Mlp Mlp Rbf Rbf Gn Gn Cart Ntn

Speaker independent vowel recognition is a di cult pattern recognition problem. Recently therehas been much research using Multi-Layer Perceptrons (MLP) and Decision Trees for this task. Thispaper presents a new approach to this problem. A new neural architecture and learning algorithmcalled Neural Tree Networks (NTN) are developed. This network uses a tree structure with a neuralnetwork at each tree node. The NTN architecture o ers a very e cient hardware implementationas compared to MLPs. The NTN algorithm grows the neurons while learning as opposed to back-propagation, for which the number of neurons must be known before learning can begin. The newalgorithm is guaranteed to converge on the training set whereas backpropagation can get stuck inlocal minima. A gradient descent technique is used to grow the NTN. This approach is more e -cient than the exhaustive search techniques used in standard decision tree algorithms. We presentsimulation results on a speaker independent vowel recognition task. These results show that the newmethod is superior to both MLP and decision tree methods.