Perceptual interactions among components of a spectral-domain voice source model

Abstract

A psychoacoustic model of the source spectrum has been proposed in which source contributions to overall voice quality can be quantified by four spectral slope components: H1-H2 (the amplitude difference between the first and second harmonics), H2-H4, H4-2000 Hz (i.e., the harmonic nearest to 2000 Hz), and 2000–5000 Hz. The natural variability of these components has been described, along with the just noticeable differences (JNDs) for each component. The goals of this study are to identify how perceptual sensitivity to each component slope varies as a function of the adjacent slope(s). The JNDs were obtained for stimuli based on synthetic copies of one female voice. The stimuli were manipulated so that spectral slope varied in 0.5 dB increments (1 dB for 2000–5000 Hz) for a particular component at “high” and “low” values of the adjacent slopes. Thirty-three listeners completed an adaptive up-down paradigm. Preliminary results suggest that sensitivity to a particular spectral component varies as a function of some adjacent components more than the other. Furthermore, sensitivity varies based on whether the adjacent component has a high or low slope. These interactions among spectral components will be interpreted with respect to the variability in spectral configuration observed in 144 voices.