Sequence Transduction with Recurrent Neural Networks, Alex Graves, 2012Proceedings of the International Conference of Machine Learning (ICML) 2012 Workshop on Representation Learning, Vol. 27 - 提出了循环神经网络换能器(RNN-T)架构,提供了一种无需显式对齐即可将输入序列直接映射到输出序列的框架,对流式ASR至关重要。