Spatial characteristics of the human voice: Directivity measurements, simulations, and perception.

• Applicant: Dr. Paul Luizard
• Subject Area: Acoustics
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 457863315

The human voice, as one of the basic elements of communication between individuals, has been studied for hundreds of years and certain aspects are well understood. Although significant effort was produced on dimensions such as sound level, rhythm, and timbre, the spatial aspects have received little attention. However, recent research has showed that the radiation of voice is a fundamental characteristic of the various phonemes that a talker can produce, making it a key feature of both speaking and singing voice.This project aims at obtaining a global understanding of voice directivity by combining together measurements, modelling and listening tests based on a large sample of phonemes, sentences, and musical pieces. The measurements will outperform the spatial resolution achieved by previous studies so far by using a large amount of MEMS microphones mounted on a circle surrounding the talker. This circular structure could be rotated to yield measurements over a hole sphere.To overcome the problem of repeatability of human talkers, an artificial head embodying a loudspeaker in the mouth with replaceable mouth shapes designed by 3D scanning and printing will yield repeatable phoneme productions to be measured along a large number of directions in space. These results will be compared to analytical solutions of a variable surface radiating sound, standing for the various open mouth sizes, on a sphere representing a simplified head. Simulations involving complex meshes representing the geometries of the artificial talking head, while pronouncing various phonemes, will estimate the sound field radiated in space by means of the boundary element method (BEM).While these measurements and simulations are intended to yield valuable fundamental knowledge on speech and singing radiation, the perception of these sounds in various spaces will also be investigated.Two listening tests will be conducted within virtual sound fields based on the previously obtained data of voice directivity. The first one focuses on the just noticeable difference (JND) of head orientation in the horizontal plane in both anechoic and slightly reverberant conditions using the ABX test framework. The second experiment aims at estimating the optimal spatial resolution in terms of order of spherical harmonics for voice rendering. A comparison task between a very high resolution and versions of the same audio scene with decreasing sound source resolution will yield the lowest order of spherical harmonics that would render a scene perceived as good as the reference. These results related to balance between sound quality of voice rendering and computation cost will strongly impact the virtual acoustics community.

• DFG Programme: Research Grants
• International Connection: France, Japan
• Cooperation Partners: Brian Katz, Ph.D.; Dr. Makoto Otani; Professor Dr. Stefan Weinzierl