Subband speech processing with neural networks

automatic processing method of the speech affected by noise comprising the steps of: - captureet digitization of speech (1) - extraction of several frames (15) corresponding to said signal (10) - chaquetrame decomposition (15) using an analysis system (20, 40) in at least two different frequency bands for two vecteursde representative parameters (45), - converting (50) vectors (45) into second vectors are relatively insensitive to noise (55) , each converter system (50) being associated with a frequency band. The learning of said converter systems (50) being formed on corpus of speech contaminated by noise (102).