Research
Communication and thought depend on constant self-correction — a slip corrected mid-word, attention shifting under load, a listener reading a half-hidden face. We study that control by changing the conditions around it, and watching how the brain adjusts.
We read the adjustment with whatever instrument the question needs — real-time voice manipulation, fNIRS, EEG, fMRI, machine learning. And because an answer is only as good as the measurement under it, we build and validate the measurement itself, alongside the experiment.
Our questions have always crossed from basic mechanism into clinical populations — vocal control in Parkinson's disease, temporal lobe epilepsy, and cerebellar disorders among them. What is new is where the methods now point: a possible speech intervention for Parkinson's, the effects of cannabis on prefrontal function, and cognitive load measured directly from the brain.
What we ask
shown · where it's headingHow the brain keeps a voice under control
Speaking and singing are governed by a fast loop: the brain monitors its own voice and corrects errors before we notice them. Altering what a talker hears during speech makes that control measurable.
Re-learning the map
ShownWhen a speaker's pitch feedback is lowered a fraction of a semitone at a time, they compensate without being aware of it. When the feedback returns to normal, the compensation persists: the brain has remapped the relationship between a vocal command and its expected sound. Speech is continuously checked against what is heard.
The limits of the loop
Under wayWhen the pitch shift oscillates rather than holding steady, the feedback loop behaves like a controller with a limited bandwidth: it follows slow oscillations and loses fast ones. Whether the loop keeps up appears to determine what the system learns from the error, which would mean that fast correction and slower re-learning draw on a shared signal.
Where the control lives, and why people differ
ShownImaging identifies which regions register a vocal error; briefly disrupting them with TMS or transcranial electrical stimulation shows which ones the brain depends on. These studies have separated the network into roles: some regions carry the error signal, while the prefrontal cortex and cerebellum impose a brake against over-reliance on feedback. Singers fall at one end of the range, detecting finer changes yet relying on feedback less.
Vocal control in Parkinson's disease
Shown → under wayIn Parkinson's disease this balance breaks down: speakers rely too heavily on feedback and over-correct for pitch errors. External cues and targeted stimulation can return those responses toward normal. We are now testing whether priming the system with slow, oscillating pitch has a similar effect, measured with portable fNIRS and pursued in partnership with Parkinson Society Southwestern Ontario, toward speech support that might one day be used at home.
What happens to a voice when the listener is a machine
Children are learning to speak in a world where some of their listeners are machines. A voice assistant does not repair a misunderstanding the way a parent does, which raises a question about how children adjust to be understood, and whether those adjustments persist.
How children adapt to a machine that mishears
SSHRC-fundedWhen a device mishears them, children adjust their speech, slowing down, over-enunciating, or shifting pitch. We track these adjustments with real-time acoustic analysis, building on earlier work showing that children's feedback control is still maturing through childhood.
Whether the machine-voice follows them home
Under wayThe open question is transfer. If a child adopts one way of speaking for a device, it may carry into conversation with people; we compare the same children speaking with machine and human partners.
Why we sometimes misread each other
We read one another in an instant, a feeling in a tone of voice or an intention in a glance. We study how that is accomplished, and what happens when the signal becomes harder to read.
How little of a voice it takes to read a feeling
Under wayBy reducing the emotion in a spoken voice toward the threshold of detection, we ask how little signal a listener needs to read a feeling. An early result revises a familiar finding: the reported female advantage in reading vocal emotion largely disappears once the speed–accuracy trade-off is taken into account, leaving a narrow advantage confined to the subtlest sadness.
Reading emotion in autism — in quiet and in noise
Shown → under wayHow a child manages emotional speech tracks their social experience. We have found that autistic children adjust their own speech to altered feedback more quickly than their peers, and that this difference relates to parent-rated social competence. Current work examines how the reading of emotion holds up against background noise, and whether musical training supports it.
Faces, voices, and masks
Shown → under wayA face and a voice are read together, and our imaging work mapped where the brain combines them. When a mask conceals the lower face, recognition falls, most sharply for disgust, and the accompanying voice does not restore it; subtle expressions suffer most.
How cognition holds up when the brain is pushed
Attention, memory, and judgment fluctuate with the demands placed on them, falling under a hard task, a long day, a substance, or stress, and recovering over different timescales. We read these changes directly from brain activity.
Cannabis and the prefrontal brake
Under wayOur account of cannabis sets aside the usual question of which regions it impairs. The proposal is that THC disrupts the timing between prefrontal networks rather than the networks themselves, degrading coordination rather than capacity. Behaviourally, regular users show a consistent slowing on a fast perceptual task even when accuracy is preserved, a cost that arises downstream of perception itself.
Reading mental workload from the brain
Under wayWhether mental effort can be measured from outside the head, as it changes, remains an open question. We train machine-learning models on fNIRS signals to estimate cognitive load and detect fatigue across tasks, with implications for education and for safety-critical work.
Markers of resilience to adversity
Shown → under waySome people prove more resilient to hardship than others, for reasons that are not obvious. In a cohort followed across the pandemic, with a rare pre-2020 baseline, the impact of lockdown predicted resilience more strongly than a lifetime of prior adversity, and earlier hardship sometimes predicted who became more resilient. We are now following the neural and cognitive markers associated with that resilience.
Selected publications
All publications →How we measure the mind.
Whatever instrument the question demands, often several at once. And because a result is only as good as the measurement behind it, we build and validate the methods themselves.