Analysis of phoneme-dependent variations in the acoustic radiation characteristics of human speech

When speech is reproduced through a loudspeaker—even with its frequency response compensated—listeners are still able to distinguish it from speech spoken directly by a human. The physical factors that enable this auditory distinction have not yet been clearly identified. In this study, we hypothesize that one key factor is the variation in acoustic radiation patterns of human speech across different phonemes. If this hypothesis holds, then the transmission path from the loudspeaker to the listener's ear remains constant, regardless of the speech content. In contrast, in natural human speech, the sound reaching the ear varies depending on the phoneme due to its directional radiation characteristics. These differences may partly explain the perceptual realism gap between reproduced and natural speech. To investigate this, we recorded human speech using a microphone array arranged omnidirectionally around the speaker’s head, capturing the spatial radiation characteristics for each phoneme. These measurements enabled an analysis of phoneme-dependent directional variation in speech radiation, providing insights into potential factors underlying perceptual differences between loudspeaker-reproduced and naturally spoken utterances. Our ultimate goal is to apply these insights to compensate for radiation mismatches and enhance the realism of synthetic speech playback through loudspeakers.