Speech enhancement is an important proceeding which it can be achieved by estimate the clean speech from noisy speech. The clean speech is corrupted by white Gaussian noise . In this paper, thresholding the wavelet coefficients, that can be done by standard deviation method for each frame by level dependent thresholding using semi soft threshold and soft threshold .The results in this work indicate that using wavelet transform in speech enhancement application provides a good quality (measured in object SNR test) as well as provides other advantages which make it more suitable for some application s such as a split of frequencies and enhancement the speech in real time applications .