Text this: Making Time Series Embeddings More Interpretable in Deep Learning