A Database for the Comparison of Measured Datasets of Human Voice Directivity

More than 200 years ago, early research on human voice radiation showed that speech has directional properties (Saunders, 1790; Wyatt, 1813). Since then, many speech directivity studies have elaborated on several specific aspects, for example, phoneme or loudness dependencies. However, up to now, only a few datasets are freely available. A database is presented that allows direct comparison and visualization of datasets from more than 15 different studies. The data is collected from tables, plots, and datasets from the supplemental material of the respective studies. Some studies present directivity patterns averaged over a whole sentence, while others report phoneme-dependent data. Furthermore, these datasets vary in their sampling grids, with many measured in the horizontal plane and just a few measured spherically. Furthermore, most datasets included in this work present frequency-band averaged values, for example, in one-third-octave bands, while a few newer studies provide the raw data in the form of transfer functions. Our database allows the visualization and comparison of directivity patterns of various datasets found in literature in different ways, such as polar plots or amplitudes over frequency. This study can thus simplify the systematic review and support a better understanding of the specific properties of human voice directivity.