This paper describes an attempt to develop a new phonetic visualization method of speech training system for the speech impairment. A problem associated with the conventional speech training system is briefly reviewed. Reasons for attempting to develop the new method are given. The proposed method uses neural network approach to visualize phonetic patterns. The method and its display capabilities are described. Also the experimental result of visualizing five Japanese vowels /A/,/I/,/U/,/E/,/O/ by using the proposed method is described.