Noraini Seman, Zainab Abu Bakar, Nordin Abu Bakar, Haslizatul Fairuz Mohamed, Nur Atiqah Sia Abdullah, Prasanna Ramakrisnan, Sharifah Mumtazah Syed Ahmad

Department of Computer Science
Faculty of Computer and Mathematical Sciences
Universiti Teknologi MARA (UiTM), Shah Alam, Selangor, MALAYSIA
{aini, zainab, nordin, fairuz, atiqah, prasanna}

Department of System and Network Department
College of Information Technology
Universiti Tenaga Nasional (UNITEN), Kajang, Selangor, MALAYSIA
This email address is being protected from spambots. You need JavaScript enabled to view it.


This paper describes speech recognizer modeling techniques which are suited to high performance and robust isolated word recognition in speaker-independent manner. In this study, a speech recognition system is presented, specifically for an isolated spoken Malay word recognizer which uses spontaneous and formal speeches collected from Parliament of Malaysia. Currently the vocabulary is limited to ten words that can be pronounced exactly as it written and control the distribution of the vocalic segments.  The speech segmentation task is achieved by adopted energy based parameter and zero crossing rate measure with modification to better locates the beginning and ending points of speech from the spoken words. The training and recognition processes are realized by using Multi-layer Perceptron (MLP) Neural Networks with two-layer feedforward network configurations that are trained with stochastic error back-propagation to adjust its weights and biases after presentation of every training data. The Mel-frequency Cepstral Coefficients (MFCCs) has been chosen as speech extraction approach from each segmented utterance as characteristic features for the word recognizer. The MLP performance to determine the optimal cepstral orders and hidden neurons numbers are analyzed. Recognition results showed that the performance of the two-layer network increased as the numbers of hidden neurons increased. Experimental result also showed that the cepstral orders of 12 to 14 were appropriate for the speech feature extraction for the data in this study.

Keywords: Multi-layer Perceptron, Feedforward, Mel-frequency Cepstral Coefficients, Hidden Neuron, Target vector.

Full Download