GameCult/AquaSynth
Watch
9
0
Fork
You've already forked AquaSynth
 0
Code Issues Pull requests 4 Projects Releases Packages Wiki Activity Actions

Adopt PanPhon realization channel

Browse source
This commit is contained in:
Meta Vens 2026-05-20 22:47:37 +01:00
commit 21c1046f9d
11 changed files with 89 additions and 27 deletions
Download patch file Download diff file Expand all files Collapse all files

4
docs/ipa-vocal-tract-roadmap.md
View file

@ -236,8 +236,8 @@ The first Weksa handoff target is `weksa.utterance_embedding_handoff.v0.1`.
AquaSynth should treat it as the training input contract:
- 1024-float `bge-m3:latest` speech text embedding for semantic text evidence;
- 256-float phonetic realization vector for IPA, phones, pronunciation, and
alien speech-shape evidence;
- 256-float PanPhon-based phonetic realization vector for IPA, phones,
pronunciation, and alien speech-shape evidence;
- 32-float deterministic prosody/emphasis hint vector;
- 64-float projected Ghostlight/Epiphany-shaped character-state vector;
- 64-float AquaSynth-owned learned utterance embedding output.