Fuzzy Logic Networks for Speech Recognition

Martin Klimo,Stefan Badura,Stanislav Foltan

doi:10.26552/com.c.2013.2.13-18

Martin Klimo, Stefan Badura + Show 1 more

Open Access

https://doi.org/10.26552/com.c.2013.2.13-18

Copy DOI

Abstract

This paper proposes a massive fuzzy logic network which can be considered as a novel model of pattern classification network. Our approach introduces fuzzy logic circuits fulfilling the function of a binary classifier at first, which are connected into fuzzy logic networks with fuzzy flip-flop circuits as memories. Genetic programming is used as a circuit designing method. In order to establish design methodology, experiments aimed at testing the suitability of fuzzy logic operation sets, fitness functions and parameters of genetic algorithm were carried out. From trained circuits a hierarchical layered structure is built, where single layers consisting of given circuits are contextually dependent. Experiments with fuzzy logic circuits and fuzzy flip-flop network show some valuable results especially in the task of audio and visual speechrecognition.

Highlights

Speech is the most natural form of human communication
Many researchers resorted to the hidden Markov model (HMM) since it performs well in audio speech recognition [7, 8, 9]
In this paper a novel approach for speech recognition was introduced. It shows a good behavior for the task of lip-reading

Summary

Introduction

Speech is the most natural form of human communication. No wonder that with the development of technology, a man has come with an idea to communicate with the machine or a computer. The history of speech recognition started [1] It was recognition of isolated words, later the development of systems recognizing continuous speech. All these systems are based on the acoustic representation of speech [2]. Most of them use artificial neural networks (ANN) or hidden Markov models (HMM). If we consider just visual speech recognition, in [3] authors recognize silence and vowels, where an Elman topology of ANN is utilized and that is constructed from 3 layers. Many researchers resorted to the hidden Markov model (HMM) since it performs well in audio speech recognition [7, 8, 9]

Objectives

Methods

Conclusion