Locating fricative categories in a synthetic acoustic space: Spectral moments measures of a perceptual English fricative system

There are several approaches to measuring spectral differences between fricatives, among which spectral moments analysis [Forrest et al., JASA 84, 115–123 (1988)] is the most common. To perceptually evaluate measurement systems, stimuli are constructed with four Gaussian noise components whose base frequency is systematically varied. Then, to vary the spectral shape, the amplitude of each component is independently varied to yield a five-dimensional space of synthetic fricatives. The stimuli were reduced to four dimensions and presented to listeners in a self-directed search protocol, in which participants located examples of /f, θ, s, ʃ, h/. Spectral moments were calculated from the stimuli, showing a complex relationship between the synthesis dimensions and spectral moments. This paper presents high-probability locations of these fricatives, regressed against the synthetic parameters and their associated spectral moments. Preliminary results indicate that acoustic measures of the chosen stimuli conform to differences found in previous fricative production studies, but also that the spectral moments measures do not outperform the synthetic parameters, and hence do not seem to be adding insight into which acoustic differences are guiding the listeners.