Computational model predictions of cues for concurrent vowel identification

Area of Science:

• Auditory perception
• Computational neuroscience
• Speech processing

Background:

• Vowel segregation and identification rely on fundamental frequencies (F0s) and formant differences.
• Listeners can identify concurrent vowels even with identical F0s, indicating the importance of other cues.

Purpose of the Study:

• To investigate the roles of F0 and formant differences in concurrent vowel identification.
• To explore how sound levels affect the contribution of these cues.
• To model the neural mechanisms underlying these effects using an auditory-nerve model.

Main Methods:

• Perceptual study with young, normal-hearing adults identifying concurrent vowels at various sound levels (25-85 dB SPL).
• Computational auditory-nerve modeling to simulate neural responses and estimate acoustic cues.
• Analysis of phase locking to F0s and formants using template contrast and synchronized rate.

Main Results:

• Vowel identification scores were lowest at the extreme sound levels (lowest and highest).
• The benefit of different F0s was reduced at the lowest sound level.
• Computational models predicted level-dependent changes in neural phase locking to F0s and formants.

Conclusions:

• Concurrent vowel identification is influenced by sound level, affecting neural processing.
• Poorer identification at low levels relates to reduced phase locking to F0s.
• At high levels, poorer identification may stem from altered formant processing due to synchrony capture.