Enhancing Word Discrimination and Matching in Query-by-Example Spoken term detection with Acoustic Word Embeddings.