Speech Encoding in the Human Auditory Periphery: Modeling and Quantitative Assessment by Means of Automatic Speech Recognition

摘要

The human ability to understand spoken language is remarkable. The response of the auditory nerve contains all the acoustic information that is available to make this feat possible. The work presented in this thesis aims at quantitatively evaluating how speech is encoded in the spike trains of the auditory system. In the first part I develop a model of the human inner ear up to the level of the auditory nerve. The study also includes a specialized neuron type found in the cochlear nucleus - the octopus cells. In the second part, I evaluate speech encoding by combining the models I developed with automatic speech recognition. I conclude that the human auditory system does not rely on a rate-place code alone but requires the abundance of fibers for precise temporal coding. A hypothesized mechanism utilizing interspike interval information improves the noise and level robustness. Further, I show that similar improvements are achieved by including cochlear nucleus neurons that inherently extract speech-relevant temporal information.