Dielectric elastomers (DE) are regarded as a potential alternative to conventional actuator technologies. They feature low weight, high strains and low material costs. Their scope of application ranges from sensors, energy generators, smart textiles to biomimetic robots and much more. A few concepts of loudspeakers using DE have been demonstrated by the research community. One of the disadvantages of previously concepts was the need for mechanical bias (e.g. by air pressure). This work proposes a new concept of loudspeaker, which does not need prestretch or other means of mechanical bias. Buckling dielectric elastomer transducers (BDET) use the area expansion of actuated DE to buckle up. This mechanism is used to construct a millimeter-scale loudspeaker with good frequency response in the upper frequency range. The concept is implemented using automatically fabricated multi-layer membranes. The multilayer structure allows to generate more force and has higher flexural rigidity than a single-layer setup. Samples with different amount of layers are fabricated and an analytical model is derived. Measurements of the static deflection, the frequency response and the total harmonic distortion validate the model. The small scale of the speaker allows it to be installed in large arrays and thus might offer a hardware platform for high-resolution beam forming or wave field synthesis.