Crossmodal ASR Error Correction with Discrete Speech Units