We examine the problem of sampling noise encountered in time-domain simulations of linear and two-dimensional spectroscopies. A new adaptive apodization scheme based on physical arguments is devised for suppressing the noise in order to allow reducing the number of used disorder realisations, but introducing only a minimum of spectral aberrations and thus allowing a potential speed-up of these types of simulations. First, the method is demonstrated on an artificial dimer system, where the effect on slope analysis, typically used to study spectral dynamics, is analysed. It is, furthermore, tested on the simulated two-dimensional infrared spectra in the amide I region of the protein lysozyme. The cross polarisation component is investigated, particularly sensitive to sampling noise, because it relies on cancelling of the dominant diagonal spectral contributions. In all these cases, the adaptive apodization scheme is found to give more accurate results than the commonly used lifetime apodization scheme and in most cases better than the gaussian apodization scheme.