def get_voxforge_total_files( vox_forge_dir, max = 3000):
CLASSES_DICT, _ = load_classes_info()
dirs = {'Italian' : os.path.join(vox_forge_dir, 'it' ),
'French' : os.path.join(vox_forge_dir, 'fr' ),
'German' : os.path.join(vox_forge_dir, 'de' ),
'English' : os.path.join(vox_forge_dir, 'en' ),
'Spanish' : os.path.join(vox_forge_dir, 'es' )
}
available_data = []
for lang, path in dirs.iteritems():
files = extract_vox_files_by_dir( path )
random.shuffle(files)
files = files[:max]
print "Available {} files for language {}".format( lang, len(files) )
lang_class = CLASSES_DICT[ lang ]
available_data.extend( [ (f, lang_class) for f in files ] )
return available_data
if __name__ == "__main__":
LIMIT = 9000
DICTIONARY_SIZE = 600
sofia_path = None # sys.argv[1]
if len(sys.argv) > 1:
VOX_FEATURES = sys.argv[1] == "use_voxforge"
else:
VOX_FEATURES = False
print "Using Vox features", VOX_FEATURES
print "LIMIT", LIMIT
print "Checking class names"
_, REVERSE_CLASSES = load_classes_info()
print "Loading train"
X_train_transformed = load_train_features("bow_train_features.pcl", extract_mfcc_features, limit=LIMIT)
X_train = np.vstack([f for (_, _, f) in X_train_transformed if f is not None])
# X_train = np.reshape(X_train, ( len(X_train_transformed), num_features ) )
print X_train.shape
# for filename, lbl, features in X_train_transformed:
# if features is None:
# print "Train",filename, "is none"
# continue
# if X_total is not None:
# X_total = np.vstack([ X_total, features ])
# else:
# X_total = features
X_total = X_train