Abstract
The new spectral analysis method in time-frequency domain was proposed by authors. Here, we performed spectral analysis of vowel from low frequency to high frequency by using the proposed method, and performed speach recognition of the vowel of unspecified speakers in each frequencies. A vowel is a wave with th periodicity which mainly consists of a fundamental and harmonics. Vocal sound /i/ has few harmonics,and subsequently the number of harmonics increases in order of /u/,/e/,/o/ and /a/. However, the feature of vowel is not so simple as dependent by only the number of harmonics. For example, /i/ and /u/ are well alike and we can not seprate /i/ and /u/ by the number of harmonics. Then we introduced the concept of the accumulation wave which made the some vowel spectrums accumulate according to frequency, and tried separatin of vowels.
