Evolvable hardware applied to voice recognition

Abstract

This paper presents some results of the application on Evolvable Hardware (EHW) in the area of voice recognition. Evolvable Hardware is able to change inner connections, using genetic learning techniques, adapting its own functionality to external condition changing. This technique became feasible by the improvement of the Programmable Logic Devices. Nowadays, it is possible to have, in a single device, the ability to change, on-line and in real-time, part of its own circuit. This work proposes a reconfigurable architecture of a system that is able to receive voice commands to execute special tasks as, to help handicapped persons in their daily home routines. The idea is to collect several voice samples, process them through algorithms based on Mel - Ceptrais theory to obtain their numerical coefficients for each sample, which, compose the universe of search used by genetic algorithm. The voice patterns considered, are limited to seven sustained Portuguese vowel phonemes (a, eh, e, i, oh, o, u).