This paper describes the universal denoising for the finite-input-general-output channel. The present work is in the case where the components of the underlying noise-free signal are still finite valued, yet their noisy observations take values in a general alphabet. This channel is assumed as discrete memoryless in this case. A discrete signal corrupted by a known discrete memoryless channel can be asymptotically optimal and practically denoised with no a-priori knowledge of statistical (or any other) properties of the signal.