To provide a device with a high recognition rate which suppresses the distortion of a speech signal of a vocal sound of low sound pressure level by extracting feature parameters from a speech signal amplified by a 2nd signal amplifying means in a section in which the sound pressure level of the speech signal is low and performing a speech recognizing process.

A 1st signal amplifier 5 amplifies the voltage value of the speech signal to a proper value and a 2nd signal amplifier 7 amplifies the speech signal so that its sound pressure level has a proper value for a consonant of low sound pressure level. Then a feature extracting means 15 extracts feature parameters by using the 1st speech signal in a section in which a decision means 13 decides that the 2nd speech signal overflows, and extracts feature parameters by using the 2nd speech signal in other sections. Then a recognizing process part 19 performs pattern matching on the basis of the feature parameters extracted by the feature extracting means 15 and the power value calculated by a power calculating means 21 to perform the recognizing process.