题目：Speech Responses in the Auditory Midbrain: A Novel Hypothesis for Vowel Coding

报告人：Laurel H. Carney ( University of Rochester ; Departments of Biomedical ; Engineering and Neurobiology & Anatomy )

时间：2013年12月03日 星期二 13:00-14:00

地点： 生命科学学院 邓佑才报告厅（101房间）

邀请人：李量

摘要：Models for neural coding of speech sounds tend to focus on the representation of the spectrum in discharge patterns of the auditory periphery or lower levels of the ascending pathway. Here we show that at the level of the midbrain, rate-based tuning for amplitude-modulation plays a role not only in the representation of temporal envelopes, but also in robust coding of voiced speech sounds.  Amplitude-modulation (AM) frequency tuning first appears in ascending auditory pathway at the level of the auditory midbrain (inferior colliculus, IC).  Approximately 50% of IC neurons have band-pass modulation transfer functions (MTFs), and the other 50% have low-pass, band-reject or high-pass MTFs. A computational model derived from the same-frequency inhibitory-excitatory model (Nelson and Carney, JASA 2004, 116:2173) is able to explain all of these MTF types. Bandpass AM tuning in the original model is derived from the dynamics of excitatory and inhibitory inputs to IC neurons, with excitation arising from the auditory brainstem (i.e. cochlear nucleus or superior olive) and similar inhibitory inputs relayed by interneurons. In the extended model presented here, the other MTF types are created by combining the brainstem excitatory inputs and inhibition from the bandpass MTFs. The best modulation frequency of the bandpass inhibition determines the shape of the other MTFs. A population model based on these model neurons provides robust representations of a wide range of vowels across a large range of sound levels and in background noise. Strategies for formant detection and speech enhancement based on this model will also be described.