def __init__(self, WindowSize=5, FeaturesDimension=41):
# features normalization values
self.mean_vect = np.load(
os.path.join(os.path.dirname(os.path.abspath(__file__)),
'data/TrainingSetMean.npy'))
self.stdev_vect = np.load(
os.path.join(os.path.dirname(os.path.abspath(__file__)),
'data/TrainingSetStDev.npy'))
# TF graph initialization
self.config = Configuration(WindowSize, FeaturesDimension)
self.graph = tf.Graph()
with self.graph.as_default():
self.feat = tf.placeholder(
dtype=tf.float32, shape=[1, self.config.audio_feat_dimension])
with tf.variable_scope('model'):
model = VAD_DNN.Model(self.feat, self.config)
logits_prob = model.softmax
# the probability of speech is given by the first dimension in the softmax
# so we slice the output accordingly
self.speech_prob = tf.slice(logits_prob, [0, 0], [-1, 1])
init_op = tf.local_variables_initializer()
saver = tf.train.Saver()
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
self.session = tf.Session(graph=self.graph, config=config)
self.session.run(init_op)
saver.restore(
self.session,
os.path.join(os.path.dirname(os.path.abspath(__file__)),
"data/datamean_nodeltas_model_epoch13.ckpt"))
Vb.main(save_dir, prj_dir, 'bDNN', 'train', dev='/gpu:' + gpu_no)
gs.freeze_graph(prj_dir + '/logs/bDNN',
prj_dir + '/saved_model/graph/bDNN',
'model_1/logits,model_1/labels')
if mode == 2:
set_path = ps.PathSetting(prj_dir, 'DNN', save_dir)
logs_dir = set_path.logs_dir
os.system("rm -rf " + logs_dir + '/train')
os.system("rm -rf " + logs_dir + '/valid')
os.system("mkdir " + logs_dir + '/train')
os.system("mkdir " + logs_dir + '/valid')
Vd.main(save_dir, prj_dir, 'DNN', 'train', dev='/gpu:' + gpu_no)
gs.freeze_graph(prj_dir + '/logs/DNN',
prj_dir + '/saved_model/graph/DNN',
'model_1/soft_pred,model_1/raw_labels')
if mode == 3:
set_path = ps.PathSetting(prj_dir, 'LSTM', save_dir)
logs_dir = set_path.logs_dir
os.system("rm -rf " + logs_dir + '/train')
os.system("rm -rf " + logs_dir + '/valid')
os.system("mkdir " + logs_dir + '/train')
os.system("mkdir " + logs_dir + '/valid')
Vl.main(save_dir, prj_dir, 'LSTM', 'train', dev='/gpu:' + gpu_no)
pred, label = Vp.main()
elif mode == 1:
Vb.test_config(c_test_dir=data_dir,
c_norm_dir=norm_dir,
c_initial_logs_dir=model_dir,
c_batch_size_eval=batch_size,
c_data_len=data_len)
pred, label = Vb.main()
elif mode == 2:
Vd.test_config(c_test_dir=data_dir,
c_norm_dir=norm_dir,
c_initial_logs_dir=model_dir,
c_batch_size_eval=batch_size,
c_data_len=data_len)
pred, label = Vd.main()
elif mode == 3:
Vl.test_config(c_test_dir=data_dir,
c_norm_dir=norm_dir,
c_initial_logs_dir=model_dir,
c_batch_size_eval=batch_size,
c_data_len=data_len)
pred, label = Vl.main()
Vb.main(prj_dir, 'bDNN', 'train')
gs.freeze_graph(prj_dir + '/logs/bDNN', prj_dir + '/saved_model/graph/bDNN', 'model_1/logits,model_1/labels')
# gs.freeze_graph(prj_dir + '/saved_model/temp', prj_dir + '/saved_model/temp', 'model_1/soft_pred,model_1/raw_labels')
if mode == 2:
set_path = ps.PathSetting(prj_dir, 'DNN')
logs_dir = set_path.logs_dir
os.system("rm -rf " + logs_dir + '/train')
os.system("rm -rf " + logs_dir + '/valid')
os.system("mkdir " + logs_dir + '/train')
os.system("mkdir " + logs_dir + '/valid')
Vd.main(prj_dir, 'DNN', 'train')
gs.freeze_graph(prj_dir + '/logs/DNN', prj_dir + '/saved_model/graph/DNN', 'model_1/soft_pred,model_1/raw_labels')
# gs.freeze_graph(prj_dir + '/saved_model/temp', prj_dir + '/saved_model/temp', 'model_1/soft_pred,model_1/raw_labels')
if mode == 3:
set_path = ps.PathSetting(prj_dir, 'LSTM')
logs_dir = set_path.logs_dir
os.system("rm -rf " + logs_dir + '/train')
os.system("rm -rf " + logs_dir + '/valid')
os.system("mkdir " + logs_dir + '/train')
os.system("mkdir " + logs_dir + '/valid')
Vl.main(prj_dir, 'LSTM', 'train')
def main(
param=0.2,
PATH_LOAD_FILE='/home/keums/Melody/dataset/adc2004_full_set/file/pop4.wav',
PATH_SAVE_FILE='./SAVE_RESULTS/pop4.txt'):
# PATH_LOAD_FILE = sys.argv[1]
# PATH_SAVE_FILE = sys.argv[2]
#==================================
# Feature Extraction
# .wav --> spectrogram
#==================================
x_test_log = myFeatureExtraction(PATH_LOAD_FILE) #path ??
#==================================
# making multi column spectrogram
# for trainging
#==================================
x_test_SF = making_multi_frame(x_test_log, num_frames=1)
x_test_MF = making_multi_frame(x_test_log, num_frames=11)
select_res_1st = 1
select_res_2nd = 2
select_res_3rd = 4
pitch_range = np.arange(min_pitch, max_pitch + 1.0 / select_res_3rd,
1.0 / select_res_3rd)
#==================================
# Melody extraction
# using DNN
#==================================
y_predict_1st = MelodyExtraction_SCDNN(x_test_MF, select_res_1st)
y_predict_2nd = MelodyExtraction_SCDNN(x_test_MF, select_res_2nd)
y_predict_3rd = MelodyExtraction_SCDNN(x_test_MF, select_res_3rd)
#==================================
# merge SCDNN
#==================================
# print 'Merging....'
ratio_res_1_3 = select_res_3rd / select_res_1st
ratio_res_2_3 = select_res_3rd / select_res_2nd
y_predict_tmp_1_3 = np.zeros(y_predict_3rd.shape)
y_predict_tmp_2_3 = np.zeros(y_predict_3rd.shape)
for i in range(y_predict_3rd.shape[0]):
for j in range(y_predict_1st.shape[1] - 1):
y_predict_tmp_1_3[i, j * ratio_res_1_3:j * ratio_res_1_3 +
ratio_res_1_3] = y_predict_1st[i, j]
y_predict_tmp_1_3[i, -1] = y_predict_1st[i, -1]
for i in range(y_predict_3rd.shape[0]):
for j in range(y_predict_2nd.shape[1] - 1):
y_predict_tmp_2_3[i, j * ratio_res_2_3:j * ratio_res_2_3 +
ratio_res_2_3] = y_predict_2nd[i, j]
y_predict_tmp_2_3[i, -1] = y_predict_2nd[i, -1]
# y_predict = (y_predict_tmp_1_3+0.0000001) *(y_predict_tmp_2_3+0.0000001) * (y_predict_3rd +0.0000001)
y_predict = 10**(np.log10(y_predict_tmp_1_3) +
np.log10(y_predict_tmp_2_3) + np.log10(y_predict_3rd))
del y_predict_tmp_1_3
del y_predict_tmp_2_3
#==================================
# singing voice detection
#==================================
voice_frame_vad = VAD_DNN(x_test_SF, y_predict_1st, param=0.2)
#==================================
# viterbi algorithm
#==================================
path_viterbi = './viterbi/'
path_prior_matrix_file = path_viterbi + 'prior_' + str(
select_res_3rd) + '.npy'
path_transition_matrix_file = path_viterbi + 'transition_matrix_' + str(
select_res_3rd) + '.npy'
prior = np.load(path_prior_matrix_file)
transition_matrix = np.load(path_transition_matrix_file)
viterbi_path = viterbi(y_predict,
transition_matrix=transition_matrix,
prior=prior,
penalty=0,
scaled=True)
pitch_MIDI = np.zeros([y_predict.shape[0], 1])
pitch_freq = np.zeros([y_predict.shape[0], 1])
for i in range(y_predict.shape[0]):
# for test : origianl
# index_predict[i] = np.argmax(y_predict[i,:])
# pitch_MIDI[i] = pitch_range[index_predict[i]]
#viterbi_path
pitch_MIDI[i] = pitch_range[viterbi_path[i]]
pitch_freq[i] = 2**((pitch_MIDI[i] - 69) / 12.) * 440
est_pitch = np.multiply(pitch_freq, voice_frame_vad)
#==================================
#adjust frame
#==================================
idx_shift = 2
shift_array = np.zeros(idx_shift)
est_pitch = np.append(shift_array, est_pitch[:-idx_shift])
#==================================
# save result
#==================================
PATH_est_pitch = PATH_SAVE_FILE
if not os.path.exists(os.path.dirname(PATH_est_pitch)):
os.makedirs(os.path.dirname(PATH_est_pitch))
f = open(PATH_est_pitch, 'w')
for j in range(len(est_pitch)):
est = "%f\t%f\n" % (0.01 * j, est_pitch[j])
f.write(est)
f.close()
print PATH_est_pitch
os.system("rm -rf " + save_dir)
os.system("mkdir " + save_dir)
os.system("mkdir " + save_dir + '/train')
os.system("mkdir " + save_dir + '/valid')
os.system(
"matlab -r \"try acoustic_feat_ex(\'%s\',\'%s\'); catch; end; quit\""
% (train_data_dir, train_save_dir))
os.system(
"matlab -r \"try acoustic_feat_ex(\'%s\',\'%s\'); catch; end; quit\""
% (valid_data_dir, valid_save_dir))
train_norm_dir = save_dir + '/train/global_normalize_factor.mat'
test_norm_dir = prj_dir + '/norm_data/global_normalize_factor.mat'
os.system("cp %s %s" % (train_norm_dir, test_norm_dir))
if mode == 0:
logs_dir = prj_dir + '/logs'
os.system("rm -rf " + logs_dir + '/train')
os.system("rm -rf " + logs_dir + '/valid')
os.system("mkdir " + logs_dir + '/train')
os.system("mkdir " + logs_dir + '/valid')
Vd.train_config(save_dir + '/train', save_dir + '/valid',
prj_dir + '/logs', batch_size, train_step, 'train')
Vd.main()
# os.system("rm -rf")
print("done")