I\'m studying the end-to-end architecture for automatic speech recognition systems. RNN transducer (RNN-T) is one of the popular end-to-end methods,but it is so difficult to