Skip to content

ADR-v2-012 — Accessibility Continuum

Status: Accepted (2026-07-02) · Wave C (experimental) Context links: [[adr-v2-000-interaction-layer]], [[adr-v2-009-personal-adapter]] (atypical-speech personalization), [[adr-015-dysfluency-friendly]] (existing), [[adr-011]]

Context

The accessibility research (internal) documents a continuum of speech abilities poorly served by mainstream dictation: quiet/whispered speech (fatigue, shared spaces), atypical/dysarthric speech, and users who need semantic rather than literal capture. YazSes already ships Dysfluency-Friendly Mode (ADR-015) and the Personal Adapter (ADR-v2-009); the gap is a coherent, opt-in continuum rather than scattered flags.

Decision

Group three opt-in accessibility capabilities under one continuum: - Low-Effort / Whisper Mode — a quiet-speech STT profile (lower VAD threshold, whisper-tuned decode) so whispered dictation is captured without shouting. - Semantic Capture — an opt-in "capture the meaning, not the exact words" pass (reuses the LLM-cleanup path, ADR-013) for users whose speech is effortful. - Adaptive pacing — endpoint/hold timings that widen for slower speakers.

Each is off by default and composes with the Personal Adapter. EXPERIMENTAL where it depends on the LLM path; --force to enable those. No new base deps; honours ADR-011.

Consequences

  • Turns scattered accessibility flags into a discoverable, documented set.
  • Reuses existing VAD, LLM-cleanup, and personalization machinery — little new surface.
  • Whisper Mode is pure config (VAD/threshold); Semantic Capture inherits the LLM extra.