Speech Translation Statistical System for Teaching Environments and Conference Speeches

Abstract

The synergic combination of different sources of knowledge is a key aspect in the development of modern statistical translators. The effect and implications of adding additional other-than-voice information in a voice translation system for teaching environments and conference speakers is described in this work. The additional information serves as the bases for the log-linear combination of several statistical models. A prototype that implements a real-time speech translation system from Spanish to English is presented. In the scenario of analysis a teacher, or presenter, as speaker giving its presentation could use a real time translation system for foreign students or participants. The speaker could add slides or class notes as additional reference to the voice translation system. Should notes be already translated into the destination language the system could have even more accuracy. In this paper, first, we present the theoretical framework of the problem, then, we summarize the overall architecture of the system, next, we specify the speech recognition module and the machine translation module, then, we show how the system is enhanced with capabilities related to capturing the additional information, and, finally, we present the performance results of the developed system.