This is a brief document in response to the request by our reviewers for a description of our goals and plans for the final period of our SPRACH project. This was deemed particularly important due to the apparent profusion of unrelated (though worthy) investigations by the research partners, as reported in our March 1998 review. We have discussed this concern, and while we largely agree with it, we note that our plan had always been to consider a wide range of approaches within our project, followed by a period of reconsolidation of the most promising ideas (or at least practical ones) into some final systems. However, this latter phase has required better definition, and now that we have a range of intermediate results it is possible for us to do this.
Following the review meeting, we discussed specific steps to address these concerns. In particular, in the remaining months we will be focusing most of our efforts on two concrete goals, both of direct relevance to the THISL project:
In particular the BN evaluation system will involve the integration of several of the new approaches developed by the partners into a single system (hopefully with an improvement in performance!)
CUED, Sheffield and ICSI will be the sites mainly collaborating on the Broadcast News system. FPMs and INESC will mainly work on the French and Portuguese systems, although FPMs will be involved in applying some of their work in WP4 and WP5 to the BN system.
Both of these goals will require a close collaboration between SPRACH partners, and both will lead to results that can be easily summarised and evaluated, in keeping with the recommendations of the reviewers.