While larger acoustic models provide better speech recognition performance, smaller models are appropriate when computational resources are limited. Knowledge distillation is used to train small ...