When the speech signal and the random noise are transformed with discrete sine transform (DST), the discrete sine transform coefficients at the low frequency regions are predominantly speech, and at the high frequency regions are predominantly noise. According to this, a new speech enhancement method based discrete sine transform is proposed in this paper. The level of the trace of the signal present in the high frequency region of DST coefficients is estimated to obtain the threshold parameter, and after the signal-to-noise ratio estimation and speech present probability is calculated, an improved method is adopted. The experiment results show that it is an effective speech enhanced method.