1 paper · 1 filter
Albert Zeyer, Robin Schmitt, Wei Zhou +2
We introduce a novel segmental-attention model for automatic speech recognition. We restrict the decoder attention to segments to avoid quadratic runtime of global attention, bette…