Skip to content

YazSes v2.8.0-dev.4 — Wave L 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 L (design/vision/v2-research/14-wave-l.md) with the robustness pair:

New since dev.3 (all off by default)

  • Breath-Paced Dictation (ADR-v2-099, [breath]) — uses your natural breath onsets as sentence/paragraph boundaries, so dictation chunks the way you actually breathe — even without silent pauses. breath_envelope + detect_breath_onsets (debounced) + segment_by_breath, all pure numpy. A different physiological signal than silence. Anchor: BreathPrint (MobiSys 2017).
  • Whisper-Aware Mode (ADR-v2-100, [whispermode]) — detects when you drop to a whisper (no voicing, flat spectral tilt) and adapts gain, VAD threshold, and the STT prompt so quiet dictation stays accurate in shared or private spaces. voicing_ratio + spectral_tilt + is_whispered + whisper_adaptation, pure numpy DSP. Anchor: Ito et al. (ICASSP 2005); wTIMIT.

yazses features now lists 113 capabilities.

Wave L remaining

Mouth-Sound Switch Access · Involuntary-Vocalization Auto-Excision.

Quality

1363 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 (envelope/onset and voicing/tilt are pure numpy; the STT-model hookup is the only downstream wiring). The default install and the v1 dictation path are unchanged.