VOICE_V7 contextual performance acceptance

Status: Accepted for VOICE_V7.

VOICE_V7 adds contextual performance, expanded synthesis dynamic range, and mouth articulation on top of the VOICE_V6 connected-phrase voice. It keeps the four PCA semantic axes as the learnable core and the V2–V6 performance channels.

Scope

• dootdoot --version reports dootdoot VOICE_V7.
• The embedded semantic mapping remains the locked token-to-axis table in assets/dootdoot_asset_v1.doot.
• New deterministic, bounded primitives: a swept-oscillator whistle gesture and wider per-gesture pitch span, a noise/breath excitation blend, role-gated long pauses with suppressible word-boundary bridging, dash/ellipsis hesitation markers, and a gated code-talkbox mouth stage.
• A deterministic discourse-performance planner assigns local phrase roles (probe, chatty_reply, hesitation, terminal_flourish, aside) and continuous performance curves, deploying the primitives by role. Affect and archetype are localized so a high-arousal utterance no longer collapses into one global Yelp; whistle/yelp are reserved for the opener and terminal accent.
• The standalone dash is control-only: it no longer appears with four-axis values in --explain, and --explain surfaces a role: annotation per voiced token.
• Neutral-curve rendering (the empty chirp and hand-built events) is byte-identical to VOICE_V6; only the planner-driven engine path changes samples.
• The committed golden WAV hashes were regenerated under VOICE_V7.

Directional Check

Reference clip: inquisitive-then-chatty.mp3. Dootdoot phrase: Hello - good morning Sandile. What are you doing today?!. Both decoded to mono 44.1 kHz; 2048-sample Hann frames, 512 hop, active frames gated at 8% of each clip's peak frame RMS; dominant spectral peak and an autocorrelation harmonicity proxy tracked per frame. The numbers are produced by the locked uv harness and reproduced with:

scripts/acoustics \
  "Hello - good morning Sandile. What are you doing today?!" \
  /path/to/bb8-clips/inquisitive-then-chatty.mp3

The VOICE_V6 column reproduces the figures from bb8-inquisitive-chatty-gap-analysis.md.

Measurement	Reference	VOICE_V6	VOICE_V7	Read
Active frame fraction	0.44	0.63	0.57	V7 leaves more rests; closer to staged ref
Max internal gap	1138 ms	232 ms	929 ms	V7 opens a real turn gap (was capped)
Dominant peak range	4264 Hz	517 Hz	2649 Hz	V7 can now sweep, not stuck low
Fraction of frames >1.1kHz	0.18	0.00	0.03	V7 reaches whistle range on accents
Harmonicity IQR	0.226	0.050	0.159	V7 noise/breath lets harmonicity swing
Spectral centroid, median	618 Hz	2214 Hz	879 Hz	V7 dropped the constant brightness floor
2–5 kHz share, max	0.834	0.366	0.920	V7 has rare, extreme upper-mid bursts
2–5 kHz share, median	0.000	0.104	0.113	residual floor; brightness is event-shaped but not lower overall

The three co-primary gaps move in the right direction: the question-to-answer timing gap widens from 232 ms to 929 ms, the tonal pitch range widens roughly 5× and now crosses the old ~1.1 kHz ceiling on terminal-flourish accents, and the excitation gains an authored noise/breath path so the harmonicity IQR widens from 0.050 toward the reference's 0.226. The global spectral centroid also falls sharply (2214 Hz → 879 Hz), so V7 does not brighten as a level — exactly the non-goal. The standalone dash is control-only.

Residual: the 2–5 kHz median is still slightly above the reference. The event-based sparkle gives brightness shape and reserves the loudest bursts for accents (max share 0.920), but the constant floor is not yet below the reference. This is left as future tuning rather than a blocking gap, consistent with the non-goal of raising global brightness as a level.

This is a directional listening aid. By-ear acceptance for reliable BB-8-family identity and the "opener, wait, answer" staging remains the final gate. The sample-level contract remains the committed golden WAV hash fixture.