Skip to main content
Open Access Publications from the University of California

UC Berkeley

UC Berkeley Electronic Theses and Dissertations bannerUC Berkeley

Predictive models of auditory perception in human electrophysiology


It has long been thought that sensory systems operate by representing information in a hierarchy of sensory features, and that these features build upon one another. From low-level information such as spectral content, to high-level information such as word content, the sensory system must rapidly extract all of these features from the world. However, the precise nature of these levels of representation, as well as how they interact with one another, is not well-understood. In audition, intermediate sensory representations are often studied in animals, using techniques that treat neurons as a linear filter for incoming sensory inputs. If those inputs are spectro-temporal features (e.g., a spectrogram), then the result is a Spectro Temporal Receptive Field (STRF). This describes how the neural unit in question (e.g., a neuron) will respond to patterns in spectro-temporal space. It has been a crucial tool in understanding sensory processing in low-level neural activity. Using this approach it is also possible to study how this neural representation changes under different experimental conditions. STRF plasticity has been shown in both reward- and context-modulated experiments in animals.

In recent years, it has been suggested that similar techniques may work in modeling the activity of neural signals recorded from humans. As we cannot generally record from single unit activity in humans, this approach relies on proxies for neural activity – specifically in the high-frequency activity (HFA) of electrocorticography electrodes. This poses a unique opportunity for two reasons: First, human language is a natural stimulus set for studying hierarchical feature representations in the brain. There are many ways to decompose speech into both auditory and linguistic components, and each of these could serve as inputs to the modeling technique described above. Second, humans are especially skilled at using high-level context such as their experience and assumptions about the world in order to change their behavior. This poses a unique opportunity to study the plasticity of speech representations in the brain.

This thesis reports several new approaches towards studying the sensory representation of speech in the human brain, as well as how these representations may change due to experience. It aims to bridge the literature in rodents and songbirds with ideas in human electrophysiology in order to pursue new approaches to studying perception in humans.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View