75d3f75b46ab272c07e449c871a1baa104440194,nnmnkwii/frontend/merlin.py,,load_labels_with_phone_alignment,#,162

Before Change

                                                                     frame_number)

        if add_frame_features:
            current_block_binary_array = np.zeros(
                (frame_number, dict_size + frame_feature_size))
            for i in range(frame_number):
                current_block_binary_array[i,
                                           0:dict_size] = label_vector

                if subphone_features == "minimal_phoneme":
                    // features which distinguish frame position in phoneme
                    // fraction through phone forwards
                    current_block_binary_array[i, dict_size] = float(
                        i + 1) / float(frame_number)
                    // fraction through phone backwards
                    current_block_binary_array[i, dict_size + 1] = float(
                        frame_number - i) / float(frame_number)
                    // phone duration
                    current_block_binary_array[i,
                                               dict_size + 2] = float(frame_number)

                elif subphone_features == "coarse_coding":
                    // features which distinguish frame position in phoneme
                    // using three continous numerical features
                    current_block_binary_array[i,
                                               dict_size + 0] = cc_feat_matrix[i, 0]
                    current_block_binary_array[i,
                                               dict_size + 1] = cc_feat_matrix[i, 1]
                    current_block_binary_array[i,
                                               dict_size + 2] = cc_feat_matrix[i, 2]
                    current_block_binary_array[i,
                                               dict_size + 3] = float(frame_number)

                elif subphone_features is None:
                    pass

After Change

                                                                     frame_number)

        if add_frame_features:
            current_block_binary_array = np.zeros(
                (frame_number, dict_size + frame_feature_size))
            for i in range(frame_number):
                current_block_binary_array[i,
                                           0:dict_size] = label_vector

                if subphone_features == "minimal_phoneme":
                    // features which distinguish frame position in phoneme
                    // fraction through phone forwards
                    current_block_binary_array[i, dict_size] = float(
                        i + 1) / float(frame_number)
                    // fraction through phone backwards
                    current_block_binary_array[i, dict_size + 1] = float(
                        frame_number - i) / float(frame_number)
                    // phone duration
                    current_block_binary_array[i,
                                               dict_size + 2] = float(frame_number)

                elif subphone_features == "coarse_coding":
                    // features which distinguish frame position in phoneme
                    // using three continous numerical features
                    current_block_binary_array[i,
                                               dict_size + 0] = cc_feat_matrix[i, 0]
                    current_block_binary_array[i,
                                               dict_size + 1] = cc_feat_matrix[i, 1]
                    current_block_binary_array[i,
                                               dict_size + 2] = cc_feat_matrix[i, 2]
                    current_block_binary_array[i,
                                               dict_size + 3] = float(frame_number)

                elif subphone_features is None:
                    pass

Instances