WHO:	John Dowding , NASA Ames Research Center
TOPIC:	A Spoken Dialogue System for Geological Exploration
ABSTRACT:	Two recent field tests have demonstrated that speech recognition and synthesis are now practical human-computer interfaces for astronauts in space suits. The RIALIST group at NASA Ames Research Center has developed a speech-in/speech-out interface to an automated geologist's field assistant. This system helps an astronaut on EVA with logging samples that have been collected, logging and uploading images and voice annotations, and tracking the astronaut's location and health status with GPS and physiological sensors. The end-product of a successful geological survey is a database of samples, images, and voice annotations, indexed by time and location, with explicit connections between correlated items. The system also supports a limited capability to command a robotic assistant that can move to known locations, carry tools and samples, and take pictures. The presentation will include slides and video from the most recent field test conducted at the Mars Desert Research Station, near Hanksville, Utah. I will give an overview of the architecture and technical approaches used in this system, and how these differ from other current commercial and research approaches to speech recognition, and emphasizing our approach to solving the open-microphone problem, using probabilistic grammar-based language models (PGBLMs). I will describe the compilation process that leads from Typed Unification Grammars to grammar-based language models, and how small amounts of training data are used to tune GBLMs to the domain. Time permitting, I will give a demonstration of this system, and/or a demonstration of a related system, CLARISSA, a procedure-reading assistant that is intended for deployment on the International Space Station in 2004.e
WHEN:	10/7/2003 2:00:00 PM
WHERE:	Computer Science Building 703

Events Homepage