Apparatus for speech recognition recognizes input speech by discriminating phonemes included in an input speech signal such that a fundamental unit for recognition is each phoneme. The apparatus comprises a standard pattern storing portion for storing phoneme standard patterns of phonemes or phoneme groups; a spectrum analyzing portion for obtaining parameters indicative of the spectrum of an input speech signal; a similarity calculating portion for calculating the degree of similarity between the results from said spectrum analyzing portion and standard patterns from said standard pattern storing portion by using a statistial distance measure; a segmentation portion for effecting segmentation by using time-dependent variation of power information and the results from said similarity calculating portion with said power information being obtained in connection with both a low-frequency range spectrum and a high-frequency range spectrum of said input speech signal; and a phoneme discriminating portion for recognizing phonemes by using the results from said similarity calculating portion.