This paper discusses the speech recognition process using the classifier Artificial Neural Network. In this work, MLP structure with error back propagation algorithm is used in which the errors are propagated backwards from the output nodes to the input nodes. Usually, one hidden layer is enough for efficient speech recognition or classification. In my research study, only one hidden layer is considered. The number of nodes in the hidden layer is adjusted empirically for the better performance of the system. The main problem here is to classify the speech sample feature vectors into several speech classes. To reduce the volume of data which is to be feed in the input layer clustering technique is used in this study. The clustering of data decides how the related data can be categorized into different classes. The K-means clustering, one of the clustering techniques is used in the present study. Speech recognition process by both ways i.e., by Speaker dependent as well as Speaker independent ways techniques are used in this work. Comparative performance evaluation is done for both mode of Speech Recognition system.