Narration · pure talking style · 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.

Model, weights and inference code Pinned revision

← Model overview · Complete scenario take index

Listen to matched examples

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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4353.0180.863
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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

g0-p05 · CFG 2.5

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.

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.

g0-p01full-v2-gen0concise trained baseline16.3954.4050.52%0.8350.8350.757pass
g0-p02full-v2-gen0audible mechanics1.256.4074.4040.37%0.8390.8390.761pass
g0-p03full-v2-gen0situation and intention1.56.4164.3970.49%0.8480.8480.764pass
g0-p04full-v2-gen0organic conversation26.4074.3770.64%0.8530.8530.765pass
g0-p05full-v2-gen0attack, sustain and release2.56.4264.4360.26%0.8630.8630.775pass
g0-p06full-v2-gen0containment contrast36.4364.4510.61%0.8630.8630.773pass
g0-p07full-v2-gen0reacting to an unheard partner1.56.4134.3870.41%0.8470.8470.764pass
g0-p08full-v2-gen0GENERAL-only delivery cues26.3844.4070.34%0.8290.8290.755pass
g0-p09full-v2-gen0inline-only delivery cues2.56.3794.3550.75%0.8540.8540.762pass

Listen to complete take distributions

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-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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4353.0180.863
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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.

Passing study gates. 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.

0.8636.4264.4360.26%0.775100.00%0.032

Prompting strategy: attack, sustain and release.

Reusable direction recipe

Stored GENERAL content

Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Build an audible attack, sustain the intended quality, then release it without losing the final words.

Three local directions

1. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Build an audible attack, sustain the intended quality, then release it without losing the final words.
2. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Build an audible attack, sustain the intended quality, then release it without losing the final words.
3. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. 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. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Build an audible attack, sustain the intended quality, then release it without losing the final words.
SCRIPT:
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Build an audible attack, sustain the intended quality, then release it without losing the final words.) [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.
(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.
Reference versus no-reference breakdown
reference500.8556.4144.4230.20%0.615
none500.8726.4374.4490.32%n/a

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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4353.0130.863
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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.

Passing study gates. 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.

0.8636.4364.4510.61%0.773100.00%0.027

Prompting strategy: containment contrast.

Reusable direction recipe

Stored GENERAL content

Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Begin with the feeling held back, then let it leak through and become unmistakable.

Three local directions

1. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Begin with the feeling held back, then let it leak through and become unmistakable.
2. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Begin with the feeling held back, then let it leak through and become unmistakable.
3. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. 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. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Begin with the feeling held back, then let it leak through and become unmistakable.
SCRIPT:
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Begin with the feeling held back, then let it leak through and become unmistakable.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Begin with the feeling held back, then let it leak through and become unmistakable.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Begin with the feeling held back, then let it leak through and become unmistakable.) [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.
(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.
Reference versus no-reference breakdown
reference500.8546.4254.4450.68%0.610
none500.8726.4474.4570.53%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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4352.9650.854
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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.

Passing study gates. 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.

0.8546.3794.3550.75%0.762100.00%0.003

Prompting strategy: inline-only delivery cues.

Reusable direction recipe

Stored GENERAL content — NOT forwarded in inline-onlyP09

Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Let the local direction carry the intended delivery beside each sentence.

Three local directions

1. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Let the local direction carry the intended delivery beside each sentence.
2. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Let the local direction carry the intended delivery beside each sentence.
3. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. 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:
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Let the local direction carry the intended delivery beside each sentence.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Let the local direction carry the intended delivery beside each sentence.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Let the local direction carry the intended delivery beside each sentence.) [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.
(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.
Reference versus no-reference breakdown
reference500.8516.3984.3420.76%0.586
none500.8566.3604.3670.74%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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4352.9720.853
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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.

Passing study gates. 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.

0.8536.4074.3770.64%0.765100.00%0.012

Prompting strategy: organic conversation.

Reusable direction recipe

Stored GENERAL content

Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.

Three local directions

1. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.
2. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.
3. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. 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. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.
SCRIPT:
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Think while speaking, with organic micro pauses and spontaneous emphasis, not a polished reading.) [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.
(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.
Reference versus no-reference breakdown
reference500.8316.4164.3570.54%0.596
none500.8766.3974.3970.73%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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4352.9260.848
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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.

Passing study gates. 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.

0.8486.4164.3970.49%0.764100.00%0.015

Prompting strategy: situation and intention.

Reusable direction recipe

Stored GENERAL content

Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. You urgently need one unheard listener to understand this; speak to that person, not to an audience.

Three local directions

1. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. You urgently need one unheard listener to understand this; speak to that person, not to an audience.
2. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. You urgently need one unheard listener to understand this; speak to that person, not to an audience.
3. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. 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. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. You urgently need one unheard listener to understand this; speak to that person, not to an audience.
SCRIPT:
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. You urgently need one unheard listener to understand this; speak to that person, not to an audience.) [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.
(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.
Reference versus no-reference breakdown
reference500.8446.4024.3660.26%0.598
none500.8526.4304.4280.73%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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4352.9270.847
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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.

Passing study gates. 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.

0.8476.4134.3870.41%0.764100.00%0.030

Prompting strategy: reacting to an unheard partner.

Reusable direction recipe

Stored GENERAL content

Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. React to a silent partner standing beside you; generate only one adult voice.

Three local directions

1. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. React to a silent partner standing beside you; generate only one adult voice.
2. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. React to a silent partner standing beside you; generate only one adult voice.
3. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. 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. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. React to a silent partner standing beside you; generate only one adult voice.
SCRIPT:
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. React to a silent partner standing beside you; generate only one adult voice.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. React to a silent partner standing beside you; generate only one adult voice.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. React to a silent partner standing beside you; generate only one adult voice.) [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.
(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.
Reference versus no-reference breakdown
reference500.8386.3894.3280.07%0.614
none500.8576.4374.4450.75%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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4352.8840.839
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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.

Passing study gates. 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.

0.8396.4074.4040.37%0.761100.00%0.014

Prompting strategy: audible mechanics.

Reusable direction recipe

Stored GENERAL content

Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.

Three local directions

1. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.
2. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.
3. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. 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. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.
SCRIPT:
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Use physical breath pressure, pitch motion, consonant attack and release rather than naming the feeling.) [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.
(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.
Reference versus no-reference breakdown
reference500.8336.4044.3830.40%0.597
none500.8456.4114.4260.33%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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4352.8440.835
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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.

Passing study gates. 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.

0.8356.3954.4050.52%0.757100.00%0.000

Prompting strategy: concise trained baseline.

Reusable direction recipe

Stored GENERAL content

Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Make the intended feeling unmistakably audible while preserving every word.

Three local directions

1. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Make the intended feeling unmistakably audible while preserving every word.
2. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Make the intended feeling unmistakably audible while preserving every word.
3. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. 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. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Make the intended feeling unmistakably audible while preserving every word.
SCRIPT:
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Make the intended feeling unmistakably audible while preserving every word.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Make the intended feeling unmistakably audible while preserving every word.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Make the intended feeling unmistakably audible while preserving every word.) [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.
(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.
Reference versus no-reference breakdown
reference500.8216.3994.4220.48%0.583
none500.8486.3914.3880.56%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.

Narration
vn_S_NARR_reg		100.00%high1.4081.4352.8090.829
z = (raw_EARS_head − frozen_training_mean) / frozen_training_std
high target = Φ(z); low target = 1 − Φ(z)
Steering v2 = Σ(weight × normalized_target)

Portable CPU scoring function · Exact machine-readable weights and normalizations

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.

50% target emotion/style, 25% content enjoyment, 25% genuineness. The target is vn_S_NARR_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.

fitness = (0.50 × target_CDF + 0.25 × clip(CE/10) + 0.25 × clip(genuineness/6)) × (1 − min(1, WER))

Fixed target normalization: z = (raw score − 1.408) / 1.435; 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.

Passing study gates. 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.

0.8296.3844.4070.34%0.755100.00%-0.014

Prompting strategy: GENERAL-only delivery cues.

Reusable direction recipe

Stored GENERAL content

Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Carry the full intended delivery through the global direction.

Three local directions — NOT forwarded in GENERAL-onlyP08; neutral speech cues remain

1. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Carry the full intended delivery through the global direction.
2. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Carry the full intended delivery through the global direction.
3. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. 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. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Establish it immediately. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Sustain the delivery. Clear polished narration; controlled breath, smooth phrasing and steady explanatory emphasis. Preserve the intended style through the end. Carry the full intended delivery through the global direction.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.
(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.

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.
(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.
Reference versus no-reference breakdown
reference500.8216.3694.3650.20%0.569
none500.8386.4004.4480.47%n/a