FEASIBILITY OF UTILIZING HIGH LEVEL KNOWLEDGE TO AID SPEECH RECOGNITION IN CONTEXT ACTIVATED DEVICES

MOST AI SYSTEMS CURRENTLY BEING DEVELOPED WILL BENEFIT FROM SPEECH RECOGNITION CAPABILITIES. SOME ARCHITECTURES BEING CONSIDERED ARE BETTER SUITED THAN OTHERS TO PERMIT THE UTILIZATION OF "HIGHER-LEVEL" KNOWLEDGE TO IMPROVE SPEECH RECOGNITION QUALITY OR EFFICIENCY. ONE SUITABLE ARCHITECTURE IS THE CONTEXT ACTIVATED MEMORY DEVICE (CAMD, PATENT PENDING). THE PROPOSED RESEARCH WILL INVESTIGATE WAYS IN WHICHTHE DIFFERENT KNOWLEDGE SOURCES OF THIS ARCHITECTURE CAN BE INCORPORATED IN A SPEECH RECOGNITION COMPONENT. ONE PROJECT GOAL IS TO ESTABLISH THE FEASIBILITY OF INCORPORATING A SPEECH RECOGNITION SYSTEM ALREADY UNDER DEVELOPMENT AS A FRONT END TO THE EXISTING DEVICE, OR MORE LIKELY, TO IDENTIFY SYSTEMS IN THE DEVELOPMENT STAGE THAT CAN MAKE USE OF KNOWLEDGE AS IT IS REPRESENTED IN THE CAMD. THIS STUDY WILL DETERMINE HOW MODIFIABLE THOSE SYSTEMS ARE IN TERMS OF THE CONTROL STRATEGIES USED IN CONTINUOUS WORD RECOGNITION. IT IS BELIEVED THAT IF EFFICIENT USE IS TO BE MADE OF THE CAMD (OR SIMILAR AI SYSTEM) KNOWLEDGE SOURCES IN THE SPEECH RECOGNITION PROCESS, THEN STRATEGIES MUST BE PROVIDED TO GUIDE THE APPLICATION OF THIS KNOWLEDGE. BASED ON THE WORK OF NUMEROUS INVESTIGATORS WHO HAVE ALREADY STUDIED THE USE OF PROSODY IN HUMAN SPEECH RECOGNITION, IT IS EXPECTED THAT PROSODIC AIDS TO SPEECH RECOGNITION MAY SERVE THIS PURPOSE. SEVERAL WAYS IN WHICH PROSODIC AIDS MIGHT BE USED TO GUIDE THE APPLICATION OF THE HIGHER LEVEL KNOWLEDGE ARE IDENTIFIED AND PROPOSED RESEARCH WILL ASSESS THEIR USEFULNESS IN SPEECH RECOGNITION SYSTEMS. THE FEASIBILITY OF USING PROSODIC AIDS IN THIS WAY IS ASSESSED WITH RESPECT TO THE RELIABILITY OF THESE CUES UNDER VARIOUS CONDITIONS RELEVANT TO THE USE OF KNOWLEDGE-BASED SYSTEMS. CONDITIONS UNDER CONSIDERATION INCLUDE SURROUNDING SIGNAL CONTEXT (OR NOISE), LINGUISTIC CONSTRAINTS, CHUNK SIZE, AND MEMORY ENSEMBLE.