def train():
# prepare data
tr_X, tr_y, _, te_X, te_y, te_na_list = pp_dev_data.GetAllData(
fe_fd, agg_num, hop, fold)
[batch_num, n_time, n_freq] = tr_X.shape
print tr_X.shape, tr_y.shape
print te_X.shape, te_y.shape
# build model
seq = Sequential()
seq.add(InputLayer((n_time, n_freq)))
seq.add(Flatten()) # flatten to 2d: (n_time, n_freq) to 1d:(n_time*n_freq)
seq.add(Dropout(0.1))
seq.add(Dense(n_hid, act=act))
seq.add(Dropout(0.1))
seq.add(Dense(n_hid, act=act))
seq.add(Dropout(0.1))
seq.add(Dense(n_hid, act=act))
seq.add(Dropout(0.1))
seq.add(Dense(n_out, act='sigmoid'))
md = seq.compile()
md.summary()
# optimizer
optimizer = Adam(1e-4)
# callbacks
# tr_err, te_err are frame based. To get event based err, run recognize.py
validation = Validation(tr_x=tr_X,
tr_y=tr_y,
va_x=None,
va_y=None,
te_x=te_X,
te_y=te_y,
batch_size=2000,
metrics=['binary_crossentropy'],
call_freq=1,
dump_path=None)
# save model
pp_dev_data.CreateFolder(cfg.dev_md_fd)
save_model = SaveModel(dump_fd=cfg.dev_md_fd, call_freq=10)
# callbacks
callbacks = [validation, save_model]
# fit model
md.fit(x=tr_X,
y=tr_y,
batch_size=2000,
n_epochs=100,
loss_func='binary_crossentropy',
optimizer=optimizer,
callbacks=callbacks,
verbose=1)
def detect():
# init paths
if type == 'home':
fe_fd = cfg.eva_fe_mel_home_fd
labels = cfg.labels_home
lb_to_id = cfg.lb_to_id_home
id_to_lb = cfg.id_to_lb_home
if type == 'resi':
fe_fd = cfg.eva_fe_mel_resi_fd
labels = cfg.labels_resi
lb_to_id = cfg.lb_to_id_resi
id_to_lb = cfg.id_to_lb_resi
n_out = len(labels)
# load model
md = serializations.load(md_path)
# do recognize for each test audio
names = os.listdir(fe_fd)
names = sorted(names)
pp_dev_data.CreateFolder(cfg.eva_results_fd)
pp_dev_data.CreateFolder(cfg.eva_results_fd + '/' + type)
# detect and write out for all audios
for na in names:
X = cPickle.load(open(fe_fd + '/' + na, 'rb'))
X = mat_2d_to_3d(X, agg_num, hop)
y_pred = md.predict(X)
outlist = pp_dev_data.OutMatToList(y_pred, thres, id_to_lb)
full_na = type + '/audio/' + na[0:4] + '.wav'
out_txt_path = cfg.eva_results_fd + '/' + type + '/' + na[
0:4] + '_detect.ann'
f = open(out_txt_path, 'w')
for li in outlist:
f.write(full_na + '\t' + str(li['event_onset']) + '\t' +
str(li['event_offset']) + '\t' + li['event_label'] + '\n')
print 'Write out detection result to', out_txt_path, 'successfully!'
f.close()
def detect_cv():
# init paths
if type=='home':
fe_fd = cfg.dev_fe_mel_home_fd
labels = cfg.labels_home
lb_to_id = cfg.lb_to_id_home
id_to_lb = cfg.id_to_lb_home
tr_txt = cfg.dev_evaluation_fd + '/home_fold' + str(fold) + '_train.txt'
te_txt = cfg.dev_evaluation_fd + '/home_fold' + str(fold) + '_evaluate.txt'
meta_fd = cfg.dev_meta_home_fd
if type=='resi':
fe_fd = cfg.dev_fe_mel_resi_fd
labels = cfg.labels_resi
lb_to_id = cfg.lb_to_id_resi
id_to_lb = cfg.id_to_lb_resi
tr_txt = cfg.dev_evaluation_fd + '/residential_area_fold' + str(fold) + '_train.txt'
te_txt = cfg.dev_evaluation_fd + '/residential_area_fold' + str(fold) + '_evaluate.txt'
meta_fd = cfg.dev_meta_resi_fd
n_out = len( labels )
# load model
md = serializations.load( md_path )
# get wav names to be detected
te_names = pp_dev_data.GetWavNamesFromTxt( te_txt )
# do recognize for each test audio
names = os.listdir( fe_fd )
names = sorted( names )
y_pred_list = []
# detect and write out to txt
pp_dev_data.CreateFolder( cfg.dev_results_fd )
file_list = []
for na in names:
if na[0:4] in te_names:
print na
gt_file = meta_fd + '/' + na[0:4] + '.ann'
out_file = cfg.dev_results_fd + '/'+na[0:4]+'_detect.ann'
X = cPickle.load( open( fe_fd+'/'+na, 'rb' ) )
X = mat_2d_to_3d( X, agg_num, hop )
y_pred = md.predict( X )
y_pred_list.append( y_pred )
out_list = pp_dev_data.OutMatToList( y_pred, thres, id_to_lb )
pp_dev_data.PrintListToTxt( out_list, out_file )
file_list.append( { 'reference_file': gt_file, 'estimated_file': out_file } )
# print results for this fold
pp_dev_data.PrintScore( file_list, labels )
# aggregate data
X3d = mat_2d_to_3d( X, agg_num, hop )
Xlist.append( X3d )
return np.concatenate( Xlist, axis=0 )
# size: n_songs*n_chunks*agg_num*n_in
def GetEvaSegData( fe_fd, agg_num, hop ):
te_Xlist = []
names = os.listdir( fe_fd )
te_na_list = []
# read one line
for na in names:
fe_path = fe_fd + '/' + na
X = cPickle.load( open( fe_path, 'rb' ) )
# aggregate data
X3d = mat_2d_to_3d( X, agg_num, hop )
te_Xlist.append( X3d )
te_na_list.append( na[0:-2] )
return np.array( te_Xlist ), te_na_list
if __name__ == "__main__":
pp_dev_data.CreateFolder( cfg.eva_fe_fd )
pp_dev_data.CreateFolder( cfg.eva_fe_mel_fd )
pp_dev_data.GetMel( cfg.eva_wav_fd, cfg.eva_fe_mel_fd, n_delete=0 )
hop = 15
fold = 1
n_labels = len(cfg.labels)
# load model
md = serializations.load(cfg.eva_md_fd + '/md10.p')
# prepare data
te_X = pp_eva_data.GetAllData(cfg.eva_fe_mel_fd, cfg.eva_csv_path, agg_num,
hop)
# do recognize and evaluation
thres = 0.4 # thres, tune to prec=recall
n_labels = len(cfg.labels)
pp_dev_data.CreateFolder(cfg.eva_results_fd)
txt_out_path = cfg.eva_results_fd + '/task4_results.txt'
fwrite = open(txt_out_path, 'w')
with open(cfg.eva_csv_path, 'rb') as f:
reader = csv.reader(f)
lis = list(reader)
# read one line
for li in lis:
na = li[1]
full_na = na + '.16kHz.wav'
# get features, tags
fe_path = cfg.eva_fe_mel_fd + '/' + na + '.f'
X = cPickle.load(open(fe_path, 'rb'))
md.summary()
# validation
# tr_err, te_err are frame based. To get event based err, run recognize.py
validation = Validation(tr_x=tr_X,
tr_y=tr_y,
va_x=None,
va_y=None,
te_x=None,
te_y=None,
metrics=['binary_crossentropy'],
call_freq=1,
dump_path=None)
# save model
pp_dev_data.CreateFolder(cfg.eva_md_fd)
save_model = SaveModel(dump_fd=cfg.eva_md_fd, call_freq=10)
# callbacks
callbacks = [validation, save_model]
# optimizer
optimizer = Rmsprop(1e-4)
# fit model
md.fit(x=tr_X,
y=tr_y,
batch_size=100,
n_epochs=1000,
loss_func='binary_crossentropy',
optimizer=optimizer,
def create_folders():
pp_dev_data.CreateFolder(cfg.scrap_fd + '/Md_eva_bob')
pp_dev_data.CreateFolder(cfg.scrap_fd + '/Results_eva')
pp_dev_data.CreateFolder(cfg.scrap_fd + '/Results_eva/bob_eer')
def train_cv_model():
# init path
if type == 'home':
fe_fd = cfg.dev_fe_mel_home_fd
labels = cfg.labels_home
lb_to_id = cfg.lb_to_id_home
tr_txt = cfg.dev_evaluation_fd + '/home_fold' + str(
fold) + '_train.txt'
te_txt = cfg.dev_evaluation_fd + '/home_fold' + str(
fold) + '_evaluate.txt'
if type == 'resi':
fe_fd = cfg.dev_fe_mel_resi_fd
labels = cfg.labels_resi
lb_to_id = cfg.lb_to_id_resi
tr_txt = cfg.dev_evaluation_fd + '/residential_area_fold' + str(
fold) + '_train.txt'
te_txt = cfg.dev_evaluation_fd + '/residential_area_fold' + str(
fold) + '_evaluate.txt'
n_out = len(labels)
# load data to list
tr_X, tr_y = pp_dev_data.LoadAllData(fe_fd, tr_txt, lb_to_id, agg_num, hop)
tr_y = sparse_to_categorical(tr_y, n_out)
print tr_X.shape
print tr_y.shape
n_freq = tr_X.shape[2]
# build model
seq = Sequential()
seq.add(InputLayer((agg_num, n_freq)))
seq.add(Flatten())
seq.add(Dense(n_hid, act='relu'))
seq.add(Dropout(0.1))
seq.add(Dense(n_hid, act='relu'))
seq.add(Dropout(0.1))
seq.add(Dense(n_hid, act='relu'))
seq.add(Dropout(0.1))
seq.add(Dense(n_out, 'sigmoid'))
md = seq.combine()
# print summary info of model
md.summary()
# optimization method
optimizer = Adam(1e-3)
# callbacks (optional)
# save model every n epoch
pp_dev_data.CreateFolder(cfg.dev_md_fd)
save_model = SaveModel(dump_fd=cfg.dev_md_fd, call_freq=5)
# validate model every n epoch
validation = Validation(tr_x=tr_X,
tr_y=tr_y,
va_x=None,
va_y=None,
te_x=None,
te_y=None,
metrics=['binary_crossentropy'],
call_freq=1,
dump_path=None)
# callbacks function
callbacks = [validation, save_model]
# train model
md.fit(x=tr_X,
y=tr_y,
batch_size=20,
n_epochs=100,
loss_func='binary_crossentropy',
optimizer=optimizer,
callbacks=callbacks)
def create_folders():
pp_dev_data.CreateFolder(cfg.scrap_fd + '/Md_dev_bob')
pp_dev_data.CreateFolder(cfg.scrap_fd + '/Results_dev')
pp_dev_data.CreateFolder(cfg.scrap_fd + '/Results_dev/bob_eer')
pp_dev_data.CreateFolder(cfg.scrap_fd + '/Results_dev/bob_eer/fold' +
str(fold))
