Two papers, one story. "Attention Is All You Need" gave the world the Transformer and reset how sequence modeling gets done, while "Listen, Attend and Spell" (LAS) did the same thing a year earlier for end-to-end speech recognition, replacing CTC-and-HMM pipelines with a single attention-based encoder-decoder. Reading them together, from a speech engineer's vantage point, makes it clear how much modern ASR owes to attention mechanics that were first road-tested on characters and phonemes.

This review lines up the two architectures, contrasts self-attention against LAS's pyramidal BLSTM encoder and content-based attention, and traces how ideas migrated between the NLP and speech communities on the way to today's Conformers, transducers, and Whisper-style seq2seq ASR. If you want the historical throughline that connects vanilla Transformers to the speech foundation models on your GPU right now, this deep-dive is a solid grounding session.