NNARX Model of Speech Signal Generating System: Test Error Subject to Modeling Mode Selection

This paper presents comparative analysis of different processing errors when modeling speech signal generating system simulated by neural network auto regressive with extra input (NNARX). Feed forward NNs were used. Vowels were taken to train and test models. Gradient descent algorithm (GDA) is used for the training, back propagation algorithm (BPA) for the testing, and optimal brain surgeon (OBS) for the NN pruning. Results are given in Tables I-IV, and compared in conclusion