Recognition of coded speech (phonocodes)

Coded speech according to a PHONOCODE, such as SOTINA, consists of words combining 3 vowel classes O (=o,u), A (=a,α), I (=i,e) and 3 consonant classes S (=s, ), T (=t,k), N (=n,m). Such coded words are designed to be internationally pronounceable by everybody, and immediately recognizable by a machine, such as a PHONO-DECODER. Human recognition tests (for orienting the best vowel and consonant combinations) are made: a) for French at the C.N.E.T. (Department J.-M. PERSON); b) for German, and next for English, at the University of Trier (Germany), (Phonetics Institute, J.-P. KÖSTER). Statistical results are given, depending on various telephonic conditions (noise immunity) and on initial, intermediate or final phoneme positions in the words (proximity immunity). Automatic recognition tests (for optimizing the systems) are made at the C.N.E.T.: a) with the PHONODECODER I, a machine which is in course of development (Department P.LORAND, team M.CARTIER); b) with DYNAMIC PROGRAMMING (Team G.Mercier). Present statistical results are given for various words combining vowels V and consonants C, such as CVV, VCV, CVCV, pronounced by men and women, without prior adaptation of the machine to every word or speaker.