The Study Of Fuzzy Classifier Based On Genetic Algorithm And Its Application On Chinese Vowels Recognition

Dysarthria is a motor speech disorder resultting from neurological injury. It ischaracterized by poor articulation of language. As the development of science, smartinformation technology can not only help pathologists make more accurate diagnosis ofpatients with dysarthria, but also help patients with dysarthria carry out more targetedphysical treament, which makes them recover sooner.Recently, there are some researches on diagnosis and treatment of dysarthria based onChinese vowel recognition. Recognizing Chinese vowel accurately to aid the diagnosis andtreatment dysarthria, is the final target of the study of this dissertation.Currently, although speech recognition technology has achieve recognizing effectively,most of this technology is time-consuming and require large memory to work, which makes ithard to work on embeded systems. Fuzzy Pattern Classifier (FPC) has the advantage of lowcomputation cost and has good recognition effect to some specific data. To recognize Chinesevowel accurately, this dissertation will extract appropriate feature to descripe Chinese voweldata and use Genetic Algorithm (GA) to optimize the Fuzzy Pattern Classifier, as a resultdevelop a high-accuracy fuzzy classifier for Chinese vowel recognition.To extract feature of Chinese vowel, this dissertation study the Linear PredictionCepstrum Coefficient (LPCC), Formants and Mel Frequency Cepstrum Coefficient (MFCC),and extract these features from the samples of Chinese vowels respectively, then find the fitestcollection of features for the fuzzy classifier.To train the classifier, according to the problems with the current Genetic Algorithm,such as premature convergence and slow convergence near optimal solution, this dissertationpropose an improved Genetic Algorithm, which deals with the problems above with theapproch of finess scaling, simularity jugement and fitness thresholding, to avoid premature ofalgorithm and help convergence near optimal solution. Experiments show that, the improvedGenetic Algorithm is more stable and performs better on searching global solution than theoriginal one.By the studying of improved Genetic Algorithm and Fuzzy Recognition, we propose aChinese vowel Fuzzy Pattern Classifier based on improved Genetic Algorithm. The classifiertakes formant and MFCC of the sample speech data as input, gives the class of the sample asout. Experiments show that, the fuzzy classifier high accuracy on Chinese vowel recognition.