Text this: Compact Acoustic Models for Embedded Speech Recognition