We present a noise-filtering scheme which works on a chaotic signal containing a certain level of noise. Our method exploits the simplicity of Rosa's filtering scheme, as well as the high applicability of the maximum likelihood method. We tested our scheme on signals from various dynamical systems and found that the noise amplitudes could be reduced up to a few percent. In the study, we are learned that our scheme could also be used for inferring the underlying dynamics of a received chaotic signal when no a-priori knowledge of the dynamics was given.