Erkennung gesprochener Zahlworte mit Funktionsmodell und elektronischer Rechenanlage

A simple function-model of the human ear is used to transfer the sound of the German spoken numbers into channel-time patterns, which are coded and fed into an electronic computer ER 56. For each number a time-normalized characteristic pattern is formed and stored in the computer. For recognition a simple matching program in the computer is used, to compare the time normalized average pattern of all 10 numbers with the individual time normalized pattern to be recognized. Using this method, 95% of more than 1000 numbers - spoken without special care - could be recognized correctly.