In this report, an algorithm is proposed for the use of a neural network as a speaker independent feature extractor. This algorithm can extract normalized features with an arbitrary number of dimensions. In order to evaluate the performances of the proposed algorithm, a combination with continuous type HMMs, for several numbers of continuous density mixtures is tested. For comparison, several phrase recognition experimental results are given. The recognition rate is around 70%, but many directions are to be investigated in the close future. It is believed that a neural network can be used as a new speaker independent feature extractor and give good results, especially in language identification.