Vowel recognition in Romanian language

Abstract

In this paper are presented results obtained in a vowel recognition task applying unsupervised and supervised fuzzy algorithms and neural networks. The vowels, uttered from 10 speakers each in 250 different contexts are recognized using as features the first three formant frequencies. After an introduction, the feature extraction is presented and then the two fuzzy algorithms (fuzzy ISODATA, fuzzy k-NN) and the two neural nets (nonlinear perceptron, Kohonen map) used for recognition are given.