Communication is multi-modal - when we interact, we speak, gesticulate, and touch. However, the neural computations and circuits that choose these different communication signals are unclear. We address this issue in Drosophila which thanks to its complex social behavior and genetic toolbox is ideal for dissecting the neural basis of communication. During courtship, male flies produce two types of communication signals to woo the female: air-borne song is produced by wing movement, while substrate-borne vibrations are transmitted via the legs. Using statistical modelling of pose tracking data alongside recordings of the song and vibration during courtship, we identified the behavioral contexts in which each signal is produced. We found that female locomotion controls the choice between song and vibration: When the female moves, the male sings, when she is stationary, he vibrates. Optogenetic control of female locomotion confirms the model results. Combining optogenetics and network models, we elucidate the circuit mechanisms underlying this choice. Two groups of sexually dimorphic neurons, “P1a“ and “pC2”, drive both signals with complex dynamics: While optogenetic activation of P1a in a solitary males primarily drives persistent vibrations, pC2 induces a sequence of first song and then vibrations. A circuit model reveals that these complex dynamics emerge from a combination of recurrence and mutual inhibition. Lastly, we show how sensory cues modify the intrinsic dynamics to support context-dependent signaling: Female movement affects male signal choice not via visual motion cues but by changing the males locomotor state, which we demonstrate by controlling the male movement and visual inputs during optogenetic stimulation. Overall, by combining behavioral and neural circuit modelling, we show how intrinsic circuit dynamics are modified by sensory cues to produce context-dependent signals during social interactions.