To provide a satisfactory target audio signal even if the audio transmission characteristic of an audio system is non-linear.

A speaker 22 is driven by an input signal xk and a microphone 23 provided in the desired place of the space collects the sound so as to obtain an observation signal yk. Noise (n) is added to xk and the speaker 22 is driven. An inverse convolution operation is executed on a difference between an observation signals yk' and yk by (n)(31). The inverse convolution operation is executed on an error Δk between the target signal (d) and yk by the result hk (32), a gain α is given to Δxk being the result, it is added to xk and the speaker 22 is driven. The error Δ(k+1) between the observation signal y(k-l-l) and (d) is obtained and it is discriminated whether the latter of the average square error of Δk and Δ(k+1) is smaller (51). When it is smaller, α.Δxk is added to xk and a new input signal is obtained.