This paper aims to establish the pivotal role of the linear non-modal dynamics of perturbations in the aerodynamic sound emission of turbulent temporal shear/mixing layers at the self-similar stage of evolution. This is achieved by comparing the results of direct numerical simulations (DNS) of the mixing layer to the linear non-modal analysis of the flow central/body part model – 3D homentropic, unbounded flow, U0(Ay, 0, 0), with shear rate, A. The non-modal analysis captures the only linear mechanism of acoustic wave generation – the linear vortex-wave mode coupling induced by the non-modal dynamics of perturbations in shear flows. The efficiency of this linear generation is uniquely determined by the mode-dependent Mach number,M= Aλx/cs, related to the convective Mach number viaM= Mcλx/πΔy. Here, λx, cs and Δy denote the streamwise wavelength, speed of sound and shear-layer width, respectively. DNS of compressible turbulent temporal mixing layers were performed for different convective Mach numbers (Mc = 0.3, 0.7, 1.4) and simulation boxes (Lx, Ly, Lz) for reliably identifying the essence of the origin of sound in the turbulent layer. The simulations clearly show the decisive role of the linear non-modal mechanism of sound generation in the flow. This novelty highlights the significance of M in the sound generation, thus ensuring maximum efficiency for harmonics with largest streamwise wavelength, λx = Lx. Accordingly, the aerodynamic sound efficiency is determined by McLx rather than Mc. Overall, our analysis points to the inadequacy of employing the modal approach to describe compressible dynamic processes in the turbulent flow.