Computational model predictions of level dependent changes in vowel identification with addition of rate-place cue

Abstract

A signal processing model based on temporal cues of auditory-nerve fibers had been developed to understand the level dependent changes in vowel identification scores. To this study, the rate-place cues of auditory-nerve fibers were added to the existing temporal model of vowel identification. The model includes the human version of the auditory-nerve model, with added rate-place cues, along with the neural network to identify vowels. The model predictions of vowel identification across levels with only temporal cues are compared with the model predictions with both temporal and rate-place cues of auditory-nerve fibers. This paper also analyses the vowel identification scores from the perspective of auditory-nerves corresponding to first and second formants (F1 and F2) besides the entire spectrum of auditory-nerve fibers. The model prediction revealed that the representation of second formant (F2) was improved with added rate-place cues especially at low-to-mid levels and could be associated with lower acoustic energy of F2. Thus, this paper possibly explains the role of rate-places cues for vowel identification scores across levels.