Graph based semi-supervised learning methods applied to speech recognition problem

Abstract

Speech recognition is the important problem in pattern recognition research field. In this paper, the un-normalized, symmetric normalized, and random walk graph Laplacian based semi-supervised learning methods will be applied to the network derived from the MFCC feature vectors of the speech dataset. Experiment results show that the performance of the random walk and the symmetric normalized graph Laplacian based methods are at least as good as the performance of the un-normalized graph Laplacian based method. Moreover, the sensitivity measures of these three semi-supervised learning methods are much better than the sensitivity measure of the current state of the art Hidden Markov Model method in speech recognition problem.