Text this: Energy-Efficient Speaker Identification with Low-Precision Networks