Using Neural Architecture Search for Speech Recognition on the Edge