Speak up

Abstract

A review of the current state of speech recognition systems reveals that such machines are increasingly becoming more receptive to human utterances. Lately, a standard system seems to have been agreed upon by software developers. Analog signals from a microphone are digitized, compared with a vocabulary and the most likely word is then written to screen. Incoming speech is continually sampled for amplitude and frequency then compared with succeeding speech to improve the chances of selecting the desired word. A statistical routine based on Markov's method estimates the probability of the next value from previous values. The software will interrogate a library of phonemes, or word parts considered as building blocks of speech, and check the speaker's voice pattern from the initial training session. A window on the screen may show homonyms or similar-sounding words instead of the words which have been spoken. This is a design to facilitate fast editing.