The work of an Automatic Speech Recognition system is to convert speech signal into a text message accurately. Transformation of the spoken words independent of the speaker, we need a microphone device to record the speech and speaker decides what to say and actually speaks a word or sentence. The Matrix laboratory (MATLAB) software then produces a speech wave form. In this project, artificial neural networks are used to do speech recognition. The paper will be investigated in two steps, consisting of the pre-processing part and the postprocessing part with Artificial Neural Networks (ANN). These two parts are explained in detail and speech recognizers using different ANN architectures will be implemented on Matlab. Speech processing is one of the most important application area in the field of digital signal processing. Different fields for research in speech processing are speaker recognition, speech recognition and speech coding. In this project, the Mel Frequency Cepstrum Coefficient (MFCC) feature has taken for constructing a text based speech recognition system. Few changes to the feature extraction technique of MFCC are also suggested to improve the speech recognition efficiency. Although speech recognition devices are already available, now in the market, their development is based on statistical techniques.