Create a scientific presentation of a paper accepted at Interspeech 2026 in a clear manner, with many visuals.

Created using ChatSlide

This document explores the significance of cross-attention in linking encoded speech to output tokens, emphasizing its role in alignment and speaker cues. It discusses the transition from attention mechanisms to explanations in speech-to-text (S2T) systems, highlighting the need for connections between words and acoustic evidence. The research questions focus on the correlation between cross-attention and input saliency, the effects of encoder context mixing, and the consistency of results...