Skip to content

YazSes v2.5.0-dev.3 — Wave I heavier tier begins (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 I (design/vision/v2-research/11-wave-i.md) into the heavier tier — pure cores that ship now, with the heavy backend deferred behind an optional extra:

New since dev.2 (all off by default)

  • Hard Contextual Biasing (ADR-v2-069, [hotwords]) — biases recognition toward your vocabulary with a hotword trie and an N-best rescorer, so rare names and jargon actually win — retraining-free, unlike the soft initial_prompt. The in-decoder logit-biasing hook (sherpa-onnx/WFST) is deferred behind a biasing extra.
  • Voice Window Management (ADR-v2-070, [windowctl]) — hands-free desktop layout: "move window left half", "tile top left", "maximize", "workspace 3", "next workspace". A motor- accessibility win untouched by Mouse Grid / Voice Pilot / Head-Pointer. Per-compositor backends (wmctrl/hyprctl/swaymsg/AppleScript/Win32) are deferred behind a windowctl extra.

yazses features now lists 81 capabilities.

Wave I remaining

Diarized Conversation Capture + Rename-by-Voice, Adaptive Latency Governor (speculative decoding), Per-Language Auto Model Switching, Citation-by-Voice from a local BibTeX/CSL library.

Quality

1232 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.