This work presents a time-frequency mask for noise reduction in cochlear implants. The proposed attenuation function is obtained from the minimization of the mean square error between the speech envelopes and its estimation, taking into account computational complexity constraints. The performance of the proposed method was evaluated using two objective intelligibility criteria, assuming contamination by babbling noise and different signal-to-noise ratios. The results indicate that the developed mask provides greater intelligibility to cochlear implant users, compared to Wiener and Wiener root masks, both widely used in the literature.