Text this: A Deep-Learning Approach to Heart Sound Classification Based on Combined Time-Frequency Representations