CNN WITH DATA AUGMENTATION FOR VOICE PATHALOGY IDENTIFICATION

Abstract

Deep learning techniques plays an important role  in speech analysis and the performance has been greatly improved in the past few years. In this work, voice disorder classification is performed using deep convolution neural networks and the disorders have been successfully classified. To identify voice disorders, the deep learning based techniques are more suitable and they typically required large training datasets. The availability of the real world speech data is limited; the speech augmentation techniques are employed to increase the training data.  Audio augmentation is proven to be useful for training the neural network and to make effective predictions. It also helps in avoiding overfitting and improves robustness of model. The classification accuracy of the model is improved from 87% to 98%, due to audio augmentation.