Abstract: A speech training method using output-layer visualization of Self-Organizing Map（SOM） is proposed.SOM is a neural
network model that can transform input data onto two-dimensional plane or curve surface of output layer neurons.The subjects guide their pronunciation through visual feedback from positional information of output layer neurons.In order to improve the clustering of SOM，the authors make strengthen training and discuss how to choose the number of neurons in the output layer.The results show the proposed speech training method is simple and straightforward.It effectively realizes “speak when seeing the picture”.