International Journal of Advanced Science and Technology Vol. 50, January, 2013 51 Automatic Speech Recognition Technique for Bangla Words * Md. Akkas Ali 1 , Manwar Hossain 2 and Mohammad Nuruzzaman Bhuiyan 1 1 Lecturer, Department of CSE & IT, University of Information Technology & Sciences (UITS), Baridhara, Dhaka-1212, Bangladesh 2 Software Engineer akkas.buet@gmail.com, manwar.cuet@gmail.com, mdnuruzzaman2001@yahoo.com Abstract Automatic recognition of spoken words is one of the most challenging tasks in the field of speech recognition. The difficulty of this task is due to the acoustic similarity of many of the words and their syllabi. Accurate recognition requires the system to perform fine phonetic distinctions. This paper presents a technique for recognizing spoken words in Bangla. In this study we first derive feature from spoken words. This paper presents some technique for recognizing spoken words in Bangla. In this work we use MFCC, LPC, GMM and DTW. Keywords: Feature extraction, speech recognition, framing, overlapping, hamming window. 1. Introduction The technique for automatic speech recognition varies for working language. Every language has its own style of pronunciation, i.e. ‘Khabor’ and ‘Kobor’ are very close to eac h other in uttering. However the meanings are very far-apart. This pronunciation difference cannot be tolerated in most of the languages i.e. French because controlling the vibration of vocal cord and movement of lips differ from language to language [1]. However in Bengali based on locality, for the same word a notable tolerance limit can be seen. So, different approach is demanded for speech recognition in Bengali language. The main task of this paper is to recognize Bengali words through speech recognition technique of our proposed model. Initially we analyze the set of speech and extract essential features based on signal processing concept. From these features we have done parametric representation, mathematical model, and signal flow diagram for stepping towards our desired result. We used appropriate technique to calculate accurate distance between feature and reference matrix [2]. Consequently we represent four speech recognition models and compare them showing recognition rate (%) and elapsed time (sec). Finally we observed that from one of our models we can get highest rate of perfection for a set of Bengali words. 2. Objectives  An efficient feature extraction method will propose to recognize Bangla isolated speech.  Measure the success rate of the proposed model.  Create a Bangla isolated speech recognizer using that feature extraction method.