Research proposal · Spring 2026
Assistive communication and silent speech interfaces.
Exploring how acoustic-sensing eyewear and language models might support communication while preserving intended meaning and user agency.
- Context
- CS 8803 · Introduction to Research
- My role
- Research proposal preparation
- Status
- Proposed
Supporting expression without losing intention.
Assistive communication raises questions about more than producing words. A system must also consider whether its output reflects what a person meant to say, how it fits into a conversation, and who controls the final expression.
This proposal explores acoustic-sensing eyewear and an LLM-assisted communication approach, with particular interest in participation in professional settings.
A concept to investigate.
The diagram below organizes the supplied proposal direction into a conceptual workflow. The specific sensing technique, model, interaction design, and implementation have not been established here.
Questions shaping the proposal
- How can assistance preserve intended meaning when the input is uncertain?
- What interaction choices could support participation in professional conversations?
- How might personalized synthetic voices affect identity and the user’s control over expression?
From interest to a researchable question.
I prepared Infrastructure of Inclusion for Speech-Impaired Persons for CS 8803 in Spring 2026. I proposed minimally obtrusive acoustic-sensing eyewear paired with LLM-enhanced vocal re-synthesis for assistive workplace communication. I formulated questions on technical information exchange, perceptions of professional authority and communicative competence, and voice cloning’s relationship to user identity.
What would need to be demonstrated?
The following are evaluation considerations for future work, rather than an approved study protocol.
| Dimension | Possible question |
|---|---|
| Intended meaning | Does the user judge the suggested expression to preserve their intention? |
| Interaction | How much effort and time does review or correction require? |
| Participation | Does the interaction support the user’s communication goals in context? |
| Agency & identity | Can the user control the wording, timing, and voice of the output? |
Any participant study would require an appropriate protocol, relevant approvals, and accessibility considerations before recruitment or data collection.
Open questions.
Sensing feasibility, ambiguity in language-model output, privacy, correction burden, and differences between communication contexts all need investigation. The value of a personalized voice should be evaluated with users rather than assumed.