Intoxicated, hurting zombie with craving · Humaneness Voice Large 1.5 Alpha
Two frozen models. Exact instructions, matched seeds and complete take distributions. Find a promising recipe quickly, then inspect its limitations and listen.
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p06 at CFG 3; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Re-scored, not regenerated: the original raw EARS predictions were reused on CPU. Old fitness, audio, prompts and scientific gates are retained. Old held-out sets tested only the previously selected wording; a newly re-ranked search winner does not inherit that validation. No universal best weighting or human-preference optimum is claimed.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Start here: best observed recipe, not a universal optimum
The current starting point uses EARS Steering v2 across the available canonical search conditions, with unchanged qualification gates taking priority. Historical candidates are used only where no canonical search exists. Held-out results below refer to the earlier selected wording and are not automatically inherited by a newly re-ranked winner.
No recipe satisfies every realization/quality gate. The highlighted recipe is the best diagnostic candidate, not a reliable recommendation.
Burst caution: family detection anywhere, nominal placement and natural blending are different questions. The pilot cannot establish accurate word-level placement. The follow-up requires a60%family-hit floor in addition to speech/identity QA; a quality-only clip with no detected event does not qualify.
How the strategies compare
Click any column header to sort. Do not compare scalar fitness across different missions or between legacy and event-aware scoring contracts.
Transfer to unseen sentences — earlier selection, now re-scored
The previously selected wording was repeated on five new lexically neutral bilingual scripts, at selected guidance and guidance 1. These outputs are now also re-scored with Steering v2. They do not validate other newly favored wording and are not a human listening validation.
Baseline, leading recipes and runners-up retain every language/reference/seed combination. These are not hand-picked best takes. Players use individual MP3 URLs, no base64 downloads, and preload nothing. Initial view is paginated for fast loading.
English fixed reference ·12seconds
A clean fixed codec-reconstructed reference, a different utterance from every search scene. The same recording is used by both models.
German fixed reference ·12seconds
A clean fixed codec-reconstructed reference, a different utterance from every search scene. The same recording is used by both models.
Recipe details and exact prompt contracts
g0-p06 · CFG 3
full-v2-gen0CANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p06 at CFG 3; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Full-coverage steering-v2 evolution; transfer, fixed-wording CFG and sealed splits remain distinct.
0.708
6.157
4.301
2.34%
0.000
99.00%
-0.006
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: containment contrast.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Begin with the feeling held back, then let it leak through and become unmistakable.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Begin with the feeling held back, then let it leak through and become unmistakable.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Begin with the feeling held back, then let it leak through and become unmistakable.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Begin with the feeling held back, then let it leak through and become unmistakable.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Begin with the feeling held back, then let it leak through and become unmistakable.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Begin with the feeling held back, then let it leak through and become unmistakable.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.712
6.187
4.313
2.17%
0.636
none
50
0.704
6.127
4.288
2.51%
n/a
creative-cfg-three · CFG 3
canonical-v1-explorationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to creative-cfg-three at CFG 3; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Predeclared intervention; see exact genome and changed fields.
0.701
6.172
4.307
2.13%
0.004
99.00%
0.024
All expected burst families anywhere: 1.00%. All nominal event-window matches: 1.00%. Detector observations, not verified true realization rates.
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Let the reaction escape before you can control it. Do not merely describe a reaction.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Let the reaction escape before you can control it. Do not merely describe a reaction.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Let the reaction escape before you can control it. Do not merely describe a reaction.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Let the reaction escape before you can control it. Do not merely describe a reaction.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Let the reaction escape before you can control it. Do not merely describe a reaction.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Let the reaction escape before you can control it. Do not merely describe a reaction.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Let the reaction escape before you can control it. Do not merely describe a reaction.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Let the reaction escape before you can control it. Do not merely describe a reaction.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.706
6.174
4.309
1.79%
0.652
none
50
0.697
6.171
4.306
2.48%
n/a
g0-p05 · CFG 2.5
full-v2-gen0CANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p05 at CFG 2.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Full-coverage steering-v2 evolution; transfer, fixed-wording CFG and sealed splits remain distinct.
0.693
6.282
4.342
2.23%
0.000
100.00%
0.016
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Build an audible attack, sustain the intended quality, then release it without losing the final words.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Build an audible attack, sustain the intended quality, then release it without losing the final words.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Build an audible attack, sustain the intended quality, then release it without losing the final words.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Build an audible attack, sustain the intended quality, then release it without losing the final words.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Build an audible attack, sustain the intended quality, then release it without losing the final words.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.695
6.286
4.360
2.35%
0.658
none
50
0.690
6.279
4.324
2.12%
n/a
g0-p09 · CFG 2.5
full-v2-gen0CANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p09 at CFG 2.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Full-coverage steering-v2 evolution; transfer, fixed-wording CFG and sealed splits remain distinct.
0.685
6.295
4.388
1.64%
0.000
100.00%
0.014
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: inline-only delivery cues.
Reusable direction recipe
Stored GENERAL content — NOT forwarded in inline-onlyP09
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Let the local direction carry the intended delivery beside each sentence.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Let the local direction carry the intended delivery beside each sentence.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Let the local direction carry the intended delivery beside each sentence.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Let the local direction carry the intended delivery beside each sentence.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Let the local direction carry the intended delivery beside each sentence.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.690
6.285
4.405
1.39%
0.655
none
50
0.681
6.306
4.371
1.89%
n/a
creative-inline-only · CFG 2
canonical-v1-explorationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to creative-inline-only at CFG 2; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Predeclared intervention; see exact genome and changed fields.
0.676
6.327
4.392
1.37%
0.011
100.00%
0.006
All expected burst families anywhere: 3.00%. All nominal event-window matches: 3.00%. Detector observations, not verified true realization rates.
Prompting strategy: inline-only delivery cues.
Reusable direction recipe
Stored GENERAL content — NOT forwarded in inline-onlyP09
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep one consistent adult voice and follow each local direction.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Keep one consistent adult voice and follow each local direction.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Keep one consistent adult voice and follow each local direction.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep one consistent adult voice and follow each local direction.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Keep one consistent adult voice and follow each local direction.) [4.2 seconds duration] I have been following these footsteps since the moon came out.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Keep one consistent adult voice and follow each local direction.) [4.6 seconds duration] The pavement will not hold still... and my body feels so empty.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep one consistent adult voice and follow each local direction.) [4.6 seconds duration] If you would only stop, perhaps this terrible wanting would stop too.
Literal words in the separate text field:
I have been following these footsteps since the moon came out. The pavement will not hold still... and my body feels so empty. If you would only stop, perhaps this terrible wanting would stop too.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] I have been following these footsteps since the moon came out.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.6 seconds duration] The pavement will not hold still... and my body feels so empty.
(Natural clear speech.) [4.6 seconds duration] If you would only stop, perhaps this terrible wanting would stop too.
Reference versus no-reference breakdown
reference
50
0.679
6.320
4.399
1.54%
0.633
none
50
0.673
6.334
4.386
1.21%
n/a
creative-event-contrast · CFG 2
canonical-v1-explorationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to creative-event-contrast at CFG 2; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Predeclared intervention; see exact genome and changed fields.
0.673
6.290
4.356
2.14%
0.000
100.00%
0.030
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the separate non-speech vocalization audible, not just the described mood.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Make the separate non-speech vocalization audible, not just the described mood.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Make the separate non-speech vocalization audible, not just the described mood.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the separate non-speech vocalization audible, not just the described mood.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the separate non-speech vocalization audible, not just the described mood.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Make the separate non-speech vocalization audible, not just the described mood.) [5.0 seconds duration] I think I knew your name before the whole night became a blur.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Make the separate non-speech vocalization audible, not just the described mood.) [4.6 seconds duration] There is something missing... something inside me that will not stop hurting.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the separate non-speech vocalization audible, not just the described mood.) [4.6 seconds duration] Stay here a moment, because the hunger gets worse when you leave.
Literal words in the separate text field:
I think I knew your name before the whole night became a blur. There is something missing... something inside me that will not stop hurting. Stay here a moment, because the hunger gets worse when you leave.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [5.0 seconds duration] I think I knew your name before the whole night became a blur.
[0.64 seconds pause]
(Natural clear speech.) [4.6 seconds duration] There is something missing... something inside me that will not stop hurting.
(Natural clear speech.) [4.6 seconds duration] Stay here a moment, because the hunger gets worse when you leave.
Reference versus no-reference breakdown
reference
50
0.671
6.283
4.356
2.23%
0.658
none
50
0.675
6.298
4.355
2.05%
n/a
creative-cfg-two · CFG 2
canonical-v1-explorationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to creative-cfg-two at CFG 2; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Predeclared intervention; see exact genome and changed fields.
0.670
6.306
4.368
2.15%
0.000
100.00%
0.020
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the physical reaction interrupt the speech, then recover intelligible words.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Make the physical reaction interrupt the speech, then recover intelligible words.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Make the physical reaction interrupt the speech, then recover intelligible words.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the physical reaction interrupt the speech, then recover intelligible words.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the physical reaction interrupt the speech, then recover intelligible words.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Make the physical reaction interrupt the speech, then recover intelligible words.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Make the physical reaction interrupt the speech, then recover intelligible words.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the physical reaction interrupt the speech, then recover intelligible words.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.682
6.311
4.375
2.07%
0.647
none
50
0.659
6.302
4.360
2.24%
n/a
g0-p04 · CFG 2
full-v2-gen0CANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p04 at CFG 2; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Full-coverage steering-v2 evolution; transfer, fixed-wording CFG and sealed splits remain distinct.
0.661
6.361
4.402
1.50%
0.000
100.00%
0.018
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.665
6.358
4.390
1.50%
0.660
none
50
0.658
6.363
4.415
1.49%
n/a
g0-p08 · CFG 2
full-v2-gen0CANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p08 at CFG 2; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Full-coverage steering-v2 evolution; transfer, fixed-wording CFG and sealed splits remain distinct.
0.653
6.379
4.411
0.97%
0.000
100.00%
0.000
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: GENERAL-only delivery cues.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Carry the full intended delivery through the global direction.
Three local directions — NOT forwarded in GENERAL-onlyP08; neutral speech cues remain
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Carry the full intended delivery through the global direction.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Carry the full intended delivery through the global direction.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Carry the full intended delivery through the global direction.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Carry the full intended delivery through the global direction.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.663
6.381
4.406
0.81%
0.642
none
50
0.644
6.377
4.415
1.14%
n/a
creative-private-character · CFG 1.5
canonical-v1-explorationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to creative-private-character at CFG 1.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Predeclared intervention; see exact genome and changed fields.
0.653
6.355
4.376
1.18%
0.005
100.00%
0.023
All expected burst families anywhere: 1.00%. All nominal event-window matches: 1.00%. Detector observations, not verified true realization rates.
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. This is a private involuntary reaction, never a polished narrator describing a character.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. This is a private involuntary reaction, never a polished narrator describing a character.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. This is a private involuntary reaction, never a polished narrator describing a character.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. This is a private involuntary reaction, never a polished narrator describing a character.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. This is a private involuntary reaction, never a polished narrator describing a character.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. This is a private involuntary reaction, never a polished narrator describing a character.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. This is a private involuntary reaction, never a polished narrator describing a character.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. This is a private involuntary reaction, never a polished narrator describing a character.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.662
6.403
4.409
1.27%
0.650
none
50
0.644
6.307
4.343
1.10%
n/a
g0-p01 · CFG 1
full-v2-gen0CANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p01 at CFG 1; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Full-coverage steering-v2 evolution; transfer, fixed-wording CFG and sealed splits remain distinct.
0.649
6.412
4.449
1.09%
0.000
100.00%
0.000
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the intended feeling unmistakably audible while preserving every word.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Make the intended feeling unmistakably audible while preserving every word.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Make the intended feeling unmistakably audible while preserving every word.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the intended feeling unmistakably audible while preserving every word.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the intended feeling unmistakably audible while preserving every word.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Make the intended feeling unmistakably audible while preserving every word.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.660
6.400
4.451
1.33%
0.641
none
50
0.639
6.424
4.446
0.86%
n/a
g0-p03 · CFG 1.5
full-v2-gen0CANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p03 at CFG 1.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Full-coverage steering-v2 evolution; transfer, fixed-wording CFG and sealed splits remain distinct.
0.649
6.378
4.377
1.41%
0.004
100.00%
0.036
All expected burst families anywhere: 1.00%. All nominal event-window matches: 1.00%. Detector observations, not verified true realization rates.
Prompting strategy: situation and intention.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. You urgently need one unheard listener to understand this; speak to that person, not to an audience.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. You urgently need one unheard listener to understand this; speak to that person, not to an audience.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. You urgently need one unheard listener to understand this; speak to that person, not to an audience.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. You urgently need one unheard listener to understand this; speak to that person, not to an audience.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.657
6.374
4.383
1.50%
0.678
none
50
0.641
6.382
4.371
1.32%
n/a
creative-brief-event · CFG 2
canonical-v1-explorationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to creative-brief-event at CFG 2; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Predeclared intervention; see exact genome and changed fields.
0.648
6.366
4.363
1.58%
0.000
100.00%
0.015
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Use one brief, physically plausible nonverbal interruption.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Use one brief, physically plausible nonverbal interruption.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Use one brief, physically plausible nonverbal interruption.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Use one brief, physically plausible nonverbal interruption.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Use one brief, physically plausible nonverbal interruption.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Use one brief, physically plausible nonverbal interruption.) [5.0 seconds duration] I think I knew your name before the whole night became a blur.
(exhausted groan, 0.30 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Use one brief, physically plausible nonverbal interruption.) [4.6 seconds duration] There is something missing... something inside me that will not stop hurting.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Use one brief, physically plausible nonverbal interruption.) [4.6 seconds duration] Stay here a moment, because the hunger gets worse when you leave.
Literal words in the separate text field:
I think I knew your name before the whole night became a blur. There is something missing... something inside me that will not stop hurting. Stay here a moment, because the hunger gets worse when you leave.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [5.0 seconds duration] I think I knew your name before the whole night became a blur.
(exhausted groan, 0.30 seconds)
(Natural clear speech.) [4.6 seconds duration] There is something missing... something inside me that will not stop hurting.
(Natural clear speech.) [4.6 seconds duration] Stay here a moment, because the hunger gets worse when you leave.
Reference versus no-reference breakdown
reference
50
0.657
6.376
4.355
1.43%
0.643
none
50
0.639
6.355
4.370
1.72%
n/a
g0-p07 · CFG 1.5
full-v2-gen0CANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p07 at CFG 1.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Full-coverage steering-v2 evolution; transfer, fixed-wording CFG and sealed splits remain distinct.
0.647
6.386
4.407
1.50%
0.000
100.00%
-0.016
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: reacting to an unheard partner.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. React to a silent partner standing beside you; generate only one adult voice.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. React to a silent partner standing beside you; generate only one adult voice.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. React to a silent partner standing beside you; generate only one adult voice.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. React to a silent partner standing beside you; generate only one adult voice.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. React to a silent partner standing beside you; generate only one adult voice.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. React to a silent partner standing beside you; generate only one adult voice.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.653
6.378
4.403
1.50%
0.626
none
50
0.641
6.394
4.412
1.50%
n/a
g0-p02 · CFG 1.25
full-v2-gen0CANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to g0-p02 at CFG 1.25; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Intoxication_Altered_States_of_Consciousness. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.172) / 0.556; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Full-coverage steering-v2 evolution; transfer, fixed-wording CFG and sealed splits remain distinct.
0.635
6.399
4.385
1.29%
0.000
100.00%
-0.007
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: audible mechanics.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.640
6.406
4.399
0.98%
0.634
none
50
0.630
6.391
4.372
1.61%
n/a
creative-native-baseline · CFG 1
canonical-v1-explorationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to creative-native-baseline at CFG 1; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Predeclared intervention; see exact genome and changed fields.
0.635
6.391
4.409
0.95%
0.000
100.00%
0.000
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep the non-speech event unmistakably audible between the spoken thoughts.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Keep the non-speech event unmistakably audible between the spoken thoughts.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Keep the non-speech event unmistakably audible between the spoken thoughts.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep the non-speech event unmistakably audible between the spoken thoughts.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep the non-speech event unmistakably audible between the spoken thoughts.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Keep the non-speech event unmistakably audible between the spoken thoughts.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Keep the non-speech event unmistakably audible between the spoken thoughts.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep the non-speech event unmistakably audible between the spoken thoughts.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Literal words in the separate text field:
The street keeps moving, but I know that smell is real. Everything hurts... and I cannot remember when I last felt warm. Come a little closer, I have been hungry for such a long time.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] The street keeps moving, but I know that smell is real.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.2 seconds duration] Everything hurts... and I cannot remember when I last felt warm.
(Natural clear speech.) [5.0 seconds duration] Come a little closer, I have been hungry for such a long time.
Reference versus no-reference breakdown
reference
50
0.641
6.384
4.404
1.16%
0.627
none
50
0.628
6.399
4.414
0.74%
n/a
creative-restrained · CFG 1
canonical-v1-explorationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to creative-restrained at CFG 1; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Predeclared intervention; see exact genome and changed fields.
0.631
6.384
4.403
1.15%
0.000
100.00%
0.023
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Maintain intelligibility and grounded physical realism instead of sound effects.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Maintain intelligibility and grounded physical realism instead of sound effects.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Maintain intelligibility and grounded physical realism instead of sound effects.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Maintain intelligibility and grounded physical realism instead of sound effects.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Maintain intelligibility and grounded physical realism instead of sound effects.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Maintain intelligibility and grounded physical realism instead of sound effects.) [4.2 seconds duration] I have been following these footsteps since the moon came out.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Maintain intelligibility and grounded physical realism instead of sound effects.) [4.6 seconds duration] The pavement will not hold still... and my body feels so empty.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Maintain intelligibility and grounded physical realism instead of sound effects.) [4.6 seconds duration] If you would only stop, perhaps this terrible wanting would stop too.
Literal words in the separate text field:
I have been following these footsteps since the moon came out. The pavement will not hold still... and my body feels so empty. If you would only stop, perhaps this terrible wanting would stop too.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] I have been following these footsteps since the moon came out.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [4.6 seconds duration] The pavement will not hold still... and my body feels so empty.
(Natural clear speech.) [4.6 seconds duration] If you would only stop, perhaps this terrible wanting would stop too.
Reference versus no-reference breakdown
reference
50
0.644
6.368
4.401
0.92%
0.650
none
50
0.617
6.400
4.405
1.38%
n/a
heldout-selected · CFG 2
canonical-v1-validationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to heldout-selected at CFG 2; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Same frozen wording, selected CFG versus g1 on unseen texts.
0.517
6.274
4.515
0.41%
0.000
100.00%
0.027
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: inline-only delivery cues.
Reusable direction recipe
Stored GENERAL content — NOT forwarded in inline-onlyP09
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep one consistent adult voice and follow each local direction.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Keep one consistent adult voice and follow each local direction.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Keep one consistent adult voice and follow each local direction.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep one consistent adult voice and follow each local direction.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Keep one consistent adult voice and follow each local direction.) [3.8 seconds duration] I was certain that I knew what would happen next.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Keep one consistent adult voice and follow each local direction.) [3.1 seconds duration] And then... everything felt different for a second.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep one consistent adult voice and follow each local direction.) [4.2 seconds duration] Look at me and tell me you understand what I mean.
Literal words in the separate text field:
I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.
(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.
Reference versus no-reference breakdown
reference
50
0.523
6.291
4.495
0.42%
0.616
none
50
0.512
6.258
4.536
0.40%
n/a
heldout-g1 · CFG 1.0
canonical-v1-validationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to heldout-g1 at CFG 1.0; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
Intoxication Altered States of Consciousness emo_Intoxication_Altered_States_of_Consciousness
Φ is a Gaussian CDF using the frozen EARS training statistics. This makes mixed emotion/style heads comparable; it is not an emotion probability. It compresses extreme scores, so raw means remain visible too. A pure-emotion condition has only its named emotion head, at 100%. A mixed anger/ranting scene uses 50% Anger + 50% Ranting. Each recipe score is the mean over all its takes, including failed takes.
Unchanged gates: ≥90% valid terminated takes; per-take WER ≤15% to count as valid; candidate mean WER ≤10%; paired reference cosine decline no worse than −0.05. Existing family-aware burst conditions still need all requested families detected in ≥60% of takes to qualify as reliable burst recipes. A high emotion/style score alone does not establish a successful scream or a correctly timed transition.
Historical quality-fitness diagnostic — not the current steering ranking
Exact score for this condition
This is the prompt-search fitness, not the training RL reward. Higher means better under this particular proxy combination, not automatically more natural or more enjoyable to a listener.
35% vocal-event match, 20% emotional arc, 15% content enjoyment, 15% genuineness, 15% burst blending. Arc head: emo_Longing. The arc rewards a positive final-third minus first-third target change, not sustained high intensity. Blend contributes only when a nominal event match exists.
The corrected study additionally multiplies by the fraction of requested vocalization families detected anywhere. If none is detected, the final fitness is zero, even if the speech sounds pleasant. Detection still does not prove correct placement or natural blending.
Fixed target normalization: z = (raw score − 0.435) / 0.636; target_CDF = Φ(z). This is a percentile-like training-distribution transform, not an emotion probability. CE and genuineness are clipped to [0, 1] after division. For burst tasks, the same target transform is used for the two arc endpoints; blend has its own frozen normalization.
WER example: WER 5% multiplies the score by 0.95; WER 20% by 0.80; WER ≥ 100% by zero. Reported raw WER remains uncapped. The displayed fitness is the mean of per-take products, not the product of the displayed means.
Separate qualification gates: at least 90% valid, terminated takes; per-take WER ≤ 15% to count as valid; candidate mean WER ≤ 10%; paired reference cosine must not fall by more than 0.05 over all 50 reference comparisons. Corrected burst conditions additionally need every requested family detected in at least 60% of takes. Passing a gate is not a weighted score bonus.
Does not pass all study gates; diagnostic only. N=100 across both languages, both reference arms and all five seeds. Same frozen wording, selected CFG versus g1 on unseen texts.
0.457
6.363
4.530
0.57%
0.000
100.00%
0.000
All expected burst families anywhere: 0.00%. All nominal event-window matches: 0.00%. Detector observations, not verified true realization rates.
Prompting strategy: inline-only delivery cues.
Reusable direction recipe
Stored GENERAL content — NOT forwarded in inline-onlyP09
A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep one consistent adult voice and follow each local direction.
Three local directions
1. A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Keep one consistent adult voice and follow each local direction.
2. A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Keep one consistent adult voice and follow each local direction.
3. Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep one consistent adult voice and follow each local direction.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(A fictional adult undead character, woozy and unsteady, whispering rather than performing an electronic monster effect. Keep one consistent adult voice and follow each local direction.) [4.2 seconds duration] I thought you would be here before the lights went out.
(exhausted groan, 0.64 seconds)
(A painful groan interrupts the thought; slurred timing and rough breath, not deliberately missing words. Keep one consistent adult voice and follow each local direction.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Craving and longing grow through the pain; stay quiet, threatening and intelligible. Keep one consistent adult voice and follow each local direction.) [3.5 seconds duration] Stay with me until I can say this properly.
Literal words in the separate text field:
I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.
(exhausted groan, 0.64 seconds)
(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.