A method of voice communication with a digital computer
Stanley R. Petrick, H. M. Willett · 1960
A pattern recognition procedure for achieving automatic identification of spoken words has been developed and instrumented using an eighteen channel vocoder and a general purpose medium scale computer, the AFCRL Cambridge Computer. The process depends upon the representation of a spoken word by a sequence of octal digits which describe the amount of instantaneous power in each of eighteen frequency bands at time intervals of 1/50 second. Essentially, recognition is achieved by matching such a digital representation of a spoken word against a set of stored word "masks", one for each word to be recognized.