OpenReviewOngoing & Published Research
Submitted1
Automatic Speech Recognition Performance in Psychiatric Speech: Linguistic, Clinical, and Architectural Factors
Abstract
Automatic speech recognition (ASR) is essential for automated speech analysis pipelines in psychosis-spectrum disorders, yet performance on clinical speech remains understudied. We evaluated three modern ASR models (Whisper large-v2, Canary-1B, Parakeet-TDT-1.1B) on speech from 145 participants (92 psychosis-spectrum, 53 healthy controls) across seven tasks spanning structured reading to spontaneous speech, yielding 1,015 recordings. Performance assessed included word error rate (WER), semantic fidelity, disfluency preservation rate, and hallucination scores. The DISCOURSE corpus is publicly available through TalkBank for future ASR models. Whisper (WER: 12.1%) and Canary (12.2%) achieved comparable accuracy; Parakeet showed higher error rates (WER: 21.8%, p < 0.001). Task type was the strongest determinant of accuracy: structured reading (8.5%) outperformed spontaneous free speech (14.3%). Patients showed modestly elevated WER compared to controls, but significant group differences emerged only in open-ended narrative tasks. Sex influenced accuracy for Whisper (p = 0.039), with males showing higher error rates. Age was significantly associated with WER for Canary only (ρ = 0.245, p = 0.003). PANSS scores did not significantly predict WER for Whisper. Second formant bandwidth variability predicted WER (ρ = -0.324, p_FDR = 0.006). All models preserved few disfluency markers (Whisper 22.6% retained, Parakeet 5.3%), with patients showing significantly higher preservation rates than controls. This was acoustically mediated by loudness variability and was specific to disfluency preservation: loudness variability showed no mediation of general WER (all p > 0.31). Findings are limited to English-speaking populations; given that immigration status is a risk factor for psychosis, multilingual validation is essential. Whisper large-v2 is recommended for clinical speech research in psychosis-spectrum disorders based on its accuracy, robustness to symptom severity, and superior disfluency preservation.
Accepted1
Futures Before Failures: Design Fictions for Anticipatory Fairness in Surgical AI
Abstract
Surgical AI is no longer a promise on the horizon. Platforms combining robotic assistance, mixed reality navigation, and real-time decision support are entering operating rooms. Yet the values embedded in these systems, the choices about whose data they train on, whose bodies they optimise for, and whose autonomy they quietly override, are rarely considered before deployment or revisited as these systems evolve. We present VIRTUES (Values and Impact for Responsible Technological Use in Environments of Surgery), a framework for anticipatory fairness evaluation in surgical AI. VIRTUES was developed through a focus group, thematic analysis, and design fictions following a fictional surgical AI platform across three temporal horizons, then evaluated by ten independent raters using structured valence coding. Across these horizons, we find that the most consequential design decisions are neither clearly beneficial nor harmful, but Mixed, simultaneously advancing some values while eroding others. Left unresolved, these tensions accumulate until they become the structural conditions of care. By surfacing these tensions before they become embedded in practice, VIRTUES supports anticipatory fairness evaluation throughout the surgical AI lifecycle.
Published1
User-Centered Design for Surgical Innovations: A Ventriculostomy Case Study
Abstract
A lack of multidisciplinary collaboration during the design phase of surgical innovation development often ignores the people whom we are developing for and therefore omits meaningful and relevant user insights that can potentially be gathered about the context-of-use of a product. To mitigate this issue, we propose a user-centered design approach to developing surgical solutions. End-user involvement during product design has been linked to the development of more useful and usable solutions as it helps create a smooth transition between research, environment, and daily practice. In this paper, we describe the user-centered design process and give an example of how it can be incorporated to enhance the development of surgical innovations. As a case study, we focus on one of the most commonly performed and error-prone neurosurgical procedures, ventriculostomy.
In Preparation3
VIRTUES: A Design-Fiction Framework for Eliciting the Values of Sustainable, Responsible Surgical Innovation
Abstract
The rapid advancement of medical and surgical technologies offers tremendous opportunities, but it also presents significant challenges. As innovations like AI-driven systems, robotic surgery, tissue engineering, and mixed reality become more prominent, it is crucial to consider their societal, ethical, and global implications. This paper explores the use of design fictions, a speculative thinking approach, to envision potential futures shaped by technological advancements in healthcare. We introduce VIRTUES (Values and Impact for Responsible Technological Use in Environments of Surgery), a ten-principle framework aligned with the United Nations Sustainable Development Goals (SDGs) to provide a global evaluation tool. Through fictitious scenarios, we examine how innovations might have unforeseen repercussions and explore their potential benefits, unintended consequences, and long-term challenges. The VIRTUES framework helps assess their impact across its ten principles, focusing not only on technical performance but also on issues like accessibility, equity, privacy, and the dehumanization of care. In doing so, we emphasize the importance of responsibility and foresight in technological development. We argue that using design fictions to envision future possibilities reaffirms human agency, challenges technological determinism, and guides medical advancements, ultimately paving the way for a just and accessible future for all.
Enhancing Speech-Based Parkinson's Detection with Celebs4PD: A Corpus of Natural, In-the-Wild, Longitudinal Celebrity Speech
Abstract
Coming soon.
Confounding Factors in Speech Biomarkers of Parkinson's Disease
Abstract
Coming soon.