|
The speech component of the system allows a user to interact with the robot using voice commands. There are two main components employed by this part of the system - ViaVoice and NAUTILUS. ViaVoice is a commercial voice recognition program from IBM. It is used to convert the voice commands into text so that NAUTILUS can analyze them. NAUTILUS is a program written in Common LISP that we received from the Naval Research Laboratory. It takes the natural language input from ViaVoice and interprets it into commands that can be sent directly to Guinness.
We use two tools included with the ViaVoice SDK for speech recognition. The first is a tool to create a customized grammar. This greatly reduces the number of words and phrases that ViaVoice tries to recognize - which helps the accuracy of the system immensely. Next, we use a program called Reco that uses our customized grammar to convert speech to text. Reco then passes any valid text commands on to the NAUTILUS system.
NAUTILUS takes these valid text commands and converts them into commands that Guinness understands. This process starts with a customized dictionary of words NAUTILUS recognizes. This allows the system to parse input words and phrases and translate them into a semantic interpretation. The semantic interpretation is then used to issue a command directly to Guinness.
More info on Nautilus can be found at the NAUTILUS description on NRL's web site.
|
