Sampling timed words out of a timed language described as a timed automaton may seem a simple task: start from the initial state, choose a transition and a delay and repeat until an accepting state is reached. Unfortunately, simple approach based on local, on-the-fly rules produces timed words from distributions that are biased in some unpredictable ways. For this reason, approaches have been developed to guarantee that the sampling follows a more desirable distribution defined over the timed language and not over the automaton. One such distribution is the maximal entropy distribution, whose implementation requires several non-trivial computational steps. In this paper, we present Wordgen which combines those different necessary steps into a lightweight standalone tool. The resulting timed words can be mapped to signals used for model-based testing and falsification of cyber-physical systems thanks to a simple interface with the Breach tool.