The challenge of understanding speech in noise, especially for listeners with sensorineural hearing loss (SNHL), cannot be explained by conventional models of neural coding. These models are typically based on average discharge rates and/or phase-locking to temporal fine structure of auditory-nerve (AN) fibers. Because these response properties are only weakly affected by mild to moderate SNHL, such models fail to explain the significant difficulties caused by relatively small amounts of hearing loss. Our recent studies of responses in the midbrain of rabbit and budgerigar have directed our attention to response features that apparently solve this problem. Midbrain neurons are well known to be strongly driven (or suppressed) by low-frequency amplitude modulations (AM). We have recently observed that selectivity of midbrain neurons to the direction of frequency chirps can match or exceed their AM sensitivity. Changes in amplitudes and frequencies of neural responses across peripheral frequency channels create strong fluctuation contrasts that encode the location of spectral peaks (formants) in voiced speech. We hypothesize that midbrain sensitivity to neural amplitude and frequency fluctuations in peripheral responses provides a robust representation of complex sounds, including speech.
Aim 1 tests this hypothesis with physiological and behavioral studies of midbrain responses to stimuli that combine these cues, including ?designer? stimuli with conflicting cues to determine how they may interact. These results will be used to test and refine a new computational model for midbrain responses with sensitivity to these cues.
Aim 2 tests the hypothesis with physiological responses of midbrain neurons to voiced speech, to directly test model predictions based on characterization of each neuron's sensitivity to these cues. Understanding the role in speech coding of the amplitude and frequency fluctuations in peripheral responses is clinically significant because these fluctuations are vulnerable to SNHL.
In Aim 3, we will test the hypothesis that amplitude and frequency fluctuations can be manipulated in synthetic speech to influence intelligibility in human listeners with or without SNHL, in quiet and in noise. In the healthy auditory system, fluctuation contrasts are mapped into a robust midbrain response, where rate and timing cues (i.e., phase-locking to the F0-related fluctuations) persist across a wide range of sound levels and in noise. Because fluctuation contrasts depend upon saturation of inner hair cells, which is in turn influenced by cochlear amplification, pathological reduction of either inner or outer hair cell sensitivity reduces the fluctuation contrasts in the periphery and ultimately in the midbrain. Model-based algorithms will manipulate the fluctuation contrasts to improve (or degrade) intelligibility in listeners. Preliminary results from modeling, physiological, and behavioral studies support the proposed hypotheses. Ultimately, our goal is to extend this approach to manipulate fluctuation contrasts in running speech, to effectively ?correct? sound for the impaired ear.
The public-health significance of the proposed work is that it will improve our understanding of speech coding by neurons in the auditory system. Using behavioral and physiological techniques, as well as computational models, we will test a novel hypothesis for robust neural coding of speech in quiet and noisy conditions. We will test the impact of hearing loss on this code and test a strategy for enhancing speech. This work will lead to novel approaches to enhance speech sounds using assistive devices such as hearing aids.
Showing the most recent 10 out of 36 publications
