We consider the ability of a very simple feed-forward neural network to discriminate phonemes based on just relative power spectrum. The network consists of two neurons with symmetric nonlinear response over a spectral range. The output of the neurons is subsequently fed to a comparator. We show that often this is enough to achieve complete separation of data. We compare the performance of found discriminants with that of more general neurons. Our conclusion is that not much is gained in passing to real-valued weights. More likely higher number of neurons and preprocessing of input will yield better discrimination results. The networks considered are directly amenable to hardware (neuromorphic) designs. Other advantages include interpretability, guarantees of performance on unseen data and low Kolmogorov complexity.