Convolution Neural Network Model for Recognition of Speech for Words used in Mathematical Expression

Abstract

Speech recognition is translation of audio signal into human readable form. Speech recognition plays a vital role in various areas such as in signal processing, dictation system, command and control, simple data entry.  Speech recognition in dictation system helps the disabled people. In this paper authors have performed the experiment for speech recognition of   mathematical words which is helpful to disabled people. Now a day’s the use of deep learning in various applications is challenging for the improvement of model. In this paper authors have used CNN model to improve the recognition accuracy. Authors have selected 17 mathematical words which are the most commonly used in mathematical expression. Rectified Linear unit activation function is used to train the CNN because of its fast computation. This paper evaluates the model for MFCC and Delta MFCC features for Adam and Adagrad optimizers. Result shows that Delta MFCC gives an accuracy of 83.33 % for both Adam and Adagrad optimizer. It indicates that Delta MFCC gives better result than MFCC. Result also shows that Adagrad with Delta MFCC trains the model earlier than Adam.