The scripts that were used to implement hill-climbing feature selection for ISOLET in David Gelbart's PhD thesis can be found here. Those scripts build on a multi-layer perceptron (MLP) based ISOLET speech recognition system, and (optionally) a noisy version of the ISOLET corpus, both of which are both available here. Diagrams of the initial and final feature vectors for hill-climbing are here. Plain text lists of the initial and final feature vectors for hill-climbing can be downloaded here.