SIDE QUEST / ACCESSIBLE COMMUNICATION / AAC
Traditional symbol communication can force a person to construct language in a compressed form such as I · GO · BATHROOM. A Voice Before Words explores an LLM/NLP layer that interprets that intent and proposes complete human sentences—such as “I need to go to the bathroom.” or, when context supports it, “I went to the bathroom.” The communicator selects the meaning that is actually theirs. The goal is fluent expression without forcing someone to sound robotic simply because symbols are their input method.
This browser experience demonstrates the product concept, not the final intelligence stack. The planned iOS application will use language models to turn compact symbol sequences into clear natural sentences, offer user-selectable human-like voices, and support hands-free eye-gaze access on compatible Apple hardware.
A symbol sequence does not have to be spoken literally. Context-aware NLP can infer plausible grammatical meanings while preserving the communicator's intent.
→ I need to go to the bathroom.Select the concepts you mean without needing to construct perfect grammar.
Generate clear, grammatical alternatives such as present, past or request forms. The user chooses what actually represents their meaning.
Future neural voice models are intended to provide more natural speech with a voice selected by the user.
The planned iOS release is designed to add eye-gaze interaction for people who cannot reliably use touch, using compatible device hardware.
The planned language layer uses modern language models and NLP to translate compact AAC concept sequences into several concise, grammatically correct interpretations. It should understand tense, articles, prepositions, requests, emotion and conversational context while keeping the final choice with the communicator.
The current browser speech button is only a technology demonstration. The intended product integrates higher-quality speech models and lets the user choose a voice that better represents how they want to sound, rather than treating one synthetic system voice as their identity.
The full application is planned for a future iOS App Store release. On compatible Apple hardware, the native application is intended to support eye-gaze interaction so people with severe motor limitations can navigate symbols and communicate without dependable hand input. The repository already explores a native SwiftUI/eye-tracking architecture; this capability is not being falsely represented as working in the browser demo.
An LLM can generate fluent language that the person did not intend. That is why the system should present interpretations rather than silently rewrite communication. The person remains the author: software expands expression, the user approves what is spoken.
JaredVogler.com demonstrates the interaction model and product direction. The intended destination is a substantially more capable native iOS application with model-backed language humanization, higher-quality selectable speech and hardware-assisted gaze input. Accessibility, privacy, latency and failure-safe behavior remain core engineering requirements before treating it as a production communication aid.