Abstract: A speech recognition circuit in which acoustic power spectrum data corresponding to an input speech signal are sequentially generated, a vowel segment is determined based on the acoustic power spectrum data, and a vowel and a consonant are recognized based on the respective acoustic power spectrum data in the vowel segment and outside the vowel segment. The vowel segment is determined by first detecting a vowel prospective period in which the acoustic power spectrum data exceed a preset value, sequentially calculating the correlation of two successively generated acoustic power spectrum data to detect a stable vowel power spectrum data which provides a maximum correlation, and then calculating the correlation of the stable vowel power spectrum data and each of other acoustic power spectrum data in the vowel prospective period to detect a period in which the correlation is larger than a predetermined value.