next up previous
Next: HYPERSPEECH Up: SPEECHWEAR: A mobile speech Previous: ADAPTING LTI FOR SPEECH

SYSTEM DESCRIPTION

The SPEECHWEAR system makes use of a Toshiba T4900ct notebook computer containing a 75MHz Pentium processor, 40Mb of RAM and running Windows NT 3.5. Input is through a head-mounted microphone and output through a small head-mounted (grey-scale) VGA display with a speaker attached to its frame. Communications is is by means of a WaveLAN transmitter.

Recognition services are provided by a real-time implementation of the SPHINX-II recognition system [7], a continuous-speech speaker-independent system based on hidden Markov modeling. Spoken language interpretation made use of the PHOENIX [2]. The system implements a ``continuous listening'' protocol[12] that allows the task to be performed hands-free. A modified mouse is provided to turn the system on and off. Figure 1 shows a diagram of the system.

The NCSA Mosaic browser[3] provides the interface to the task hypertext document. It was modified by merging the spoken language code into it to create a single multi-threaded application. Inspection data was recorded through the use of FORMs embedded in the task document. As the interface is a speech-enhanced version of the Mosaic browser, communication is through the standard http protocol and makes use of servers and CGI[4] scripts to implement the inspection system.



Alex Rudnicky
Thu May 30 19:32:28 EDT 1996