The Digital Loudspeaker Array (DLA) is an electroacoustic transducer which receives as input a digital signal and performs the analog conversion directly into the air. It consists of a plurality of radiating elements arranged in a matrix. These elements will be designated by the term “speaklet” when they are reduced in size. The acoustic radiation of a DLA is indeed very sensitive to the size of the matrix due to differences in path length, which makes it especially suitable for MEMS technology. This thesis is on the study of a piezoelectric MEMS DLA. After an introduction that is increasingly focused on the subject, the thesis addresses the multiphysics modeling of the DLA, dimensioning of the speaklets and experimental tests. Analytical formulas, numerical simulations and finite element models are developed and used to predict the mechanical behavior of the presented speaklets, the pressure radiated by the DLA and the electrical power consumption. The speaklet are then dimensioned from the technological stack (set in advance) in order to maximize the pressure level. Experimental tests involving the use of an anechoic chamber, an optical interferometer, a vibrometer and an impedancemeter validate most of the models. Otherwise, these tests are usefull for improving some of them or for showing their limitations. The results have shown the importance of the residual stresses, which cause an initial deformation of the speaklets and modify their resonance frequencies, thus rendering ineffective the use of large radii. In accordance with the models, the static deflection of the speaklets is nonlinear but their dynamic behavior is linear. This enables characterizations using transfer functions. Theory and sound recordings show that a DLA made of such speaklets can produce in the best case the same pressure to that generated by the same matrix driven in an analog way. In our case, more distortions were obtained in digital reconstructions because of non-uniform responses of the speaklets, due to different access resistances. However, the presented DLA has other advantages, the most important being the very low power consumption it is theoretically possible to achieve using the adiabatic charge principle. The piezoelectric MEMS DLA thus appears as a promising technology. The optimization of our first prototype using the developed tools should indeed lead to a DLA able to generate an equivalent presure to that obtained with analog control, but with a far greater electroacoustic efficiency. Future work should then focus on the design of nonlinear speaklets and on the shaping of the pulse of pressure they generate, in order to increase the total pressure level.