arXiv:2611.06043v1 Announce Type: cross Abstract: Streaming speech recognition is increasingly consumed by machines, such as translators and voice agents, that act on each token as it appears. A token that is later revised costs wasted work or a wrong action.
