To develop a voice-operated procedure browserand astronaut assistant for the International Space Station.

As the project has progressed, the followingkey objectives have emerged:

• Procedure-independent
  The browser should be a general tool that can accept any suitably formatted new procedure as a data object.
  
• Flexible command vocabulary
  The user should be able to move freely around the current procedure and carry out other related activities, entirely using voice commands.
  
• Hands-free operation
  The system should be usable in an "open-mic" mode, where it can distinguish between voice commands requiring a response and speech directed at other people.
• Robust dialogue
  It should be possible to undo or correct any misrecognition or inappropriate response with a single voice command.
• High-quality spoken output
  Text read by the browser must be clearly comprehensible, even in noisy background conditions.

The current experimental version of Clarissa runs on a standard ISS laptop, and browses procedures written in an XML format. The markup language includes constructs for branch points, conditional steps, querying the user for numerical values, and linking to subprocedures. For our initial experiments, five procedures have been converted into Clarissa-compatible XML form: three for sample collection and testing of the Station's potable water supply, and two for space suit checkout.

The system has a vocabulary of about 260 words, and supports about 75 different commands. The user can scroll forwards or backwards, move to an arbitrary new step, preview or read out a non-current step without losing their place in the procedure, open a sub-procedure, and read safety-critical portions of the procedure in a mode which checks aggressively that steps have not been skipped. Other commands include support for recording, playing and deleting voice notes, setting timers and alarms, and querying status. Any misrecognition can be undone or corrected, using a command like "undo" or "no, I said go to step fourteen".

Speech recognition is in "open mic" mode. In the deployed system, the error rate for distinguishing between voice commands and non-system-directed speech is about 10%; together with Jean-Michel Renders from Xerox Research Center Europe, we have developed experimental methods using Support Vector Machines, which reduce the error rate to about 5%. This work may be integrated into a future version of Clarissa. Spoken output is performed using a recorded voice.

Clarissa has been implemented mainly using SICStus Prolog and a speech recognition toolkit provided by Nuance Communications. Application-specific spoken command grammars were constructed using the Regulus platform. Other programming languages and software packages used include Java, C, and SRI International's Gemini and Open Agent Architecture.

• First version of software tested by Astronauts at JSC, December 2002.
  
  
• STDO version of software delivered to JSC, July 2004.
  
  
• Software installed on ISS, January 13, 2005.
  
  
• Clarissa first used by Expedition 11 Science Officer and Flight Engineer John Phillips, June 27, 2005.

To the best of our knowledge, Clarissa is the first spoken dialogue system in space.