Evaluation of various sets of acoustic cues for the perception of prevocalic stop consonants. II. Modeling and evaluation

The purpose of the study presented in this paper and the accompanying paper [Smits et al., J. Acoust. Soc. Am. 100, 3852–3864 (1996)] is to evaluate whether detailed or gross time-frequency structures are more relevant for the perception of prevocalic stop consonants. To this end, first a perception experiment was carried out with ``burst-spliced'' stop-vowel utterances. This experiment is described in the accompanying paper. The present paper describes the second part of the investigation, i.e., the simulation of the behavior of the listeners in the perception experiment. First, a number of detailed and gross cues are measured on the stimuli. Next, these cues are mapped onto the observed perceptual data using a formal model of human classification behavior. The results show that in all cases the detailed cues, such as formant transitions, give a better account of the perceptual data than the gross cues, such as the global spectral tilt and its initial change. The best-performing models are interpreted in terms of the acoustic boundaries which are associated with the perceived linguistic contrasts. These boundaries are highly interpretable linear functions of five or six acoustic cues, which give a quantitative description of the often-discussed ``trade-off'' relation between the various cues for perception of place of articulation in stop consonant