Skip to content

YazSes v2.8.0-dev.2 — 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 two more pure cores — one robustness, one mainstream-reach:

New since dev.1 (all off by default)

  • Beam-Steered Spatial VAD (ADR-v2-098, [spatialvad]) — with a stereo/2-mic input, estimates a sound's direction of arrival and drops anything not coming from your seat (a TV, a colleague, a passer-by). gcc_phat (a reverberation-robust phase-transform FFT cross-correlation, pure numpy) + tdoa_to_angle + spatial_gate. Enrollment-free and geometric — an independent second gate that composes with the Cocktail Filter. Anchor: Knapp & Carter GCC-PHAT.
  • Prosodic Auto-Punctuation (ADR-v2-104, [prosodypunct]) — inserts periods, commas, and question marks from prosody alone (pause duration, pitch reset, terminal rise) — no need to say "comma" or "period". punctuate_from_prosody. Distinct from Voice Punctuation (acoustic vs spoken keywords). Anchor: Cho et al., Leveraging Prosody for Punctuation Prediction (Interspeech 2022).

yazses features now lists 109 capabilities.

Wave L remaining

Mouth-Sound Switch Access · Breath-Paced Dictation · Whisper-Aware Mode · Hesitation-Hold Endpointing · Involuntary-Vocalization Auto-Excision · Pitch-Contour Vocal Gestures.

Quality

1350 tests green across Linux × macOS × Windows on Python 3.11 and 3.12. Both new cores are unit-tested to 100% — the Spatial VAD's GCC-PHAT is pure numpy (no new dependency); Prosodic Auto-Punctuation's F0/energy extraction is an optional lazy extra while the decision logic is dependency-free. The default install and the v1 dictation path are unchanged.