Improving a phoneme classification neural network through problem decomposition

The authors discuss how a methodology called problem decomposition can be applied to an AP-net, a neural network for mapping acoustic spectra to phoneme classes. The network's task is to recognize phonemes from a large corpus of multiple-speaker, continuously spoken sentences. The authors review previous AP-net systems and present results from a decomposition study in which smaller networks trained to recognize subsets of phonemes are combined into a larger network for the full signal-to-phoneme mapping tasks. It is shown that, by using this problem decomposition methodology, comparable performance can be obtained in significantly fewer arithmetic operations.< >