Sequential System for Recognizing Spoken Digits in Real Time

A real time system for recognizing the ten digits spoken in Italian language is described.

An analyzer, composed of 17 filters uniformly covering the frequency range from 110 Hz to 5600 Hz and the corresponding rectifying and level-detecting units, yields a set of 17 binary outputs as functions of time. These outputs are processed by a real time system composed of ten units, each of which performs the classification of one of the ten digits. Each unit is characterized by a few combinational networks, recognizing the vowels or some consonants or some typical transitions of the word, and by a sequential circuit analyzing the time evolution of the outputs of the combinational networks.

The performance of the system is without faults in recognizing the ten recorded voices used in the learning process. Much less efficiency may be found in recognizing other voices, but for many speakers some training is sufficient to obtain better performances.