In general, the implementation of several parameter groups is seen to be effective in clearing up the nature of a phoneme, because each of them might have a typical feature which does not appear in the others. A speech visualization system has been developed as a speech training aid for the hearing impaired. Beside the formant tracker, the filter bank, and the neural network for manner-of-articulation, the system adopts neural networks for extracting cues of place-of-articulation. It was shown in a previous experiment that the information extracted by all of those devices are very useful for human judgment of the visual pattern. In this study of word recognition, in which the vocabulary is freely and easily constructed, an attempt was made to add cues of place-of-articulation to improve the performance. The results show that the addition of parameters was very effective, and the recognition rates using the proposed parameters were much higher than the usual rates. When 50 place names uttered by 30 males from the JEIDA database were tested, the recognition rates were 97.8% in the Bayes-decision-rule distance.