In recent works, we proved that some dielectric elastomer (DE) actuator topologies, normally used as linear actuators at low frequencies (LFs), can produce sound taking advantage of high-frequency (HF) structural vibrations of the DE membrane surface. Because structural vibrations take place along different deformation directions compared to those involved in LF actuation, these DE actuators (DEAs) can generate sound even when their LF pumping motion is constrained, or while they are driven to produce a concurrent LF pumping motion. This observation can be used to develop acoustic buttons, which produce sound while being deformed by a user, or multi-function audio-tactile interfaces, which provide combined HF acoustic and LF tactile feedbacks through multi-chromatic voltage inputs. In this paper, we propose a self-sensing approach to estimate LF deformations of multi-function DEAs. In contrast with traditional self-sensing approaches, in which an additional sensing signal is superimposed to the main driving signal, here we solely rely on the HF acoustic voltage input, which we also use as the sensing signal. We prove that self-sensing of LF deformations can be achieved even in the presence of complex HF driving signals, such as soundtracks. This allows reconstructing LF deformations induced either by a LF voltage excitation (superimposed to the driving acoustic signal), or by variable external forces (e.g., user touches, such as in user interfaces). In the future, this self-sensing approach might be used to build multi-functional sound interfaces that adapt their output based on a user-driven deformation, or for virtual reality rendering applications.