Skip to content

YazSes v2.6.0-dev.3 — Wave J continues (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 J (design/vision/v2-research/12-wave-j.md) with two more pure cores:

New since dev.2 (all off by default)

  • Self-Learning Correction Dictionary (ADR-v2-079, [corrdict]) — learns the ASR errors you keep fixing (e.g. "yaz says" → "YazSes") from your own accepted edits and applies them automatically to future output. mine_substitutions (support-gated, dominant-only — ambiguous edits dropped) + apply_corrections (longest-match, word-boundary-guarded). High-precision by construction: it can only apply corrections it has actually seen. The seq-tagger for generalization is deferred.
  • Voice Fuzzy File Open (ADR-v2-080, [fileopen]) — "open the notes about the mortgage" → fuzzy-matches your local file index (difflib ratio + content-token overlap, stopwords stripped) and opens the best match. Mouse-free file opening for motor-impaired users. The EmbeddingGemma semantic tier is deferred behind a semanticfs extra.

yazses features now lists 91 capabilities.

Wave J remaining

Crowd-Proof Dictation (target-speaker extraction), Voice Jump-to-Symbol, Recording Import (batch transcription), Spoken Shell Pipeline Builder.

Quality

1279 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; the default install and the v1 dictation path are unchanged.