To search, Click below search items.


All Published Papers Search Service


Thai Word Recognition Using Hybrid MLP-HMM


Maleerat Sodanil, Supot Nitsuwat, Choochart Haruechaiyasak


Vol. 10  No. 3  pp. 103-110


The Hidden Markov Model (HMM) is a popular model for speech recognition systems. However, one of the difficulties in applying HMM is the estimation of the emission probabilities for constructing the Gaussian Mixture Models (GMMs). In this paper, we propose a method to estimate the state emission probabilities in HMM framework using Artificial Neural Networks (ANNs), particularly the Multi-Layer Perceptrons (MLPs). The proposed method can be considered as a hybrid MLP-HMM. Furthermore, tone information is one of highly potential features which could increase the recognition accuracy of tonal languages such as the Thai. Therefore, both MFCC features and tone features were extracted and served as the inputs for the MLP-HMM and the tone classifier. The posterior probabilities of outputs for each phone are represented as the state emission probabilities of the continuous density HMM framework. The experimental results showed that using the proposed hybrid MLP-HMM to train a Thai word recognition model helped improve the performance over the baseline system in terms of word error rates.


Hybrid MLP-HMM, Thai speech recognition, tonal language