Dynamic trajectories in the vocal space : mapping basic acoustic features of the voice with listeners' subjective perception in a public figure identification tasók

Abstract

Using the source-filter model of voice production, previous research has utilized multidimensional maps to quantify differences between speaker voices and explain them in relation to various perceptual dimensions. These approaches may be extended by integrating dynamic aspects of the voice physical signal. It is unclear whether the informational similarity between different voices’ acoustic distributions leads to subjective perception of such signals as being from the same person. In this work, we examine a method for representing voices in their important acoustic dynamic dimensions and evaluate these representations in terms of subjective judgment. A database of speech stimuli from famous public figures (targets) was created, along with a collection of verbatim imitations produced by professional impersonators. Imitation pair sets were generated before and after listening to the reference target stimulus. We quantified acoustic similarities between targets and all imitations using two-dimensional probability maps of joint pitch and the average inter-formant distance dimensions, which represent the source and filter stages of vocal production, respectively. Subjective identification data were collected in a large-scale online experiment, where participants evaluated the correspondence of each voice stimulus with the target person. The results showed that the degree of acoustic similarity an imitation has with the original target voice is related to the likelihood of identifying such imitation as a genuine voice. This demonstrates that acoustic similarity mappings are relevant for subjective voice perception and provide a valid way to represent voices. This may open new research directions in voice perception by integrating the dynamic trajectories of the voice’s spectral features during production.