PURPOSE: To emphasize a consonant with which a noise is mixed and improve the recognition rate of a speech, specially, consonants by making the consonant with the vowel following the consonant by utilizing the tasking effect of reverse masking.

CONSTITUTION: This device is equipped with a masking process part 3 as a masking means which calculates the mask quantity of the reverse masking effect on an inputted speech and performs an output process for the speech according to the mask quantity, an analytic part 4 as an analyzing means which analyzes the speech outputted from the masking process part 3, and a speech recognition part 5 as a recognizing means which recognizes the speech on the basis of the analytic result of the analytic part 4. The level of the speech is compared with the mask quantity and the speech itself or a specific-level signal is outputted to the masking process part 3 on the basis of the comparison result. Therefore, when there is a consonant which has small power in front of a vowel having large power, the noise mixed with the consonant is masked by the reverse masking by the vowel.