The study of computer-aided speech training method for deaf children based on Learning Vector Quantization

Learning Vector Quantization (LVQ) is a neural network model that has strong ability of data mining, which topological information prevalent in high dimensional input data can be transformed onto a low dimensional layer of neurons. In order to help the deaf children improve the ability of pronunciation, a method based on LVQ is proposed. We explored the effects of different parameters for the mapping, such as speech feature vectors, the structure and parameters of the neural network. Comparing the proposed method with SOM, the results show LVQ network made competitive supervised learning effectively and the clustering ability of LVQ had much better than SOM The regional expression made the borderline of map better.