Skip to content

YazSes v2.8.0-dev.3 — 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 turn-taking pair:

New since dev.2 (all off by default)

  • Hesitation-Hold Endpointing (ADR-v2-101, [hesitation]) — when you trail off with a filled hesitation ("annnd… uhh…") instead of clean silence, YazSes recognizes you're still composing and holds the turn open instead of cutting you off. is_filled_pause (flat F0 + stable spectrum) + endpoint_decision (extends the commit threshold right after a filled pause). Especially helps slow/aphasic/aging speakers. Anchor: Chatziagapi et al., filled-pause detection (ACII 2022).
  • Pitch-Contour Vocal Gestures (ADR-v2-103, [contour]) — hum a pitch shape as a word-free, language-independent command: rising = confirm, falling = cancel, rise-fall = undo, flat = repeat. normalize_contour (semitone deviations) + classify_contour + gesture_to_command. Robust in noise; for users who can phonate melody but not clear speech. Anchor: Vocal Joystick line + Meyer, Whistled Languages (2015).

yazses features now lists 111 capabilities. (Pitch-Contour Gestures is experimental — features enable requires --force.)

Wave L remaining

Mouth-Sound Switch Access · Breath-Paced Dictation · Whisper-Aware Mode · Involuntary-Vocalization Auto-Excision.

Quality

1356 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 (F0/spectral extraction is an optional lazy extra); this drop also fixed a real contour-classification bug (z-score normalization amplified flat pitch — switched to semitone deviations). The default install and the v1 dictation path are unchanged.