Recognition of Vowel Sounds from a Clipped Speech Signal

A method of obtaining features and an algorithm for the recognition of vowel sounds are described. Before clipping the speech signal passes through circuits in one of which mainly the range of the first formant is amplified, in the other the mean audio frequencies. The number of pulses at the outputs of these circuits corresponds approximately to the first two formant frequencies and characterizes the individual vowels well (with the exception of the discrimination of $[i]$ and $[y]$). The reliability of recognition of the hard variants of the vowels and $[i]$ by two features amounts to 88.2%. The soft variants of the vowels, having an inhomogeneous (diphthongal) structure, are recognized in more than 90% of cases as one of their component sounds.