摘要 |
According to the method, the fibres of the auditory nerve are modelled, learning sounds are coded according to the same model, in order to constitute the lexicon of classes of parameters which are indicative of the learning sounds and these classes are isolated from one another, the sounds of the incident speech are coded according to the same model, for these sounds the values of the same parameters are determined and those of the classes which contain the parameters of the incident sounds are sought. The method of the invention permits an excellent recognition score and is well adapted for interrogation of data bases. <IMAGE> |