With the rise of low power speech-enabled devices, there is a growing demand
to quickly produce models for recognizing arbitrary sets of keywords. As with
many machine learning tasks, one of the most challenging parts in the model
creation process is obtaining a sufficient amount of tr