04-design-e4

design.py on boyspace boyspace_e5_s12000 · 11 voices × 10 lines
Round 04 — FIRST DESIGNED VOICES (blended speaker embeddings on the epoch-4 snapshot; ratios are PROVISIONAL — embeddings still move as training continues, so judge direction, not final identity). shigure anchors the boy register; kokoro adds boyish energy; ami_pun/ayaka dilute shigure's baked-in breathiness (measured: shigure aperiodicity 0.721 = highest in the pool); aoba adds prosodic liveliness. robotic_flat = the hanamaru-neutral direction (intonation 0.4). excited = the nekotsuka direction (intonation 1.5, slightly faster). bright_up = pitch +8%. koko_anchor = the inverse test: kokoro-led with shigure seasoning. Tell me: which 2-3 rows point the right way, whether robotic_flat or excited is closer to HIM, and whether breathiness is sufficiently gone in the diluted rows.
voice title進め、僕の無敵マシン!!!the title line (JP marketing title; the shouted VO line becomes its conlang translation) shortえっ嘘でしょ。ITA EMOTION100_001 — short exclamation casual時間はあるんだから、安全運転してくれよ。ITA EMOTION100_066 — casual boyish register emotion私たちは恐怖の中で生きていて、苦しんでいるの。ITA EMOTION100_073 — emotional narrate家具商人のフィシェルは、荷車と仔馬を貸してくれた。ITA RECITATION324_010 — neutral narration + loanword effortはぁっ!やあっ!うわあっ!wordless combat efforts — expect TTS to struggle; this is the pack's job nonsense1チャタンナオン、モチノッギカー。generated nonsense, Japanese core phonotactics (wordgen ja-core seed 7) nonsense2ソッメンキッヌケデムネーシャメ。generated nonsense, Japanese core phonotactics (wordgen ja-core seed 7) extended1オーナキチェ、ファッドゥコントゥノヌン。generated nonsense, extended loanword kana (wordgen ja-extended seed 7) extended2フッケバツグトチュ。generated nonsense, extended loanword kana (wordgen ja-extended seed 7)
shigure_pureboyspace boyspace_e5_s12000 · blend shigure×1.0
shig70_koko30boyspace boyspace_e5_s12000 · blend shigure×0.7+kokoro×0.3
shig50_koko50boyspace boyspace_e5_s12000 · blend shigure×0.5+kokoro×0.5
shig70_amipun30boyspace boyspace_e5_s12000 · blend shigure×0.7+ami_pun×0.3
shig60_koko20_pun20boyspace boyspace_e5_s12000 · blend shigure×0.6+kokoro×0.2+ami_pun×0.2
shig60_koko20_aoba20boyspace boyspace_e5_s12000 · blend shigure×0.6+kokoro×0.2+aoba_szn×0.2
shig50_koko30_ayk20boyspace boyspace_e5_s12000 · blend shigure×0.5+kokoro×0.3+ayaka×0.2
robotic_flatboyspace boyspace_e5_s12000 · blend shigure×0.6+kokoro×0.2+ami_pun×0.2 · intonation_scale=0.4
excitedboyspace boyspace_e5_s12000 · blend shigure×0.5+kokoro×0.3+ami_pun×0.2 · intonation_scale=1.5, length=0.92
bright_upboyspace boyspace_e5_s12000 · blend shigure×0.6+kokoro×0.2+ami_pun×0.2 · pitch_scale=1.08
koko_anchorboyspace boyspace_e5_s12000 · blend kokoro×0.6+shigure×0.2+ami_pun×0.2