Author/Editor     Kaiser, Janez; Kalčič, Zdravko
Title     Učenje prikritih modelov Markova z genetskimi algoritmi
Translated title     Training of hidden Markov models with genetic algorithms
Type     članek
Source     In: Zajc B, editor. Zbornik 7. elektrotehniške in računalniške konference ERK'98. Zvezek B. Računalništvo in informatika, umetna inteligenca, robotika, razpoznavanje vzorcev, biomedicinska tehnika, močnostna elektrotehnika, didaktika, študentski članki; 1998 sep 24-26; Portorož. Ljubljana: Slovenska sekcija IEEE,
Publication year     1998
Volume     str. 125-8
Language     slo
Abstract     In this paper training of Hidden Markov Models (HMM) with Genetic Algorithms (GA) is presented. Genetic algorithms are robust search and optimization techniques, based on the mechanics of evolution and natural genetics. The main aim of using GA for training of HMM's instead of common Viterbi and Baum-Welch reestimation methods is the fact that the results of the later methods axhibit strong dependency on initial conditions. As GA exploit entire search space, this dependency should be overcome. Compared with Baum-Welch method, we achieved 0.53% increase in recognition results for phonemes from database SNABI.
Descriptors     MARKOV CHAINS
SPEECH ACOUSTICS
ALGORITHMS
ARTIFICIAL INTELLIGENCE
CROSSES, GENETIC