Section: New Results

Speech-to-Speech Translation and Langage Modeling

Participants : Kamel Smaïli, David Langlois, Sylvain Raybaud.

Our work on Confidence Measures is now published in Machine Translation [9] . Now we are working on Speech-to-Text translation. We have proposed a method to segment audio input stream for machine translation. First, audio stream is splitted into overlapping segments useful for speech recognition; then, these segments are transcribed and regrouped; last, the obtained text flow is segmented into machine translation-friendly segments and translated. We incorporated this work into our speech-to-text machine translation system, and we evaluated our system for french-english broadcast news translation [34] . The following step consists in integrating Confidence Measures into the system in order to improve the integration between the both recognition-translation processes.

Moreover, we pursued our collaboration with Chiraz Latiri from the URPAH Team, University of Tunis. Running on our previous works (based on word-based machine translation system) we compared our respective methods in the scope of phrase-based machine translation [30] .