
# Whitening the matrix as a pre-processing step
whiteMatrix = utils.whitenMatrix(matrix)
X = whiteMatrix
# Find the individual components one by one
vectors = []
for i in range(0, X.shape[0]):
# The FastICA function is used as is from FastICA_image.py, and the it works out of the box
vector = FastICA(X, vectors, eps)
vectors.append(vector)
# Stack the vectors to form the unmixing matrix
W = np.vstack(vectors)
# Get the original matrix
S = np.dot(W, whiteMatrix)
eps = 0.00000001
# Read the images from ./images/mixed
names = ["unos", "dos", "tres"]
images = utl.listImages(names, "mixed")
# The images are mean centered
centImages = []
for image in images:
rescaleImage = image
centImage = rescaleImage - np.mean(rescaleImage)
centImages.append(centImage)
# The images are whitened, the helper function is in utilities.py
whiteImages = utl.whitenMatrix(utl.list2matrix(centImages))
# Uncomment the lines below to plot the images after whitening
# utl.plotImages(utl.matrix2list(whiteImages), names, "../white_tranform", True, False)
# The images are now converted into time series data
# X is a 3*image_size matrix, with each row representing a image
X = whiteImages
# Find the individual components one by one
vectors = []
for i in range(0, len(images)):
vector = FastICA(X, vectors, eps)
# print vector
vectors.append(vector)
plt.title("data2")
plt.show()
# Creating a matrix out of the signals
signals = [data1, data2]
plt.figure()
plt.plot(signals[0], signals[1], '*b')
plt.ylabel('Signal 2')
plt.xlabel('Signal 1')
plt.title("Original data")
matrix = np.vstack(signals)
# Whitening the matrix as a pre-processing step
whiteMatrix = utl.whitenMatrix(matrix)
X = whiteMatrix
#Display the whitened matrix
plt.figure()
plt.plot(X[0], X[1], '*b')
plt.ylabel('Signal 2')
plt.xlabel('Signal 1')
plt.title("Whitened data")
plt.show()
# Find the individual components one by one
vectors = []
for i in range(0, X.shape[0]):
def separate_sound3(b, name):
name = ["1", "2", "3"]
eps = 0.00000001
#Read the mixed signals
rate1, data1 = wavfile.read(b[0])
rate2, data2 = wavfile.read(b[1])
rate3, data3 = wavfile.read(b[2])
#Centering the mixed signals and scaling the values as well
data1 = data1 - np.mean(data1)
data1 = data1 / 32768
data2 = data2 - np.mean(data2)
data2 = data2 / 32768
data3 = data3 - np.mean(data3)
data3 = data3 / 32768
#Creating a matrix out of the signals
signals = [data1, data2, data3]
matrix = np.vstack(signals)
#Whitening the matrix as a pre-processing step
whiteMatrix = utl.whitenMatrix(matrix)
X = whiteMatrix
#Find the individual components one by one
vectors = []
for i in range(0, X.shape[0]):
#The FastICA function is used as is from FastICA_image.py, and the it works out of the box
vector = FastICA(X, vectors, eps)
vectors.append(vector)
#Stack the vectors to form the unmixing matrix
W = np.vstack(vectors)
#Get the original matrix
S = np.dot(W, whiteMatrix)
#Unmixing matrix through FOBI
fobiW = FOBI(X)
#Get the original matrix using fobiW
fobiS = np.dot(fobiW.T, whiteMatrix)
#Plot the separated sound signals
utl.plotSounds([S[0], S[1], S[2]], ["1", "2", "3"], rate1, "separated")
#Write the separated sound signals, 5000 is multiplied so that signal is audible
wavfile.write(
"C:/SPEECH_SEPARATION--FinalYear/Separated/separate-" + name[0] +
".wav", rate1, 5000 * S[0].astype(np.int16))
wavfile.write(
"C:/SPEECH_SEPARATION--FinalYear/Separated/separate-" + name[1] +
".wav", rate1, 5000 * S[1].astype(np.int16))
wavfile.write(
"C:/SPEECH_SEPARATION--FinalYear/Separated/separate-" + name[2] +
".wav", rate1, 5000 * S[2].astype(np.int16))
#Plot the separated sound signals
utl.plotSounds([fobiS[1], fobiS[0], fobiS[2]], ["1", "2", "3"], rate1,
"separated_FOBI")
#Write the separated sound signals, 5000 is multiplied so that signal is audible
wavfile.write(
"C:/SPEECH_SEPARATION--FinalYear/Separated/separate-" + name[0] +
".wav", rate1, 5000 * fobiS[1].astype(np.int16))
wavfile.write(
"C:/SPEECH_SEPARATION--FinalYear/Separated/separate-" + name[1] +
".wav", rate1, 5000 * fobiS[0].astype(np.int16))
wavfile.write(
"C:/SPEECH_SEPARATION--FinalYear/Separated/separate-" + name[2] +
".wav", rate3, 5000 * fobiS[2].astype(np.int16))