Active automata learning (AAL) algorithms infer accurate automata models of black box applications, letting developers verify the behavior of increasingly complex real-time systems (RTS). However, learning models of larger RTS often takes very long or is not feasible at all. We introduce Mealy machines with local timers, a new class of Mealy machines that permit multiple location-bound timers and that can be learned efficiently. We design an efficient learning algorithm for them and validate our method across diverse case studies ranging from automotive systems to smart home appliances, where we drastically reduce runtimes compared to the state-of-the-art approach, thus, making AAL available for a wide range of RTS.