The method of complex networks has been proposed as a novel approach to analyze time series from a new perspective. However, only few studies have applied this methodology to certain types of pseudo-periodic signals. In this article, the network-based technique is applied on voice signals, a kind of pseudo-periodic signals which has not been analyzed using complex networks, to differentiate between a healthy subject and subjects with pathological disorders. The results obtained demonstrated that through a set of statistic computed from the complex networks is possible to differentiate between healthy and non-healthy subjects, contrary to what was observed using well known non-linear statistics, such as Lempel-Ziv complexity and sample entropy. We conclude that by seeing voice signals as complex networks new information can be extracted from the time series that may help in the diagnosis of pathologies.