This paper deals with voice communication models for disaster area scenarios. The goal is to design models that can be used to generate realistic push to talk traffic for single talk groups. The modelling is based on an analysis of empirical measurements during a catastrophe maneuver. The analysis shows that the time series comprise heavy load periods and significant correlations. Based on these characteristics, different Markov and Semi-Markov models are considered. Synthetic traffic streams for the different models are generated and evaluated by visual and statistical analysis. Finally, a case study outlines the impact of the different traffic models in network performance simulation.