native stress prominence gate

Labnote 010 native stress prominence gate🔗

Status: Complete negative result. Native lexical stress did not create reliable directional discourse prominence.

This experiment determines whether Misaki primary/secondary stress produces reliable word-level acoustic prominence rather than merely clean lexical pronunciation. It uses all three eligible Labnote 009 contrasts across both voices and whole-utterance Kokoro generation from token lists, preserving model-generated token timestamps.

For each competing focus word, primary stress must beat secondary stress in at least two of local RMS energy, duration, and median F0 by 3 percent, with no dimension below 0.90. A pair/voice trial passes only when both words pass. Campaign intake requires at least 5/6 trials overall and at least 2/3 for each voice. No DSP is permitted, and no human review is prepared from a failed campaign.

Result🔗

The approval-gated scheduled run completed synthesis and analysis, then exited with the intentional failed-gate status. Zero of six pair/voice trials passed; only two of twelve individual focus words cleared the two-dimension rule. Some words changed duration or energy, but effects were inconsistent, and median F0 remained approximately flat. Neither voice cleared a trial.

This explains Labnote 009: Misaki primary/secondary stress can produce clean native speech, but its lexical-stress annotation does not reliably implement discourse-level contrastive focus. Do not spend more listener labor or tune thresholds around this mechanism. A future prosody compiler requires a model or interface with explicit utterance-level style/prosody conditioning.