Macroscopic versus microscopic spectral weighting of speech stimuli

A spectral weighting experiment is presented, designed to assess whether listeners adopt different listening weights for consonant sounds depending on the sound presented and whether weights depend on the level of analysis -- for individual stimuli (“microscopic”) level or for the stimulus set (“macroscopic”). Three stimuli, modeled after the voiceless fricatives /f/, /s/, and /∫/, were used. Subjects heard one of the three stimuli, each containing random level perturbations across multiple frequency bands. They then made an identification decision. For each stimulus, the weighting strategy was estimated by correlating observers’ responses with the random perturbations. This procedure allows for the estimation of listeners’ strategies applied to each stimulus in the set. This method differs from other speech weighting studies, which assume that listeners apply a general filter to speech stimuli, enhancing some bands and attenuating others (e.g., the Articulation Index). Weighting functions for individual stimuli are compared to those using a global strategy to determine the importance of analyzing decisions on an individual-stimulus basis. Results indicate that the listening strategy depends greatly on the stimulus presented, and that the flexibility of the auditory system in adapting to changing stimuli can be assessed using these methods.