Cochlear implants (CIs) are neuroprosthetic devices that are surgically implanted to restore functional hearing in deaf and hard-of-hearing individuals. Most CI users can understand speech well in quiet situations, yet, it becomes quite challenging for them to understand speech in crowded environments, especially when multiple people are speaking simultaneously. This dissertation investigated whether such difficulties are related to the poor representation of voice cues in the implant arising from degraded spectral and temporal resolution from signal processing strategies. Human voices are characterized by their pitch (F0), in addition to a second dimension called the vocal-tract length (VTL). This dimension directly scales with the size of the speaker and, therefore, plays a crucial role in the distinction between male and female talkers, or between adults and children. The research questions were: whether CI users’ speech intelligibility in the presence of a competing talker (speech-on-speech; SoS) is related to their sensitivity to the F0 and VTL differences between the speakers, whether this relationship is influenced by the spectral resolution in the implant, and whether optimizing signal processing algorithms could improve the perception of such voice cues. The data showed that CI users’ SoS intelligibility was related to how sensitive they were to both F0 and VTL differences, and that this relationship was influenced by the spectral resolution in the implant. The data also provided evidence that CI users can draw a benefit from voice differences between male and female speakers, but not between female speakers and children. In addition, spectral enhancement techniques and optimization of some implant parameters were both shown to contribute to an improvement in SoS intelligibility and VTL sensitivity, respectively. These findings lay the foundations for future optimizations of the implant to improve CI users’ speech intelligibility in noisy settings.