Accession Number : AD0622988

Title :   SOUND RECOGNITION IN A NEURAL NETWORK.

Descriptive Note : Master's thesis,

Corporate Author : AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OHIO SCHOOL OF ENGINEERING

Personal Author(s) : Minard,Leon Dale ,Jr.

Report Date : AUG 1965

Pagination or Media Count : 127

Abstract : The sound patterns for eight sustained phonemes are used as input to a single-level network of eight Steele neuromimes. Each pattern is a loudness (neuron firing rate) versus frequency representation of the output from the cochlear section of an electrical analog of the ear. Recognition of a pattern occurs when one of the eight outputs of the network, as designated, becomes greater than all the rest. To meet the requirement for recognition, the learning process involves a matrix transformation that allows the elements of the matrix to change. The transformation used was 20, the dimensionality of each pattern, to 8, the dimensionality of the network output. The network, after careful selection of network parameters, achieved continuous recognition for three consecutive cycles of the eight patterns. (Author)

Descriptors :   (*PATTERN RECOGNITION, ACOUSTIC SIGNALS), (*AUDITORY NERVE, SIMULATION), (*ELECTRICAL NETWORKS, AUDITORY NERVE), (*BIONICS, SPEECH RECOGNITION), EAR, ANALOG SYSTEMS, LEARNING, COMPUTER LOGIC, NERVOUS SYSTEM, HEARING, COMPUTER PROGRAMMING, MATRICES(MATHEMATICS)

Distribution Statement : APPROVED FOR PUBLIC RELEASE