Analyzing the Directivity Patterns of Human Speakers

The directivity pattern of the human voice plays an important role in the realization of communication scenarios in virtual acoustic environments. In contrast to many other sound sources, a specific characteristic of the human voice is its dynamic directivity, i.e. time-dependent changes in articulation or singing. As it is nearly impossible for subjects to perform exactly the same articulations for numerous measurements, capturing the dynamic directivity with sequential measurements is not feasible. Thus, the directivity must be measured simultaneously for all directions, e.g. by using a surrounding spherical microphone array. Generally, such microphone arrays only have a limited number of microphones and the spatial resolution of the captured directivity patterns is low. To obtain a high-density directivity pattern, methods for spatial upsampling are required. In this context, we recently introduced the so-called SUpDEq (Spatial Upsampling by Directional Equalization) method, which makes it possible to generate high-resolution voice directivities from sparse measurements. In this study, we now apply the method previously already evaluated for a head and mouth simulator to human speaker directivities. Based on individual measurements, we examine the spherical directivity patterns regarding their interindividual differences and analyze the articulation-dependent influences for different vowels and fricatives.