YazSes v2.10.0-dev.3 — Wave N: Diacritize + SafeGlyph (developer preview)¶

Pre-release · not published (publish workflows are pinned to v0.x/v1.x; v2 dev tags do not ship). v1.4.1 remains the stable release. Every feature is OFF by default.

Continues Wave N (design/vision/v2-research/16-wave-n.md) with the i18n & glyph-integrity tier.

New in dev.3 (all off by default)¶

• Diacritize (ADR-v2-122, [diacritize]) — restore the diacritics ASR drops: "cafe" → "café", "naive" → "naïve", "creme brulee" → "crème brûlée". restore_diacritics — whole-word and longest-first phrase replacement from an unambiguous built-in lexicon, case-preserving, with a user mapping merged on top. Ambiguous words ("resume", "expose") are deliberately left alone — those need context and stay with an optional neural backend. Anchor: NAACL 2024 diacritics restoration.
• SafeGlyph (ADR-v2-123, [safeglyph]) — flag homoglyph hazards in dictated identifiers, URLs, and secrets before they are injected: Cyrillic "а" vs Latin "a", Greek look-alikes, fullwidth forms, zero-width invisibles, and mixed-script words. scan_confusables (findings with index + ASCII suggestion) + normalize_ascii (UTS-39-style skeleton). Anchor: Unicode UTS-39.

yazses features now lists 131 capabilities.

Wave N remaining¶

WordFind (reverse-dictionary) · LoadGuard (cognitive-load guardrails) · Echo (own-audio replay) · SRPace (screen-reader pacing).

Quality¶

1437 tests green across Linux × macOS × Windows on Python 3.11 and 3.12. Both new cores are unit-tested to 100% with no new dependency (pure lexicon + pure UTS-39-subset tables). The default install and the v1 dictation path are unchanged.