Provided are an interested audio source cancellation method, a voice recognition method using the interested audio source cancellation method, computer-readable recording media recording programs for implementing the methods, and a voice recognition apparatus for performing voice recognition by using the interested audio source cancellation method. The interested audio source cancellation method includes steps of: receiving input mixed signals from two microphones and transforming the signals into a time-frequency domain through short-term Fourier transformation; setting an interested audio source cancellation vector for cancelling an interested audio source signal from the input mixed signals in the time-frequency domain; and generating a mixed noise signal by cancelling the interested audio source signal from the input mixed signals by using the interested audio source cancellation vector.