PROLOG-BASED EXPERT PATTERN RECOGNITION SYSTEM SHELL FOR TECHNOLOGY INDEPENDENT,USER-ORIENTED CLASSIFICATION OF VOICE TRANSMISSION QUALITY

摘要

An operational,prototype Expert System Shell is presented for objectively determining voice transmission quality directly from channel input/output speech.Thus the technique is system and technology independent with application to both analog and digital systems.The Expert System allows for direct input of additional user knowledge,including bias.It was developed in a Turbo Prolog* environment since it is user friendly and provides a natural language interface to the FORTRAN modules.The shell provides relatively easy adaption to changes suggested by future research.This system currently processes speech,extracts parameter values,classifies the speech into one of five classes,and provides reasoning for obtaining that classification.The performance objective is to minimize the average probability of error where objective classifications are compared to subjective panel classifications on training data.