To cancel noise with higher precision and to detect the speech of a speaker (speaker's voice) by reducing an arithmetic load for a processing system with limited arithmetic ability.

The configuration includes a first equalizer 14 for inputting a voice signal comprising a noise signal of a prescribed frequency characteristic and making the frequency characteristic of the noise signal flat, an adaptive filter 16 for emphasizing a linear spectrum based on the output signal of the first equalizer 14, a third equalizer 15 for changing the frequency characteristic of the output signal from the adaptive filter 16 so as to compensate the frequency characteristic of the first equalizer 14, a second equalizer 21 for detecting the voice signal based on the output signal of the adaptive filter 16, a sound pressure converting part 23 and a call state detecting part 45.