This paper proposes a novel speech enhancement approach for a single-microphone system to meet the demand of quality noise reduction algorithms. The proposed system incorporates a perceptually motivated stationary wavelet packet filter-bank (PM-SWPFB) and improved spectral oversubtraction (I-SOS) algorithm together to enhance the speech degraded by non-stationary or colored noise environment. The PM-SWPFB is obtained by adjusting the uniformly spaced stationary wavelet packet tree in order to most closely mimic the critical-bands of the psycho-acoustic model. The PM-SWPFB is, firstly, used to decompose the input noisy speech signal into nonuniform sub-bands. Then, I-SOS algorithm is used to estimate of speech from each sub-band. The I-SOS algorithm uses a new noise estimation approach, to estimate noise power from each sub-band without the need of explicit speech silence detection. The sub-band noise estimate is updated by adaptively smoothing the noisy signal power. The smoothing parameter is controlled by a function of the estimated signal-to-noise ratio (SNR). The performance of the proposed speech enhancement system is evaluated objectively by SNR, Itakura-Saito distortion measure and subjectively by informal listening test. The results confirm that the proposed speech enhancement system is capable of reducing noise with little speech degradation remains acceptable in real-world environments, and the overall performance is superior to several competitive methods.Index Terms-speech enhancement, stationary wavelet packet transforms, critical-band rate scale, spectral-over subtraction, noise estimation.