Intelligibility Assessment of Dysarthric Speech Using Extreme Learning Machine

Abstract

Artificial neural networks are known for their superior performance in many applications. Weights obtained during training are fine-tuned to the training data used. As a result the performance is reduced when tested with unseen data. Also, the use of backpropagation algorithms used in conventional neural networks is time consuming. Extreme learning machine classifiers are known for better generalization and quicker training compared to the neural networks with back propagation. In this work, the performance of the above mentioned classifiers on both seen and unseen data when used with cepstral coefficients is compared.