def load_vox_forge_files(vox_dir, max_files, vox_file = 'voxforge.pcl', feature_func = extract_mfcc_features):
feature_storage = FeatureStorage( name = vox_file, base_dir = FEATURE_STORAGE )
if feature_storage.exists():
print "Loading voxforge from pickle"
return feature_storage.load()
else:
pool = Pool()
files = get_voxforge_total_files(vox_dir, max_files)
print "Started processing voxforge data {}".format( len(files) )
voxforge_features = pool.map( feature_func, files )
pool.close()
pool.terminate()
print "Dumping voxforge features"
feature_storage.save( voxforge_features )
return voxforge_features
def load_train_features( train_filename, feature_func, limit = None ):
feature_storage = FeatureStorage( name = train_filename, base_dir = FEATURE_STORAGE )
if not feature_storage.exists( ):
all_train_data = return_all_train_files()
if limit is not None:
all_train_data = all_train_data[:limit]
print "Started processing train"
pool = Pool()
X_train_transformed = pool.map( feature_func, all_train_data )
pool.close()
pool.terminate()
print "Dumping train features"
feature_storage.save( X_train_transformed )
else:
print "Loading train from cache"
X_train_transformed = feature_storage.load()
return X_train_transformed
def load_test_features(test_filename, feature_func, limit = None):
pool = Pool()
test_data = get_all_test_data()
if limit:
test_data = test_data[:limit]
feature_storage = FeatureStorage( name = test_filename, base_dir = FEATURE_STORAGE )
if not feature_storage.exists():
print "Loading test from scratch"
for_features = [ (path, None) for (path, filename ) in test_data ]
X_test_transformed = pool.map( feature_func, for_features )
print "Dumping test features"
feature_storage.save( X_test_transformed )
print "Finished dumping"
else:
print "Loading test from cache"
X_test_transformed = feature_storage.load()
pool.close()
pool.terminate()
return X_test_transformed