We propose a new methodology to denoise MRS spectrum with a focus on the acquisition time diminution. Using a discrete orthogonal polynomials, we detect two types of areas : homogenous and non-homogenous (metabolite peaks). Once these areas detected, we compute the Noise Level Function (NLF). Then, using the NLF, we use orthogonal polynomials to reconstruct a signal with a strategy for each type of area. As results, a denoising method is provided and it helps to correct the noise due to the acquisition time diminution with a good metabolite peaks conservation.