def split_data_by_s5(src_dir, des_dir, keywords=['train_si284', 'test_eval92', 'test_dev93']): count = 0 for key in keywords: wav_file_list = os.path.join(src_dir, key+'.flist') label_file_list = os.path.join(src_dir, key+'.txt') new_path = check_path_exists(os.path.join(des_dir, key)) with open(wav_file_list, 'r') as wfl: wfl_contents = wfl.readlines() for line in wfl_contents: line = line.strip() if os.path.isfile(line): shutil.copyfile(line, os.path.join(des_dir, key, line.split('/')[-1])) print(line) else: tmp = '/'.join(line.split('/')[:-1]+[line.split('/')[-1].upper()]) shutil.copyfile(tmp, os.path.join(des_dir, key, line.split('/')[-1].replace('WV1', 'wv1'))) print(tmp) with open(label_file_list, 'r') as lfl: lfl_contents = lfl.readlines() for line in lfl_contents: label = ' '.join(line.strip().split(' ')[1:]) with open(os.path.join(des_dir, key, line.strip().split(' ')[0]+'.label'), 'w') as lf: lf.writelines(label) print(key, label)
optimizer_fn = optimizer_functions_dict[FLAGS.optimizer] batch_size = FLAGS.batch_size num_hidden = FLAGS.num_hidden num_feature = FLAGS.num_feature num_classes = get_num_classes(level) num_epochs = FLAGS.num_epochs lr = FLAGS.lr grad_clip = FLAGS.grad_clip datadir = FLAGS.datadir logdir = FLAGS.logdir savedir = os.path.join(logdir, level, 'save') resultdir = os.path.join(logdir, level, 'result') loggingdir = os.path.join(logdir, level, 'logging') check_path_exists([logdir, savedir, resultdir, loggingdir]) mode = FLAGS.mode keep = FLAGS.keep keep_prob = 1 - FLAGS.dropout_prob print('%s mode...' % str(mode)) if mode == 'test': batch_size = 100 num_epochs = 1 train_mfcc_dir = os.path.join(datadir, level, 'train', 'mfcc') train_label_dir = os.path.join(datadir, level, 'train', 'label') test_mfcc_dir = os.path.join(datadir, level, 'test', 'mfcc') test_label_dir = os.path.join(datadir, level, 'test', 'label') logfile = os.path.join(