This paper describes monosyllabic recognition, specifically stop consonants, using dynamic critical band spectra and a three-layered neural network. Parameters of stop consonants were estimated by utilizing an adaptive digital filter with fast convergence property, and were converted into critical band spectra similar to human hearing. The spectra were programmed into a three-layered neural network, and the stop consonants were identified. The learning property of the neural network was improved by augmenting common bias to the input-output function of the network.