PROCESS FOR RECOGNISING VOICE SIGNALS AND DEVICE FOR IMPLEMENTING IT

摘要

The invention relates to a voice recognition system operating in the following manner: In a training phase, reference voice signals corresponding to the voice signals to be recognised are processed as follows: From the reference voice signal available for processing are obtained first characteristic vectors which are then subjected to vector quantification in which the individual vectors are compared with a plurality of cluster vectors having an address and stored in a code book and a first cluster vector is found which corresponds most closely to the characteristic vector concerned in relation to a predetermined latitude. The address of the first cluster vector found is then stored in an address store. In the recognition phase the voice signals to be recognised are processed as follows: Two characteristic vectors are obtained from the voice signals available for processing and then subjected to classification in which the individual vectors are compared with the first cluster vectors characterised by the addresses in the address store. By means of the comparison, that reference voice signal is identified as the voice signal to be recognised, the cluster vectors of which correspond most closely to the second characteristic vectors in relation to the latitude.