SBIR-STTR Award

Today there is an incredible and rapidly growing amount of Arabic broadcast data available over the air and digitally. Furthermore this media is of high interest to governments and corporations worldwide. In the proposed work, StreamSage proposes to develop a new type of recognition and translation system that would greatly improve the accuracy with which the deluge of broadcast content can be automatically transcribed and translated for use by non-Arabic speakers. At the crux of the proposed system are new approaches for greatly reducing the errors in Arabic speech recognition and reducing the effect of recognition errors on translation accuracy. The goal of the project is to translate Arabic speech to English text with nearly the same accuracy as translating Arabic text to English by constructing a translation system from the ground up with a focus on the unique difficulties presented by spoken language