Hidden Markov Model and Learning Vector Quantization for Indonesian Speech Recognition: A Comparative Study

摘要

A characteristic of a language is different one another. In order to make an application of speech recognition required a robust method for recognizing the characteristic of a language. However, the method, that proven has a high accuracy for English speech recognition, results highaccuracy for Indonesian speech recognition. The aim of this research is to compare two methods including Hidden Markov Model (HMM) and Learning Vector Quantization (LVQ). The research consists of 2 steps, i.e., (i) the development of Indonesian speech recognition based on HMM and LVQ and (ii)performance comparison against both methods. Meanwhile, there are 4 sub-processes in the speech recognition development, namely input process, Mel-Frequency Cepstral Coefficients (MFCC) feature extraction process, training process, and testing process. Subsequently, the results of both testingprocesses are compared each other. The experiment is conducted for 300 data and the final results show that the performance of HMM method is about 97.67% for 20 MFCC coefficients and 11 HMM states.