Despite good performance in quiet environments, there are still significant gaps in speech perception in noise between normal-hearing listeners and hearing-impaired listeners using devices like hearing aids or cochlear implants (CIs). Much effort has been invested to develop noise reduction algorithms that could fulfill these gaps, but few of them have the ability to enhance speech intelligibility without any prior knowledge of the speech signal, including both statistical properties and location information. In this study, a single-channel noise reduction algorithm, based on a noise tracking algorithm and the binary masking (BM) method, was implemented for CI users. The noise tracking algorithm was able to catch detailed spectral information of the noise with a fast noise tracker during the noise-like frames and update the estimated accumulative noise level with a slow noise tracker during speech-like frames. Next, this noise tracking algorithm was used to estimate the signal-to-noise ratio (SNR) of each temporal-spectral region, termed “time-frequency unit” in the BM method, to determine whether to eliminate or retain each unit. Finally, a sentence perception test was employed to investigate the effects of this noise reduction algorithm in various types of background noise and input SNR conditions. Results showed that the mean percent correct for CI users is improved in most conditions by the noise reduction process. Improvements in speech intelligibility were observed at all input SNR conditions for the babble and speech-shaped noise conditions; however, challenges still remain for the non-stationary restaurant noise.