Verbal Input of a Set of Commands into the Besm-3M High Speed Electric Computer

摘要

Up to 50 instructions may be entered aurally into the BESM-3M digital computer by an arbitrary speaker in a room with noisy background with the aid of a speech recognition system which operates on signals divided into voice spectrum sub-bands and clipped speech parameters. In this system channel number 1 passes speech energy which exceeds a certain threshold value and which occupies a band below 500 Hz; channel number 2 reacts to signals whose frequencies are above 500 Hz. Other channels (consisting of bandpass filters) are used for distinguishing vowel sounds. Thus, channel number 5 recognizes 'a' and 'o' and partially 'e' and 'u' sounds and channel number 6 may distinguish 'i' and 'y' and partially 'u' sounds. The resulting pulses are counted, an action which identifies affricates and fricatives with sufficient accuracy. The resulting speech indices, coded in binary form, are interrogated by the BESM-3M computer input unit at a rate of 100 Hz. This information is subsequently stored in the BESM-3M main memory. It takes up to 1 sec to enter an aural command into the computer. All aural signals may be delivered to an octal output printer whose output makes it possible to study word structure and individual sounds.