Abstract: Based on the property that signal phase matching principle can extract signal frequency exactly,three sub-array signal phase matching principle is applied to extract the speech signal from strong background noise,and a speech quality evaluation method is proposed based on average power density spectrum distance.The subjective mean opinion scores(MOS),the signal-noise ratio(SNR) and the average power density spectrum distance as indexes are used to assess the speech qualities which processed by three sub-array signal phase matching principle,spectral subtraction and least mean square error estimation,and the noise reduction performance of these algorithms are compared.The simulation results demonstrate the validity of three sub-array signal phase matching principle in speech noise reduction.