Vowel recognition with first, second and third order autocorrelation

The efficiency of an autocorrelation algorithm for speech recognition was evaluated experimentally, using vowels pronounced by Italians. The results were compared with those from other methods, such as average logarithmic spectral distance, cepstral distance, the likelihood ratio method, and cosH measurement. It is shown that higher order autocorrelations are necessary for attaining acceptable precision but because of the length of the computation the function is practical only for n = 2.