Aiming at both speaker independence and robustness with respect to recognition errors in the spoken queries, we have implemented a two-pass system for spoken web search. In the first pass, unconstrained phone recognition ...