seq2seq is short for "sequence-to-sequence", referring to a model. E.g. voice-to-text and text translation are seq2seq models. Similar terms are seq2vec and vec2seq, where "vec" stands for "vector". E.g. a sentence embedding model is a seq2vec model.