UNDERSTANDING

THE PROJECT WILL FIRST INVESTIGATE MEANS BY WHICH CONTENT-ADDRESSABLE MEMORY TECHNIQUES COULD BE EMPLOYED IN C3I SPEECH RECOGNITION. IN THE SECOND PHASE, A HARDWARE/SOFTWARE IMPLEMENTATION OF SELECTED TECHNIQUES WILL BE CONSTRUCTED. THE SPECTRAL AND TEMPORAL CHARACTE RISTICS OF SPEECH ARE REGISTERED AS ELEMENTS IN "FEATURE VECTOR SPACES." THE RESEARCH IN PHASE I WILL FOCUS ON UNDERSTANDING HOW THESE SPACES ARE POPULATED BY ACTUAL TEST AND REFERENCE PATTERNS. THIS DEEPER UNDERSTANDING WILL BE USED AS GUIDANCE FOR (1) DEVELOPING ALGORITHMS THAT CAN EASILY BE IMPLEMENTED WITH PARALLEL PROCESSING AND (2) DEVELOP SPEECH SIGNAL TRANSFORMATIONS THAT UNIQUELY IDENTIFY WORDS, PHRASES, ETC. VIA A CONTENT-ADDRESSABLE MEMORY. IN PHASE II A HARDWARE/SOFTWARE IMPLEMENTATION WILL BE CONSTRUCTED TO EVALUATE TECHNIQUES IDENTIFIED IN PHASE I.