AUDITORY MODELING FOR SPEECH RECOGNITION

THE GOAL OF THIS PROJECT IS TO DEVELOP TECHNIQUES FOR COMPUTER SPEECH RECOGNITION USING PARALLEL DISTRIBUTED PROCESSING MODELS. THESE MODELS ARE BASED UPON THEORIES OF HUMAN COGNITIVE PROCESSING AND HAVE SEVERAL USEFUL PROPERTIES. THEY ARE VERY EFFECTIVE AND EFFICIENT AT INTEGRATING TOGETHER IMPRECISE INFORMATION, AND THUS FOR INTEGRATING MULTIPLE SOURES OF KNOWLEDGE FOR SPEECH RECOGNITION. THEY CAN ORGANIZE THEMSELVES AS THE RESULT OF EXPERIENCE, AND THUS LEARN TO USE SPEECH MECHANISMS FOR RECOGNITION THAT ARE ONLY POORLY UNDERSTOOD.THEY ARE NATURALLY PARALLEL AND ESPECIALLY SUITED FOR MASSIVELY PARALLEL MACHINE ARCHITECTURES, AND ARE ALSO RELATIVELY INSENSITIVE TO DAMAGE TO MACHINE HARDWARE. THESE PROPERTIES ALLOW SPEECH RECOGNITION SYSTEMS TO BE BUILT THAT CAN MAKE USE OF SPEECH INFORMATION THAT IS DIFFICULT TO DO WITH CONVENTIONAL SYMBOLIC COMPUTATION. THEY ALSO ALLOW SYSTEMS THAT CAN OPERATE IN HARSH ENVIRONMENTS SUCH AS THAT OF A MILITARY AIRCRAFT. THIS PROJECT INVESTIGATES ARTIFICIAL INTELLIGENCE ARCHITECTURES FOR COMPUTER SPEECH RECOGNITION BASED UPON THESE MODELS, BY TRAINING A MODEL RECOGNITION SYSTEM, EVALUATING ITS PERFORMANCE, AND ANALYZING THE ORGANIZATIONAL STRUCTURE IT DEVELOPS.