Modeling the neural encoding of vowel formants in the midbrain

The first (F1) and second (F2) formants of a vowel determine its identity, making it essential to understand how these formant regions are encoded in the brain. We investigated the neural encoding of vowel formants using computational model responses of the auditory nerve and midbrain to natural vowel tokens Hillenbrand et al., JASA 97, 3099 (1995). The first formant peak can be difficult to identify based on model population responses. Responses in the F1 region are more affected by the surrounding harmonic structure, whereas F2 often has a more straightforward representation in the model response. This result is surprising given the sensitivity of listeners for F1 discrimination compared to F2, reflecting Vowel Dispersion Theory Liljencrants & Lindblom, Language 48, 839 (1972), which favors the F1 dimension over F2. While peaks are difficult to identify, complex patterns of neural discharge rates observed across low-frequency midbrain responses may work to encode F1. These results suggest revisiting assumptions of the neural representations of formants and highlight the need for a system that evaluates overall response trends to formant regions instead of searching for formant peaks. Neural models can play an important role in understanding the structure of vowel systems. Work supported by NIDCD-R01-010813.