High-resolution spherical directivity of live speech from a multiple-capture transfer function method

Abstract

Although human speech radiation has been a subject of considerable interest for decades, researchers have not previously measured its directivity over a complete sphere with high spatial and spectral resolution using live phonetically balanced passages. The research reported in this paper addresses this deficiency by employing a multiple-capture transfer function technique and spherical harmonic expansions. The work involved eight subjects and 2522 unique sampling positions over a 1.22 or 1.83 m sphere with 5° polar and azimuthal-angle increments. The paper explains the methods and directs readers to archived results for further exploration, modeling, and speech simulation in acoustical environments. Comparisons of the results to those of a KEMAR head-and-torso simulator, lower-resolution single-capture measurements, other authors' work, and basic symmetry expectations all substantiate their validity. The completeness and high resolution of the measurements offer insights into spherical speech directivity patterns that will aid researchers in the speech sciences, architectural acoustics, audio, and communications.

Near-Field Evaluation of Reproducible Speech Sources

“…Though some evidence indicates that directivity of fricative sounds can be slightly asymmetric [13], the directivity of speech was assumed to be symmetric along the forwardlooking direction [5,14], and therefore speech was captured for a semicircle in the horizontal plane. On the ring, nine DPA 4060 miniature omnidirectional microphones were arranged from a 0°direction, in 22.5°increments, until 180°as shown in Fig.…”

“…The development of reproducible speech devices is intrinsically connected to the initial studies on the sound field of speech around the human head, or speech directivity [7,8]. A considerable amount of research has been published relating to speech directivity including large directivity samples [9], directivity while singing [10,11,5], the directivity of low and high levels of speech [11], the directional characteristics of specific phonemes [12,13], as well as modern methods to capture [14,5] and spatially up-sample [15] speech directivity.…”

“…However, it only investigated the frontal direction and not directivity. This paper does not focus on the performance of the mouth simulator for the purposes of speech intelligibility, as mouth simulators have already been proven to work adequately for these kinds of tasks [14].…”

“…Methods for acoustic holography [25] used for capturing, analyzing, and emulating the radiation properties of sources [26] have been applied to the directivity of speech using arrays of microphones [13,15,14]. By decomposing the discrete pressure surface emanating from a speaker using a set of orthonormal basis of functions, the continuous definition of speech directivity patterns can be reconstructed up to some spatial aliasing limit.…”

Combining multiple sparse measurements with reference data regularization to create spherical directivities applied to voice data

“…Over the decades, researchers have employed numerous measurement techniques to obtain increasingly finer resolutions approaching compatibility with standardized resolutions for loudspeakers [3] that are commonly employed in architectural acoustics simulation packages [4,5]. Although initial efforts often employed a single moving microphone [6,7], most modern techniques make use of rotations of microphone arrays or rotations of subjects within microphone arrays to leverage higher-resolution, spherical or partial-spherical results [8][9][10][11][12].…”