REMOTE VOICE INTERFACE FOR COMPUTER CONTROL

AFTER MANY YEARS OF RESEARCH, SPEECH RECOGNITION IS BECOMING PRACTICAL. EMERSON & STERN RECENTLY INTRODUCED SOLILOQUY (STABILIZER) VOICE INTERFACE SOFTWARE, WHICH MEETSMANY OF THE MARKET'S DEMANDS FOR SPEAKER-INDEPENDENT, NATURAL LANGUAGE, CONTINUOUS SPEECH INPUT. SINCE SOLILOQUY'S UNIQUE APPROACH REQUIRES ONLY 1/10 TO 1/100TH THE PROCESSING POWER OF TRADITIONAL APPROACHES TO RECOGNITION, IT IS ATTRACTIVE FOR INTEGRATION WITH THE NEW GENERATION OF MULTIMEDIA MACHINES. BECAUSE THE CAPABILITY DID NOT EXIST UNTIL NOW, NO ONE HAS ADDRESSED THE PROBLEM OFHANDLING USERS' SPONTANEOUS CORRECTION OF THEIR OWN ERRORS OR OF THOSE MADE BY A LISTENER (IN THIS CASE A RECOGNIZER). THE PROPOSED RESEARCH WILL DETERMINE THE FEASIBILITY OF SOLVING THIS PROBLEM WITH THE NATURAL LANGUAGE FACILITIES INHERENT IN SOLILOQUY. RESEARCH OBJECTIVES AND METHODS INCLUDE BUILDING A LINGUISTIC MODEL FOR ERROR CORRECTION, TAKING TELEPHONE NUMBERS AS AN EXAMPLE. LINGUISTIC DATA WILL BE COLLECTED ON HUMAN/HUMAN ERROR CORRECTION OVER THE PHONE AND WITH A "WIZARD OF OZ" STYLE EXPERIMENT IN WHICH CALLERS BELIEVE THEY ARE SPEAKING TO A COMPUTER, BUT ARE REALLY TALKING TO APERSON. BASED ON THE RESULTS, VOCABULARY AND GRAMMAR MODELSFOR ERROR CORRECTION WILL BE CREATED, USING EXISTING TOOLS, AND INTEGRATED INTO SOLILOQUY. NEW CODE WILL ALLOW THE RECOGNIZER TO SHIFT CONTEXT ON-THE-FLY BASED ON THE RESULTS OF RECOGNITION, AS OPPOSED TO ISSUING AN EXTERNAL COMMAND TO SHIFT CONTEXT. DATA WILL BE COLLECTED AND ANALYZED FROM A FINAL EXPERIMENT IN WHICH CALLERS REALLY ARE TALKING TO A COMPUTER. ANTICIPATED RESULTS INCLUDE CREATING A LINGUISTIC MODEL THAT CAN BE EXTENDED IN A PHASE II TO OTHER APPLICATIONS, INCLUDING REMOTE DATA ENTRY, DATABASE INQUIRY, COMPUTER CONTROL, AND CALL SORTING. SPEECH RECOGNITION FITS WITHIN THE HIGH PERFORMANCE COMPUTING INITIATIVE. TRULY FLEXIBLE RECOGNITION COULD PUT U.S. TECHNOLOGY SIGNIFICANTLY AHEAD OF RECOGNITION RESEARCH EFFORTS IN JAPAN AND EUROPE.