Sensory feedback plays a crucial role in speech production in both healthy individuals and in individuals with production-limited speech. However, the vast majority of research on the sensory consequences of speech production has focused on auditory feedback while relatively little is known about the role of vocal tract somatosensory feedback. The body of this dissertation investigates speech and nonspeech production in the absence of vocal tract somatosensory feedback by training subjects to use a touchscreen-based speech production platform. Contact with the touchscreen results in instant playback of a vowel or complex tone dependent on the location selected. Because the axes of the touchscreen are associated with continuous F2 and F1 frequencies, every possible vowel within a wide formant range can be produced. Participants with no initial knowledge of the mapping of screen areas to playback sounds were asked to reproduce auditory vowel or complex tonal targets. Their responses were evaluated for accuracy and consistency, and in some cases participants underwent functional neuroimaging via MEG during training. Following training, participants were capable of using the touchscreen to produce speech and nonspeech sounds in the absence of the vocal tract. Their increased accuracy and consistency as they learned to produce speech and nonspeech sounds indicates the development of new audiomotor maps, as does significant changes in their task-based functional neuroimaging over the course of training. While participants demonstrated learning in both speech and nonspeech production, the neural and behavioral differences indicated different learning processes. We hypothesize that these differences can at least partially be attributed to the presence of an existing audiomotor network for producing speech sounds. This would account for more rapid learning rates in the speech variants of the task, the presence of generalization in touchscreen-based speech production, and the neural similarities to vocal speech production in the touchscreen-produced speech sounds that presented differently in touchscreen-produced nonspeech sounds.

Introduction: MEG recordings are prone to interferences. Lots of efforts have been made to deal with interference rejection in MEG data, but still there are some kinds of interferences could not be handled well so far, e.g. large amplitude interferences like those caused by vagus nerve stimulators (VNS) implant in epilepsy patients, and dealing with these interferences well are of clinical importance for these patients. In this study, we illustrated a novel MEG interference rejection algorithm called dual signal subspace projection (DSSP) that could potentially help deal with such interferences, and evaluated its performance through computer simulation data and clinical data.

Method: Computer simulation is performed with several types of interferences varying in localization and signal type, and DSPP-processed data was compared with the true signal for performance evaluation. The performance of DSSP in the clinical data was mainly tested through a retrospective cohort study from epilepsy patients with VNS implant receiving MEG for interictal spike mapping. The evaluation was based on direct MEG recording reading for epileptic spike identification, successful spike localization using the clinical standard dipole fitting model, and spike localization and activity time-series recovery using a more reliable MEG source reconstruction algorithms called Champagne.

Results: DSSP handled various kinds of interference setting in computer simulation well, but also showed preference towards periodic interference outside region of interest. Direct MEG recording reading showed that MEG recordings became more readable after DSSP processing and more epileptic spikes could be identified. More spike could be localized using the dipole fitting method with DSSP-processed data, and localization results differed from that achieved with DSSP- unprocessed data. Evaluation with Champagne algorithm showed that DSSP-processed data had a higher chance to achieve successful spike localization with reasonable location and meaningful recovered activity time-series data. The localization result achieved with Champagne algorithm also differed from the previous spike mapping result using the DSSP-unprocessed data.

Conclusion: DSSP is a valuable novel interference rejection algorithm to be explored. It potentially works better with interference periodic and located outside the region of interest, and could help to recover the distorted MEG recordings from epilepsy patient with VNS impact.

When we speak, there’s a complex sequence of events occurring within us. A symphony, if you will, with the brain as the conductor; the respiratory muscles pushing air out of our bellow-like lungs; this exhaled air setting the vocal cords into vibration like the strings of a musical instrument; the tongue, lips, nose, teeth and jaw acting as an ensemble and obstructing this air flow to produce speech sounds. The smooth execution of this sequence of events requires the human brain to monitor sensory feedback during speech, correct for any errors and learn from any past errors. This phenomenon is called sensorimotor integration and is essential for efficient speech motor control. Various theoretical and computational models of speech production explain how sensorimotor integration occurs but many aspects of this process still remain underexplored. This dissertation starts by investigating how sensorimotor learning of vowels depends on the complex relationship between articulatory dimensions and acoustic space. Specifically, formant adaptation or response to altered formant frequency feedback depends on the direction of the shift in two-dimensional F1-F2 vowel space. Using magnetoencephalographic imaging, I then investigate how sensorimotor integration is affected during speaking in a voice disorder called Laryngeal Dystonia (or Spasmodic Dysphonia). Significant differences in neural activity at various nodes of the speech motor control network were observed in patients with Laryngeal Dystonia at various time points around the act ofphonation. Lastly, the dissertation describes cortical dynamics of the speech motor control network in a neurodegenerative disorder affecting speech and language called the non-fluent variant of Primary Progressive Aphasia (nfvPPA). These patients have significant motor speech impairments which were investigated using a pitch perturbation experimental paradigm. Neural and behavioural results showed that sensorimotor integration is severely impacted in patients with nfvPPA. Taken together, the work in this dissertation aims to help inform current computational models of speech production and underlines the important role of sensorimotor integration in human speech.

Sensory feedback plays an important role in maintaining steady and fluent speech production. So far, a majority of research in speech has focused on auditory feedback while not a lot has been done on somatosensory feedback. Thus, the current study aims to further explore the effect of vocal tract somatosensory feedback on auditory cortex activity during the process of audiomotor map learning. Due to extensive evidence for a phenomenon known as motor-induced suppression (MIS), the current study hypothesizes that cortical activity will be reduced in subjects after the establishment of learning. Using an MEG touchscreen speech synthesizer set- up, subjects heard a target vowel sound and were asked to touch a location on the touchscreen that matched the sound they heard. With each trial, subjects were given feedback in the form of the sound that corresponded to the location on the screen they touched, resulting in them eventually learning to map out each vowel sound to a target location on the screen. This set-up allowed a paradigm to test for the effect of audiomotor map learning. Then, using the NUTMEG software and Champagne source localization algorithm, data from each subject was analyzed before and after learning, as well as any potential differences between auditory and motor feedback or left and right auditory cortices were noted. Findings across all 4 conditions (left auditory cortex with auditory feedback, right auditory cortex with auditory feedback, left auditory cortex with auditory and somatomotor feedback, right auditory cortex with auditory and somatomotor feedback) in almost all the subjects found a statistically significant decrease in cortical activity following the establishment of audiomotor map learning. The current study sets up future work in which a variety of patient populations as well as different forms of feedback can be studied using a touchscreen speech synthesizing platform.

In the last few decades there have been major advances in the technology of function brain imaging, allowing for insight into the functions of the human brain previously elusive to neuroscientists. These advances have been primarily on the hardware end and developing effective software to interpret the data collected by neuroimaging machines is a current challenge to the use of the technology. Magnetoencephalography (MEG), in particular, requires solving an ill-posed inverse problem in order to uncover the brain areas active during a task. While the solution to this inverse problem is not unique, there are many methods to estimate its solution and this is a field of active research. In Chapter Two of this thesis, we derive an algorithm that solves the inverse problem for MEG, and the related imaging method, electroencephalography (EEG). Our method improves upon existing algorithms in that it incorporates noise suppression into the estimation procedure and is theoretically and empirically robust to correlated sources. In Chapter Three, we show the results from extensive testing of our algorithm using simulated and real M/EEG data and we show our algorithm's results in comparison to the benchmark algorithms. Chapter Four explores variants of the algorithm, including its application to data sets without pre-stimulus data. In Chapter Five, we present methods to statistically threshold the inverse solution results using nonparametric statistics. Finally, in Chapter Six, we provide some concluding remarks and ideas for future research directions. As a whole, the work presented in this thesis improves the interpretation and analysis of M/EEG data.

Speech is central to human communication. One of the features of speech we control to make distinguishable speech is formants, which are the spectral frequencies in our speech that we vary to make different vowel sounds. It is well-understood that the control of speech features, including formants, depends heavily on feedback information. However, the exact mechanisms of feedback control of formants are not fully understood. The body of this dissertation investigates the feedback control of formants in different scenarios to explore its relationship with other speech control mechanisms. We first investigated the relationship of feedback control and the adaptation of feedforward control of formants and found that they have distinct mechanisms. We then explored the neural substrate of feedback processing for formants as compared to pitch using MEG, which allowed us to investigate network activity changes with a high temporal resolution. We found similar neural regions involved in feedback processing for both pitch and formants, though we observed opposite lateralization for pitch and formants across different frequency bands. Finally, we explored the interaction between auditory and somatosensory feedback information in the control of formants by attenuating oral somatosensory feedback using lidocaine while perturbing auditory feedback. We found a direction-dependent interaction between the two feedback modalities, specifically where we observed an increase in compensation responses to downward auditory feedback perturbations but a decrease in compensation responses to upward auditory feedback perturbations when the oral somatosensory feedback was attenuated.

The nature of the integration of auditory and motor processes in the brain has been a subject of long-standing interest in the cognitive sciences and neuroscience(Price, 2012). Dysfunction in auditory-motor integration underlies pathology in a wide range of disease processes including stroke, stuttering, dyslexia, and neuropsychiatric disorders such as schizophrenia. Understanding the functional integration of the neural circuits that subserve the different stages of auditory-motor transformation and coordination remains a major challenge in both basic and clinical human systems neuroscience.

Here, we report on the neural oscillatory correlates of speech auditory encoding, working memory and verbal reproduction (the "phonological loop") in healthy human subjects, and schizophrenic patients both before and after computerized cognitive training. We present a high-resolution spatiotemporal description of the brain network that supports the functions of the phonological loop in the healthy subjects, show the aberrant behavior of this network in schizophrenia, reveal novel compensatory mechanisms deployed in the cortex in schizophrenics, and show how some of the significant abnormalities are normalized through cognitive training.

In the healthy subjects, we demonstrate that Area Spt and Broca's Area act as dissociable input and output buffers in a reverberating bi-directional dorsal stream for phoneme perception and production. These buffers interact in time along a tightly controlled schedule as part of a speech-motor feedback loop that operates before speech onset. In the schizophrenic patients, we find that the normal dorsal stream pathway is disrupted, and that schizophrenics activate the ventral visual stream in compensation, and in particular the visual word form area (VWFA), in a manner mediated by an overall suppressed dorsal lateral prefrontal cortex, the activity level of which in turn correlated with overall patient functional status. Both the posterior dorsal stream deficits and the VWFA compensation are greater in patients with severe auditory hallucinations, reflecting the significant underlying lesion to the auditory-motor system in these patients. Finally, we demonstrate that computerized cognitive training acts to normalize the oscillatory deficits of the dorsal stream and that this normalization correlates with improvements in both task performance and long-term verbal learning and memory scores.

Objective

Approach

Background: Post-traumatic stress disorder (PTSD) is a mental health disorder characterized by symptoms such as insomnia, irritability, issues with memory, difficulty concentrating, and poor decision-making abilities. With symptoms that closely resemble those of other anxiety disorders, it is very difficult to accurately diagnose. More research is needed to identify structural and functional imaging biomarkers to aid in diagnosis.

Methods: Ten right-handed male subjects (5 combat-exposed veterans, 5 healthy civilian controls) underwent magnetoencephalographic recording for this study. MEG data were acquired with a 275-channel whole-head CTF Omega 2000 system. Resting-state and tasked-based (Stroop Color-Naming Task) data were acquired. Voxel-based time-frequency analysis was subsequently performed using NUTMEG and SPM8.

Results: Significant differences were found between the two groups at rest (in delta, theta, gamma, and high-gamma neural oscillatory frequency bands) and during the Stroop Color-Naming task (in alpha, beta, and gamma, and high-gamma frequency bands).

Conclusions: Despite the small sample size, we were able to replicate some aspects of previous MEG research in veterans with PTSD. Not only does this result substantiate the use of MEG for population studies, but it also shows that PTSD is a mental disorder that is physical in nature and can be characterized through passively observing electromagnetic neuronal activity.