Transducer Consistency Regularization for Speech to Text Applications