Learning Efficient Representations for Keyword Spotting with Triplet Loss - View it on GitHub