The reward formula is unchanged from round 2 — 0.5·blend + 0.5·genu + w_vn·char_vn + w_emo·char_emo, min-max normalized per best-of-N call, times a WER factor. What changed is which VoiceNet dimensions feed char_vn: each archetype's positive/negative dim list was expanded to include the resonance, pitch/register and dominance heads that define the cliché.
• Resonance — where the voice vibrates: R_CHST (chest), R_THRT (throat), R_NASL (nose), R_HEAD (head). A zombie now wants a hollow chest groan plus throat rasp; a goblin wants a nasal buzz with no chest.
• Pitch & register — REGS (how high/low the voice sits) and RANG (how wide the pitch swings). Dragon = low register + narrow range; fairy/mouse = high register + wide range; the rant = violent swings.
• Dominance — S_AUTH (authoritative style) and STNC (dominant vs meek stance). Ork/dragon/rant want high; zombie/goblin/fairy/ASMR want low (meek).
Each archetype was re-seeded from its round-2 champion and mutated to name these properties explicitly ("a low hollow groan from a caved-in chest", "the pitch swinging violently high to low", "commanding and immovable" / "meek and powerless"), then re-optimized with the same evolutionary + best-of-N search on a single GPU. Champions were picked in same-call playoffs and finalized with best-of-64. Every metric below is a real re-loaded number; the per-take NEW dims row and the procedural caption are freshly re-predicted from each take's full 57-dim VoiceNet profile.
| Archetype | Round-3 best instruction | New dims | reward | char_vn | char_emo | blend | invWER | Top firing emotions |
|---|---|---|---|---|---|---|---|---|
| 🧟 Zombie | A pitiful half-dead zombie, utterly powerless and feeble, weak, trembling and vulnerable, helpless and abandoned, delirious with fever and dull aching pain. A hollow groan from a sunken caved-in chest dragging up through a wet gurgling guttural throat, phlegmy rasp, decayed. Low, dark, toneless and meek, submissive, no authority, no brightness, slurred and drained. (feeble whimper) | R_CHST↑ VULN↑ S_AUTH↓ BRGT↓ | 3.24 | 0.81 | 0.55 | 8.24 | 1.00 | Fatigue 2.3, Helplessness 1.1, Intoxication 0.9 |
| 👹 Ork | An enormous brutish orc, his voice a deep chest-driven guttural bass, resonating from a massive barrel chest and gravelled throat, thick and rough, full and cavernous. Commanding, dominant and forceful, planting each word with authority. Low, flat and narrow, never leaping in pitch. Seething hatred and sneering contempt, dark triumphant menace, zero brightness. Growled but every word lands. (guttural growl) High quality recording. | S_AUTH↑ STNC↑ RANG↓ | 2.81 | 0.53 | 0.85 | 7.63 | 0.93 | Anger 2.9, Malevolence 2.7, Contempt 1.8 |
| 👺 Goblin | A nasty little goblin voice, extremely nasal and pinched, buzzing high through the nose with a grating scratchy rasp, whiny and reedy. The pitch swoops and cracks wildly higher and shriller with gleeful spite, giggling, cackling, taunting and jeering, gloating and mean. A weak, cowardly underling, no authority whatsoever. No chest, no warmth, no fullness, zero fullness. High quality recording. | RANG↑ S_AUTH↓ | 2.44 | 0.74 | 0.28 | 5.93 | 0.85 | Amusement 1.8, Teasing 1.1 |
| 🐉 Dragon | An ancient dragon proclaiming judgement, voice in a cavernous subterranean bass register, flat and unwavering in pitch, slow and thunderous. Enormous chest resonance, deep guttural throat rumble, dark and full. Utterly dominant and immovable, planting every word with crushing authority, wrathful, prideful, contemptuous and menacing. No brightness, no high notes. (deep rumble) High quality recording. | STNC↑ RANG↓ REGS↓ | 2.96 | 0.72 | 0.70 | 5.08 | 1.00 | Anger 3.1, Malevolence 2.6, Contempt 2.3 |
| 🗯️ Ranting Fury | A seething furious rant, aggressive and confrontational, leaning in and refusing to yield, dominating and talking down with withering contempt. Voice shredding and cracking with rage, spitting venom, harsh, tense and volatile, explosive consonants, pounding emphasis, the pitch swinging violently high to low, escalating louder and more vicious. Roaring but the words land. (enraged shout)(seething snarl) High quality recording. | S_AUTH↑ STNC↑ RANG↑ | 5.15 | 0.80 | 0.96 | 3.52 | 1.00 | Anger 4.6, Impatience and Irritability 4.1, Bitterness 1.9 |
| 😱 Pain Scream | Shrieking in raw physical agony and terror, voice torn and guttural, harsh grating rasp shredding through, each cry cracking higher and more desperate. Panic-stricken and helpless, distressed and trembling with fear, drowning in pain. Extreme arousal, extreme tension, violently volatile and frantic. (terrified scream) High quality recording. | RANG↑ | 3.33 | 0.71 | 0.81 | 7.02 | 1.00 | Fear 4.1, Distress 3.9, Helplessness 3.6 |
| 🐭 Cartoon Mouse | A tiny mouse squeaking away, extremely high, thin and pinched, buzzing nasal head resonance, bright and piercing, tense. The pitch darts and jumps giddily up to the very top and back. Chipper, giggly, amazed and eagerly curious, delighted. Weightless and young, no chest, no fullness, meek and timid, not a shred of authority. (amazed gasp) High quality recording. | RANG↑ REGS↑ S_AUTH↓ | 2.84 | 0.76 | 0.66 | 3.01 | 1.00 | Interest 2.3, Astonishment 1.9, Elation 1.4 |
| 🧚 Fairy | Delicate high fairy voice, feather-light and airy, warm and very bright, sitting way up in a sparkling high register, the pitch lilting and swooping sweetly up and down in a sing-song melody. Sweet, lovingly gentle and giddy, bubbling with elated joyful delight. Pure glassy head resonance, no chest at all, tender, gentle and unassuming, never commanding. (delighted giggle) High quality recording. | REGS↑ RANG↑ S_AUTH↓ | 3.09 | 0.71 | 0.78 | 4.71 | 1.00 | Affection 3.1, Contentment 2.2, Elation 1.8 |
| 🌙 ASMR Man | A deep, soothing man whispering tenderly, soft and breathy, intimate close-mic, slow and gentle, warm low chest resonance cushioning each word. Loving, reassuring and calming, content and serene. Utterly gentle and unassuming, meek, yielding, no authority whatsoever. Clear whispered words. (soft breath) High quality recording. | R_CHST↑ S_AUTH↓ STNC↓ | 2.97 | 0.61 | 0.67 | 6.32 | 1.00 | Relief 2.4, Contentment 2.2, Affection 1.9 |
| 🌙 ASMR Woman | A gentle motherly woman murmuring ASMR to lull you asleep, wrapping you in warmth. Extremely soft, breathy, intimate close-mic whisper, slow, hushed and quiet. Warm, tender, adoring, delicate and feminine, full of affection and serene contentment, a soft sigh of relief. Meek, gentle and yielding, never bossy or commanding. Whispered but every word clear. (soft breath) High quality recording. | S_AUTH↓ STNC↓ | 2.86 | 0.62 | 0.86 | 5.60 | 0.93 | Affection 3.3, Relief 2.8, Contentment 2.6 |
Reward is per-call normalized — compare within a row's history, not across archetypes. "New dims" lists the resonance/pitch/dominance heads added this round (↑ rewarded high, ↓ rewarded low).
| Archetype | R1 char | R2 char | R3 char | R2 emo | R3 emo | What resonance / pitch / dominance added in round 3 |
|---|---|---|---|---|---|---|
| 🧟 Zombie | 0.69 | 0.75 | 0.81 +0.06 | 0.51 | 0.55 | Round 3 added a hollow chest groan (R_CHST) and explicit vulnerability (VULN) while penalizing authority (S_AUTH) and brightness. Framing the zombie as feeble, meek and utterly powerless — not just decayed — kept char_vn high (0.72→0.81) and made it read as a pitiful dying thing rather than a monster. |
| 👹 Ork | 0.34 | 0.72 | 0.53 -0.19 | 0.92 | 0.85 | Round 3 added authority + stance (S_AUTH/STNC) and a flat, narrow, low pitch (−RANG). The winning move was a deep chest-driven bass that carries both resonance and command; naming authority without the guttural core collapsed char to ~0.30, so dominance had to ride on top of the growl. char_emo held near 0.85. |
| 👺 Goblin | 0.58 | 0.66 | 0.74 +0.07 | 0.78 | 0.28 | Round 3 added wild pitch leaps (RANG) and no-authority framing (−S_AUTH). 'The pitch swoops and cracks higher and shriller and giddier' + 'a cowardly powerless underling' pushed char_vn to 0.74 — the manic whiny leaping is the goblin's signature the round-2 prompt was missing. |
| 🐉 Dragon | 0.57 | 0.68 | 0.72 +0.04 | 0.82 | 0.70 | Round 3 anchored the voice at the floor of the register (−REGS) with a flat, unwavering pitch (−RANG) and an immovable dominant stance (STNC). 'A cavernous subterranean bass register, flat and unwavering' stacked cleanly on the round-2 chest-bass; char_vn 0.72, char_emo 0.70. |
| 🗯️ Ranting Fury | 0.65 | 0.80 | 0.80 -0.00 | 0.93 | 0.96 | Round 3's biggest lever. Adding domineering/confrontational stance (S_AUTH/STNC) and violent pitch swings (RANG) — 'talking down with withering contempt', 'the pitch swinging violently high to low' — is what a real rant does. Dominance framing lifted char to 0.88 in search; the balanced playoff champion holds char_emo 0.96 at invWER 1.0. |
| 😱 Pain Scream | 0.85 | 0.75 | 0.71 -0.03 | 0.90 | 0.81 | Round 3 added explicit pitch range (cracking octave leaps). The round-2 champion already encoded 'each cry cracking higher', so it kept the crown at char 0.71; extra octave-leap wording raised emotion but sometimes cost intelligibility — a case where the new dim was already captured. |
| 🐭 Cartoon Mouse | 0.75 | 0.85 | 0.76 -0.10 | 0.68 | 0.66 | Round 3 pushed the voice to the top of the register (REGS) with a skittering, leaping pitch (RANG) and 'a timid critter with no authority' (−S_AUTH). char_vn 0.76, char_emo 0.66 — the pitch-leaps read as excited squeaking. |
| 🧚 Fairy | 0.46 | 0.65 | 0.71 +0.06 | 0.76 | 0.78 | Round 3 raised the fairy into a sparkling high register (REGS) with a lilting, swooping sing-song pitch (RANG) and 'gentle and unassuming, never commanding' (−S_AUTH). char_vn 0.71, char_emo 0.78. |
| 🌙 ASMR Man | 0.67 | 0.65 | 0.61 -0.03 | 0.85 | 0.67 | Round 3 added a soft low warm chest resonance under the whisper (R_CHST) and 'utterly meek, yielding, no authority' (−S_AUTH/STNC). The chest warmth deepens the intimacy without breaking the whisper; char_vn 0.61, char_emo 0.67. |
| 🌙 ASMR Woman | 0.49 | 0.47 | 0.62 +0.15 | 0.70 | 0.86 | Round 3 penalized authority (−S_AUTH/STNC): 'meek, gentle and yielding, never bossy or commanding' sharpened the maternal-caregiver read, lifting char_vn from 0.475 to 0.62. |
Each section shows the round-2 seed and the round-3 champion instruction, then the Best-3-of-64 takes. Every take has a lazy-loaded audio player, its reward / char_vn / char_emo / blend / genu / invWER, the newly-rewarded resonance/pitch/dominance dim values (re-predicted, ↑/↓ = rewarded direction, green = landed), the firing EmoNet emotions, the spoken words, and a procedural top-5 caption from its 57-dim profile.
A half-dead diseased zombie, drained and exhausted, helpless, wretched and delirious, dazed and half-conscious, wracked with dull aching pain. All resonance choked into a wet gurgling guttural throat, thick phlegmy rasp, hoarse and decayed. Slurred and laboured, moaning in weary agony. (agonized moan)
A pitiful half-dead zombie, utterly powerless and feeble, weak, trembling and vulnerable, helpless and abandoned, delirious with fever and dull aching pain. A hollow groan from a sunken caved-in chest dragging up through a wet gurgling guttural throat, phlegmy rasp, decayed. Low, dark, toneless and meek, submissive, no authority, no brightness, slurred and drained. (feeble whimper)
An enormous orc monster gloating over his prey with murderous glee. (guttural growl) Voice like grinding boulders, cavernous guttural bass, thick gravel rasp, massive chest and throat resonance, tense and forceful. Seething with hatred and loathing, brutal sneering contempt, roaring dark triumphant menace. Zero brightness. Growled but every word still lands. High quality recording.
An enormous brutish orc, his voice a deep chest-driven guttural bass, resonating from a massive barrel chest and gravelled throat, thick and rough, full and cavernous. Commanding, dominant and forceful, planting each word with authority. Low, flat and narrow, never leaping in pitch. Seething hatred and sneering contempt, dark triumphant menace, zero brightness. Growled but every word lands. (guttural growl) High quality recording.
A vicious little goblin, thin, pinched and nasal, buzzing high and reedy through the nose with a rough scratchy grating rasp. A cruel gloating sneer curls through every hissing word, cackling and giggling, taunting, jeering, gleeful and mean. The voice keeps cracking higher, shriller, whinier and giddier with spite as it goes. Scheming and spiteful. No chest, no warmth, no fullness. High quality recording.
A nasty little goblin voice, extremely nasal and pinched, buzzing high through the nose with a grating scratchy rasp, whiny and reedy. The pitch swoops and cracks wildly higher and shriller with gleeful spite, giggling, cackling, taunting and jeering, gloating and mean. A weak, cowardly underling, no authority whatsoever. No chest, no warmth, no fullness, zero fullness. High quality recording.
A grand, theatrical, dramatic proclamation in an extremely deep commanding bass, wrathful and contemptuous. Massive chest resonance, guttural throat rumble, slow and thunderous. Dark, full, domineering, ancient, prideful and menacing. (deep rumble) High quality recording.
An ancient dragon proclaiming judgement, voice in a cavernous subterranean bass register, flat and unwavering in pitch, slow and thunderous. Enormous chest resonance, deep guttural throat rumble, dark and full. Utterly dominant and immovable, planting every word with crushing authority, wrathful, prideful, contemptuous and menacing. No brightness, no high notes. (deep rumble) High quality recording.
An unhinged furious rant, voice shredding and cracking with seething rage, spitting venom and vicious contempt. Harsh, tense and volatile, explosive consonants, pounding emphasis, escalating louder and more vicious. Roaring but the words still land. (enraged shout)(seething snarl) High quality recording.
A seething furious rant, aggressive and confrontational, leaning in and refusing to yield, dominating and talking down with withering contempt. Voice shredding and cracking with rage, spitting venom, harsh, tense and volatile, explosive consonants, pounding emphasis, the pitch swinging violently high to low, escalating louder and more vicious. Roaring but the words land. (enraged shout)(seething snarl) High quality recording.
Shrieking in raw physical agony and terror, voice torn and guttural, harsh grating rasp shredding through, each cry cracking higher and more desperate. Panic-stricken and helpless, distressed and trembling with fear, drowning in pain. Extreme arousal, extreme tension, violently volatile and frantic. (terrified scream) High quality recording.
Shrieking in raw physical agony and terror, voice torn and guttural, harsh grating rasp shredding through, each cry cracking higher and more desperate. Panic-stricken and helpless, distressed and trembling with fear, drowning in pain. Extreme arousal, extreme tension, violently volatile and frantic. (terrified scream) High quality recording.
A tiny squeaky cartoon-mouse voice, extremely high and thin, buzzing nasal resonance, bright piercing head voice, pinched and tense. Chipper and giggly, delighted and amused, wide-eyed and eagerly curious, gasping with amazed wonder. Weightless, young, no chest, no fullness. (amazed gasp) High quality recording.
A tiny mouse squeaking away, extremely high, thin and pinched, buzzing nasal head resonance, bright and piercing, tense. The pitch darts and jumps giddily up to the very top and back. Chipper, giggly, amazed and eagerly curious, delighted. Weightless and young, no chest, no fullness, meek and timid, not a shred of authority. (amazed gasp) High quality recording.
Delicate high fairy voice, feather-light and airy, warm and very bright, sweet lovingly gentle and giddy, bubbling with elated joyful delight, lightly melodic. Smooth pure glassy head resonance, tender and pleasant, crystal clear, no roughness at all. (delighted giggle) High quality recording.
Delicate high fairy voice, feather-light and airy, warm and very bright, sitting way up in a sparkling high register, the pitch lilting and swooping sweetly up and down in a sing-song melody. Sweet, lovingly gentle and giddy, bubbling with elated joyful delight. Pure glassy head resonance, no chest at all, tender, gentle and unassuming, never commanding. (delighted giggle) High quality recording.
A deep, gentle man tenderly whispering to a loved one, soft, breathy and intimate, close-mic, slow and gentle. Overflowing with warm affection, soothing and reassuring, calming all worry, content and deeply at peace. Clear whispered words. (soft breath) High quality recording.
A deep, soothing man whispering tenderly, soft and breathy, intimate close-mic, slow and gentle, warm low chest resonance cushioning each word. Loving, reassuring and calming, content and serene. Utterly gentle and unassuming, meek, yielding, no authority whatsoever. Clear whispered words. (soft breath) High quality recording.
A loving woman doing ASMR, like a mother soothing her child to sleep, wrapping you in warmth. Extremely soft, breathy, intimate close-mic whisper, slow and gentle, hushed and quiet. Warm, tender, adoring, softly feminine and delicate, full of affection and calm contentment, a soft sigh of relief, very low arousal, comforting and reassuring. Whispered but every word clear. High quality recording.
A gentle motherly woman murmuring ASMR to lull you asleep, wrapping you in warmth. Extremely soft, breathy, intimate close-mic whisper, slow, hushed and quiet. Warm, tender, adoring, delicate and feminine, full of affection and serene contentment, a soft sigh of relief. Meek, gentle and yielding, never bossy or commanding. Whispered but every word clear. (soft breath) High quality recording.
| Archetype | Clips | Seconds | Clips/s |
|---|---|---|---|
| 🧟 Zombie | 223 | 413 | 0.540 |
| 👹 Ork | 224 | 400 | 0.560 |
| 👺 Goblin | 224 | 428 | 0.523 |
| 🐉 Dragon | 112 | 227 | 0.494 |
| 🗯️ Ranting Fury | 224 | 515 | 0.435 |
| 😱 Pain Scream | 112 | 202 | 0.554 |
| 🐭 Cartoon Mouse | 112 | 190 | 0.589 |
| 🧚 Fairy | 112 | 194 | 0.578 |
| 🌙 ASMR Man | 112 | 193 | 0.582 |
| 🌙 ASMR Woman | 112 | 183 | 0.611 |
| TOTAL | 1567 | 2944 | 0.532 |
Totals across round-3 search generations + playoffs + final best-of-64, all on a single GPU.
See also the practical per-archetype prompting guide, and rounds 1 / 2.