This paper describes an adaptive filter construction technique and a spectrum estimation technique which uses a model for the variations of speech characteristics. The filter construction technique is applied to the adaptive analysis for the spectrum estimation technique. These techniques are the bases for the speaker independent vowel recognition which is tolerant of the variations of the speech features. Although the model is simple and efficient for the calculation, it is experimentally shown that the model yields well approximated spectra for the varied vowel parts in CV syllables and for the speaker dependent variations of isolated vowels.