def mixgauss_init(M, data, cov_type, method='kmeans'):
'''
% MIXGAUSS_INIT Initial parameter estimates for a mixture of Gaussians
% function [mu, Sigma, weights] = mixgauss_init(M, data, cov_type. method)
%
% INPUTS:
% data(:,t) is the t'th example
% M = num. mixture components
% cov_type = 'full', 'diag' or 'spherical'
% method = 'rnd' (choose centers randomly from data) or 'kmeans' (needs netlab)
%
% OUTPUTS:
% mu(:,k)
% Sigma(:,:,k)
% weights(k)
'''
if isinstance(data, list):
data = np.hstack(data)
elif data.ndim == 3:
O, T, N = data.shape
data = np.reshape(np.transpose(data, (0, 2, 1)), (O, T * N))
d, T = data.shape
if method == 'rnd':
C = np.atleast_2d(np.cov(data))
Sigma = np.transpose(np.tile(np.diag(np.diag(C)) * 0.5, (M, 1, 1)),
(2, 1, 0))
# Initialize each mean to a random data point
indices = np.arange(T)
np.random.shuffle(indices)
mu = data[:, indices[0:M]]
weights, _ = normalise(np.ones((M, 1)))
elif method == 'kmeans':
gmm = GMM(n_components=M,
covariance_type=cov_type,
thresh=1e-2,
min_covar=1e-3,
n_iter=5,
n_init=1,
params='wmc',
init_params='wmc')
gmm.fit(data.T)
mu = gmm.means_.T
weights = np.asmatrix(gmm.weights_).T
covars = gmm.covars_
Sigma = np.zeros((d, d, M))
for m in range(M):
if cov_type == 'diag':
Sigma[:, :, m] = np.diag(covars[m, :])
elif cov_type == 'full':
Sigma[:, :, m] = covars[:, :, m]
elif cov_type == 'spherical':
Sigma[:, :, m] = covars[m] * np.eye(d)
return mu, Sigma, weights
def mixgauss_init(M, data, cov_type, method='kmeans'):
'''
% MIXGAUSS_INIT Initial parameter estimates for a mixture of Gaussians
% function [mu, Sigma, weights] = mixgauss_init(M, data, cov_type. method)
%
% INPUTS:
% data(:,t) is the t'th example
% M = num. mixture components
% cov_type = 'full', 'diag' or 'spherical'
% method = 'rnd' (choose centers randomly from data) or 'kmeans' (needs netlab)
%
% OUTPUTS:
% mu(:,k)
% Sigma(:,:,k)
% weights(k)
'''
if isinstance(data, list):
data = np.hstack(data)
elif data.ndim==3:
O, T, N = data.shape
data = np.reshape(np.transpose(data, (0, 2, 1)), (O, T*N))
d, T = data.shape
if method=='rnd':
C = np.atleast_2d(np.cov(data))
Sigma = np.transpose(np.tile(np.diag(np.diag(C))*0.5, (M, 1, 1)), (2, 1, 0))
# Initialize each mean to a random data point
indices = np.arange(T)
np.random.shuffle(indices)
mu = data[:,indices[0:M]]
weights, _ = normalise(np.ones((M,1)))
elif method=='kmeans':
gmm = GMM(n_components=M, covariance_type=cov_type,
thresh=1e-2, min_covar=1e-3,
n_iter=5, n_init=1, params='wmc', init_params='wmc')
gmm.fit(data.T)
mu = gmm.means_.T
weights = np.asmatrix(gmm.weights_).T
covars = gmm.covars_
Sigma = np.zeros((d,d,M))
for m in range(M):
if cov_type=='diag':
Sigma[:,:,m] = np.diag(covars[m,:])
elif cov_type=='full':
Sigma[:,:,m] = covars[:,:,m]
elif cov_type=='spherical':
Sigma[:,:,m] = covars[m] * np.eye(d)
return mu, Sigma, weights
def build_models(data,class_name,file_name):
print("Building EM")
#Gausian_EM=GMM(n_components=6,n_init=1000)
Gausian_EM=GMM(n_components=6)
Gausian_EM.fit(data)
print("Building SVM")
#SVM_Model=SVC(probability=True,kernel='RBF')
#SVM_Model=SVC(probability=True,kernel='poly')
SVM_Model=SVC(probability=True)
print(SVM_Model.fit(data,class_name))
print("Building Gausian Naive")
Gausian_Naive_Model=GaussianNB()
Gausian_Naive_Model.fit(data,class_name)
print("Building Kmeans")
K_means_model=KMeans(n_clusters=7)
K_means_model.fit(data)
return Gausian_EM,SVM_Model,Gausian_Naive_Model,K_means_model
#min_max_scaler=preprocessing.MinMaxScaler(feature_range=(-1000, 10))
features_scaled = data
'''
print(features_scaled.shape)
print(features_scaled.min(axis=0))
print(features_scaled.max(axis=0))
scatter(features_scaled[:,0],features_scaled[:,1])
'''
#labels=model.fit_predict(features_scaled)
#print(labels)
print("checking for the song", file_name[2])
print("original class", class_name1[2])
model = GMM(n_components=7)
model.fit(data)
print("EM predict class", model.predict(data[2]))
print("EM predict class", model.predict_proba(data[2]))
#print("EM score",model.score(data, class_name1))
model = SVC(probability=True)
#model.fit(data,class_name1)
#print("SVM predict class",model.predict(data[2]))
#print("SVM predict class",model.predict_proba(data[2]))
#print("SVM score",model.score(data, class_name1))
model = GaussianNB()
#model.fit(data,class_name1)
#print("Gausian Naive predict class",model.predict(data[2]))
#print("Gausian Naive predict class",model.predict_proba(data[2]))
#print("GNB score",model.score(data, class_name1))
# print W
# plot projection
pca = PCA(n_components=2)
pca.fit(omegas[:,1:4])
projs = pca.transform(omegas[:,1:4])
plt.plot(projs[:,0], projs[:,1], 'bo')
plt.savefig('projs.png')
plt.close()
# fit mixture of gaussians
N = omegas.shape[0]
for k in range(1,10):
g = GMM(n_components=k, covariance_type='full')
mixture = g.fit(omegas[0:(int(0.75*N)),1:4])
print k, ":", sum(mixture.score_samples(omegas[(int(0.75*N)):N,1:4])[0])
g = GMM(n_components=3, covariance_type='full')
mixture = g.fit(omegas[:,1:4])
print "weights:", mixture.weights_
print "means:", mixture.means_
print "covariances:", mixture.covars_
A = np.transpose(pca.transform(np.eye(3)))
print A
new_means = np.zeros((3, 2))
new_covars = np.zeros((3, 2, 2))
print mixture.means_.shape
def train_user():
print('start capturing video.....')
time.sleep(3)
cam = cv2.VideoCapture(0)
cam.set(3, 640) # set video width
cam.set(4, 480) # set video height
face_detector = cv2.CascadeClassifier(
'haarcascade_frontalface_default.xml')
# For each person, enter one numeric face id'''
face_id = input('enter user id=')
name = input('enter user-name=')
path = '/Users/parthpatel/Desktop/Face-Voice-Recognition/dataset/' + face_id
os.mkdir(path)
print("[INFO] Initializing face capture. Look at the camera and wait ...")
# Initialize individual sampling face count
count = 0
while (True):
ret, img = cam.read()
# img = cv2.flip(img, -1) # flip video image vertically
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = face_detector.detectMultiScale(gray, 1.3, 5)
for (x, y, w, h) in faces:
cv2.rectangle(img, (x, y), (x + w, y + h), (255, 0, 0), 2)
count += 1
# Save the captured image into the datasets folder
cv2.imwrite(
"/Users/parthpatel/Desktop/Face-Voice-Recognition/dataset/" +
face_id + "/User." + str(face_id) + '.' + str(count) + ".jpg",
gray[y:y + h, x:x + w])
cv2.imshow('image', img)
k = cv2.waitKey(100) & 0xff # Press 'ESC' for exiting video
if k == 27:
break
elif count >= 30: # Take 30 face sample and stop video
break
# release camera
print("[INFO] Exiting Program .......")
cam.release()
cv2.destroyAllWindows()
# Path for face image database
path = '/Users/parthpatel/Desktop/Face-Voice-Recognition/dataset/' + face_id
recognizer = cv2.face.LBPHFaceRecognizer_create()
detector = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")
# function to get the images and label data
def getImagesAndLabels(path):
imagePaths = [os.path.join(path, f) for f in os.listdir(path)]
faceSamples = []
ids = []
for imagePath in imagePaths:
if imagePath == path + '.DS_Store':
continue
PIL_img = Image.open(imagePath).convert(
'L') # convert it to grayscale
img_numpy = np.array(PIL_img, 'uint8')
id = int(os.path.split(imagePath)[-1].split(".")[1])
faces = detector.detectMultiScale(img_numpy)
for (x, y, w, h) in faces:
faceSamples.append(img_numpy[y:y + h, x:x + w])
ids.append(id)
return faceSamples, ids
print("\n [INFO] Training faces. It will take a few seconds. Wait ...")
faces, ids = getImagesAndLabels(path)
recognizer.train(faces, np.array(ids))
train_path = '/Users/parthpatel/Desktop/Face-Voice-Recognition/trainer/' + face_id
os.mkdir(train_path)
# Save the model into trainer/trainer.yml
recognizer.write(
'/Users/parthpatel/Desktop/Face-Voice-Recognition/trainer/' + face_id +
'/trainer.yml') # recognizer.save() worked on Mac, but not on Pi
# Print the numer of faces trained and end program
print("\n [INFO] {0} faces trained. Exiting Program".format(
len(np.unique(ids))))
print(
'--------------------------------------------------------------------------------------------'
)
time.sleep(5)
print(
'--------------------------------------------------------------------------------------------'
)
# We'll run the script for different people and store
# the data into multiple files
# Voice authentication
CHUNK = 1024
FORMAT = pyaudio.paInt16
CHANNELS = 2
RATE = 44100
RECORD_SECONDS = 3
WAVE_OUTPUT_FILENAME = "./voices/" + face_id
os.mkdir(WAVE_OUTPUT_FILENAME)
p = pyaudio.PyAudio()
stream = p.open(format=FORMAT,
channels=CHANNELS,
rate=RATE,
input=True,
frames_per_buffer=CHUNK)
print("* recording")
print("...........")
frames = []
for i in range(0, int(RATE / CHUNK * RECORD_SECONDS)):
audio_data = stream.read(CHUNK)
frames.append(audio_data)
print("* done recording")
stream.stop_stream()
stream.close()
p.terminate()
# # saving wav file of speaker
waveFile = wave.open(
WAVE_OUTPUT_FILENAME + '/' + name + '_sample_' + face_id + '.wav',
'wb')
waveFile.setnchannels(CHANNELS)
waveFile.setsampwidth(p.get_sample_size(FORMAT))
waveFile.setframerate(RATE)
waveFile.writeframes(b''.join(frames))
waveFile.close()
print("Done")
# # path to training data
source = WAVE_OUTPUT_FILENAME
# # path to save trained model
dest = "./gmm_model/"
files = [
os.path.join(source, f) for f in os.listdir(source)
if f.endswith('.wav')
]
print(files[0])
features = np.array([])
sr, audio = read(files[0])
# convert audio to mfcc
vector = extract_features(audio, sr)
features = vector
# for more accurate weights
features = np.vstack((features, vector))
features = np.vstack((features, vector))
print(features.shape)
#Gaussian-mixture-model to save gmm-model
gmm = GMM(n_components=16, n_iter=200, covariance_type='diag', n_init=3)
gmm.fit(features)
# # picklefile = f.split("\\")[-2].split(".wav")[0]+".gmm"
# # model saving..
pickle.dump(gmm, open(dest + name + '_' + face_id + '.gmm', 'wb'))
print(name + ' ' + 'added......')
features = np.asarray(())
for token in test_tokens: if token in w2v_model: test_vectors.append(w2v_model[token]) elif token in unknown_words: test_vectors.append(unknown_words[token]) else: unknown_vec = w2v_model.seeded_vector(token) unknown_words[token] = unknown_vec test_vectors.append(unknown_vec) # Train GMM print 'Starting GMM training' words = w2v_model.vocab.keys() word_vectors = w2v_model.syn0 gmm_model = GMM(n_components=num_topics, n_iter=num_gmm_iterations, covariance_type='diag') gmm_model.fit(word_vectors) # joblib.dump(gmm_model, gmm_output_file) print 'Done GMM training' # Get the likelihood of each word vector under each Gaussian component scores = gmm_model.score(test_vectors) print scores ll = sum(scores) print "LL: "+str(ll) # Print topics if desired if print_topics: log_probs = log_multivariate_normal_density(word_vectors, gmm_model.means_, gmm_model.covars_, gmm_model.covariance_type) print np.min(log_probs) _, num_col = log_probs.shape for col in xrange(num_col):
test_vectors.append(w2v_model[token])
elif token in unknown_words:
test_vectors.append(unknown_words[token])
else:
unknown_vec = w2v_model.seeded_vector(token)
unknown_words[token] = unknown_vec
test_vectors.append(unknown_vec)
# Train GMM
print 'Starting GMM training'
words = w2v_model.vocab.keys()
word_vectors = w2v_model.syn0
gmm_model = GMM(n_components=num_topics,
n_iter=num_gmm_iterations,
covariance_type='diag')
gmm_model.fit(word_vectors)
# joblib.dump(gmm_model, gmm_output_file)
print 'Done GMM training'
# Get the likelihood of each word vector under each Gaussian component
scores = gmm_model.score(test_vectors)
print scores
ll = sum(scores)
print "LL: " + str(ll)
# Print topics if desired
if print_topics:
log_probs = log_multivariate_normal_density(
word_vectors, gmm_model.means_, gmm_model.covars_,
gmm_model.covariance_type)
print np.min(log_probs)
print(features_scaled.shape)
print(features_scaled.min(axis=0))
print(features_scaled.max(axis=0))
scatter(features_scaled[:,0],features_scaled[:,1])
'''
#labels=model.fit_predict(features_scaled)
#print(labels)
print("checking for the song",file_name[2])
print("original class",class_name1[2])
model=GMM(n_components=7)
model.fit(data)
print("EM predict class",model.predict(data[2]))
print("EM predict class",model.predict_proba(data[2]))
#print("EM score",model.score(data, class_name1))
model=SVC(probability=True)
#model.fit(data,class_name1)
#print("SVM predict class",model.predict(data[2]))
#print("SVM predict class",model.predict_proba(data[2]))
#print("SVM score",model.score(data, class_name1))
model=GaussianNB()
#model.fit(data,class_name1)
#print("Gausian Naive predict class",model.predict(data[2]))
#print("Gausian Naive predict class",model.predict_proba(data[2]))
#print("GNB score",model.score(data, class_name1))
