In this paper, a speech enhancement algorithm for dual-microphone mobile phones is im- proved. The proposed method is based on the exploitation of the coherence function and the quadrature mirror filter banks. This latter is designed by the genetic algorithm as cri- teria for noise reduction. Thus, the quadrature mirror filter banks are used as a mother wavelet in the speech enhancement algorithm. The adopted algorithm benefits of non- interfering signals statistics prediction. Hence, it can be applied with two closely micro- phones in small cellular mobile phones. Finally, referred to the objective measures and the time domain waveforms, the proposed method proves its effectiveness in term of speech quality and intelligibility compared with other algorithms when the speech is corrupted by different noise types.