Using bands of frequencies for vowel recognition for Polish language

Abstract

This paper presents a simple and useful method of vowel recognition of the Polish language. It relies on determining some characteristic bands of frequencies for each vowel. These bands are chosen so as to provide maximum separability of all vowels. Within each band we determine three parameters: average, a standard deviation and a maximum value. Comparing these values with the previously designated boundary values, we can classify a given vowel. As shown by the test, this method has a low percentage of an incorrectly recognized vowel. An additional advantage is its efficiency. It is four times faster than the method based on the formants.