Intelligible intimate whispering · 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 repair-canonical_both at CFG 2.0; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
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
repair-canonical_both · CFG 2.0
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 repair-canonical_both at CFG 2.0; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
0.548
6.104
4.403
1.13%
0.604
99.00%
-0.053
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.) [3.8 seconds duration] We should probably leave before they notice we are missing.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.) [3.5 seconds duration] Although... I am quite happy hiding here with you.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.) [5.0 seconds duration] Give me one more quiet minute, and then I promise we will go.
Literal words in the separate text field:
We should probably leave before they notice we are missing. Although... I am quite happy hiding here with you. Give me one more quiet minute, and then I promise we will go.
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] We should probably leave before they notice we are missing.
(Natural clear speech.) [3.5 seconds duration] Although... I am quite happy hiding here with you.
(Natural clear speech.) [5.0 seconds duration] Give me one more quiet minute, and then I promise we will go.
Reference versus no-reference breakdown
reference
50
0.560
6.081
4.404
0.74%
0.533
none
50
0.536
6.126
4.402
1.51%
n/a
heldout-selected · CFG 2.0
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-selected at CFG 2.0; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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. Same frozen wording, selected CFG versus g1 on unseen texts.
0.507
6.082
4.479
0.37%
0.590
100.00%
-0.048
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.) [3.8 seconds duration] Now... I am not quite sure where I should begin.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.
Literal words in the separate text field:
I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.
(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.
(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.
Reference versus no-reference breakdown
reference
50
0.514
6.063
4.476
0.55%
0.546
none
50
0.500
6.102
4.483
0.20%
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.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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. Same frozen wording, selected CFG versus g1 on unseen texts.
0.310
6.351
4.531
0.78%
0.499
100.00%
0.000
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.) [3.8 seconds duration] Now... I am not quite sure where I should begin.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.
Literal words in the separate text field:
I remember how ordinary everything looked a moment ago. Now... I am not quite sure where I should begin. Give me a little time and I will explain the whole thing.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [3.5 seconds duration] I remember how ordinary everything looked a moment ago.
(Natural clear speech.) [3.8 seconds duration] Now... I am not quite sure where I should begin.
(Natural clear speech.) [4.6 seconds duration] Give me a little time and I will explain the whole thing.
Reference versus no-reference breakdown
reference
50
0.328
6.347
4.511
0.68%
0.594
none
50
0.293
6.355
4.552
0.89%
n/a
g1-p01 · CFG 2.5
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 g1-p01 at CFG 2.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.665
5.568
4.291
2.51%
0.635
98.00%
-0.078
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper focus on the breath entering and leaving the mouth no discernible vocal projection.
Three local directions
1. Focus on the intake of breath a soft almost inaudible hiss.
2. Keep the whisper extremely close to the mic like a secret shared in darkness.
3. Let the breath itself carry the weight of the words not the vocal cords.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper focus on the breath entering and leaving the mouth no discernible vocal projection.
SCRIPT:
(Focus on the intake of breath a soft almost inaudible hiss. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Keep the whisper extremely close to the mic like a secret shared in darkness. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Let the breath itself carry the weight of the words not the vocal cords. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper focus on the breath entering and leaving the mouth no discernible vocal projection.
SCRIPT:
(Focus on the intake of breath a soft almost inaudible hiss.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Keep the whisper extremely close to the mic like a secret shared in darkness.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Let the breath itself carry the weight of the words not the vocal cords.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.672
5.568
4.282
2.44%
0.508
none
50
0.658
5.567
4.300
2.58%
n/a
g1-p07 · CFG 2.5
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 g1-p07 at CFG 2.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.658
5.575
4.314
2.67%
0.631
97.00%
-0.077
Prompting strategy: GENERAL-only delivery cues.
Reusable direction recipe
Stored GENERAL content
Start as a barely audible breathy whisper. Focus on the texture of the air moving past the lips. Maintain this extreme quiet throughout the entire take.
Three local directions — NOT forwarded in GENERAL-onlyP08; neutral speech cues remain
1. Begin with a breathy whisper almost inaudible to the listener.
2. Maintain that fragile breath quality while keeping the consonants incredibly soft and close to the throat.
3. Fade to the absolute edge of audibility keeping the emotional weight present in the stillness.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Start as a barely audible breathy whisper. Focus on the texture of the air moving past the lips. Maintain this extreme quiet throughout the entire take.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Start as a barely audible breathy whisper. Focus on the texture of the air moving past the lips. Maintain this extreme quiet throughout the entire take.
SCRIPT:
(Natural clear speech.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Natural clear speech.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Natural clear speech.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.659
5.545
4.299
1.86%
0.509
none
50
0.657
5.605
4.329
3.47%
n/a
g0-p06 · CFG 3.0
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 3.0; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.636
5.626
4.241
1.71%
0.625
97.00%
-0.144
Prompting strategy: containment contrast.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Start with the feeling held back, then let it escape through the voice.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Start with the feeling held back, then let it escape through the voice.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Start with the feeling held back, then let it escape through the voice.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Start with the feeling held back, then let it escape through the voice.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Start with the feeling held back, then let it escape through the voice.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Start with the feeling held back, then let it escape through the voice. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Start with the feeling held back, then let it escape through the voice. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Start with the feeling held back, then let it escape through the voice. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Start with the feeling held back, then let it escape through the voice.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Start with the feeling held back, then let it escape through the voice.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Start with the feeling held back, then let it escape through the voice.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Start with the feeling held back, then let it escape through the voice.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.634
5.650
4.252
1.56%
0.442
none
50
0.638
5.602
4.230
1.85%
n/a
g0-p09 · CFG 2.5
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.620
5.942
4.393
1.17%
0.634
100.00%
-0.056
Prompting strategy: inline-only delivery cues.
Reusable direction recipe
Stored GENERAL content — NOT forwarded in inline-onlyP09
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Let the local directions alone carry the changing emotion.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Let the local directions alone carry the changing emotion.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Let the local directions alone carry the changing emotion.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Let the local directions alone carry the changing emotion.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Let the local directions alone carry the changing emotion. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Let the local directions alone carry the changing emotion. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Let the local directions alone carry the changing emotion. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Let the local directions alone carry the changing emotion.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Let the local directions alone carry the changing emotion.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Let the local directions alone carry the changing emotion.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.614
5.970
4.398
1.00%
0.530
none
50
0.626
5.914
4.388
1.35%
n/a
g0-p05 · CFG 2.5
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.619
5.755
4.301
1.27%
0.625
99.00%
-0.045
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Build an attack, a sustained emotional middle, and a clear release.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Build an attack, a sustained emotional middle, and a clear release.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Build an attack, a sustained emotional middle, and a clear release.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Build an attack, a sustained emotional middle, and a clear release.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Build an attack, a sustained emotional middle, and a clear release.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Build an attack, a sustained emotional middle, and a clear release. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Build an attack, a sustained emotional middle, and a clear release. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Build an attack, a sustained emotional middle, and a clear release. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Build an attack, a sustained emotional middle, and a clear release.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Build an attack, a sustained emotional middle, and a clear release.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Build an attack, a sustained emotional middle, and a clear release.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Build an attack, a sustained emotional middle, and a clear release.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.621
5.788
4.313
1.14%
0.541
none
50
0.617
5.722
4.289
1.40%
n/a
g0-p08 · CFG 2.0
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 2.0; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.607
5.958
4.419
1.82%
0.625
99.00%
-0.016
Prompting strategy: GENERAL-only delivery cues.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Carry the complete intended performance through the global direction.
Three local directions — NOT forwarded in GENERAL-onlyP08; neutral speech cues remain
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Carry the complete intended performance through the global direction.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Carry the complete intended performance through the global direction.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Carry the complete intended performance 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. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Carry the complete intended performance through the global direction.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Carry the complete intended performance through the global direction.
SCRIPT:
(Natural clear speech.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Natural clear speech.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Natural clear speech.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.596
5.989
4.428
1.47%
0.570
none
50
0.618
5.927
4.409
2.17%
n/a
g0-p04 · CFG 2.0
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 2.0; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.599
6.033
4.434
0.92%
0.630
99.00%
-0.036
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
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 repair-historical_control at CFG 2.0; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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. Predeclared intervention; see exact genome and changed fields.
0.599
6.033
4.434
0.92%
0.630
99.00%
-0.036
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list. [3.8 seconds duration]) We should probably leave before they notice we are missing.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list. [3.5 seconds duration]) Although... I am quite happy hiding here with you.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list. [5.0 seconds duration]) Give me one more quiet minute, and then I promise we will go.
Literal words in the separate text field:
We should probably leave before they notice we are missing. Although... I am quite happy hiding here with you. Give me one more quiet minute, and then I promise we will go.
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]) We should probably leave before they notice we are missing.
(Natural clear speech. [3.5 seconds duration]) Although... I am quite happy hiding here with you.
(Natural clear speech. [5.0 seconds duration]) Give me one more quiet minute, and then I promise we will go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Speak organically and spontaneously, thinking while speaking, never reading a list.) [3.8 seconds duration] We should probably leave before they notice we are missing.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Speak organically and spontaneously, thinking while speaking, never reading a list.) [3.5 seconds duration] Although... I am quite happy hiding here with you.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Speak organically and spontaneously, thinking while speaking, never reading a list.) [5.0 seconds duration] Give me one more quiet minute, and then I promise we will go.
Reference versus no-reference breakdown
reference
50
0.600
6.018
4.425
0.82%
0.550
none
50
0.599
6.048
4.443
1.02%
n/a
g1-p06 · CFG 2
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 g1-p06 at CFG 2; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.597
6.048
4.468
1.42%
0.627
100.00%
-0.011
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
The lowest possible register whisper. The emotion should be deeply internal, felt more than heard. Absolute restraint on volume.
Three local directions
1. Near-silent whisper. Focus on the internal resonance of the sound, not projection.
2. Maintain the lowest audible level. Consonants dissolve into the breath texture.
3. Emotionally present but physically minimal. A sustained, almost imperceptible murmur.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. The lowest possible register whisper. The emotion should be deeply internal, felt more than heard. Absolute restraint on volume.
SCRIPT:
(Near-silent whisper. Focus on the internal resonance of the sound, not projection. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Maintain the lowest audible level. Consonants dissolve into the breath texture. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Emotionally present but physically minimal. A sustained, almost imperceptible murmur. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. The lowest possible register whisper. The emotion should be deeply internal, felt more than heard. Absolute restraint on volume.
SCRIPT:
(Near-silent whisper. Focus on the internal resonance of the sound, not projection.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Maintain the lowest audible level. Consonants dissolve into the breath texture.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Emotionally present but physically minimal. A sustained, almost imperceptible murmur.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.602
5.974
4.414
1.53%
0.575
none
50
0.591
6.121
4.522
1.32%
n/a
g0-p03 · CFG 1.5
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.573
6.081
4.468
1.42%
0.616
98.00%
0.001
Prompting strategy: situation and intention.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. You urgently need an unheard listener to understand what just happened.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. You urgently need an unheard listener to understand what just happened.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. You urgently need an unheard listener to understand what just happened.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. You urgently need an unheard listener to understand what just happened.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. You urgently need an unheard listener to understand what just happened.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. You urgently need an unheard listener to understand what just happened. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. You urgently need an unheard listener to understand what just happened. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. You urgently need an unheard listener to understand what just happened. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. You urgently need an unheard listener to understand what just happened.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. You urgently need an unheard listener to understand what just happened.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. You urgently need an unheard listener to understand what just happened.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. You urgently need an unheard listener to understand what just happened.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.579
6.087
4.441
1.58%
0.587
none
50
0.567
6.075
4.495
1.27%
n/a
g0-p07 · CFG 1.5
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.543
6.137
4.471
1.38%
0.603
99.00%
-0.019
Prompting strategy: reacting to an unheard partner.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. React to a silent person right beside you; generate only one voice.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. React to a silent person right beside you; generate only one voice.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. React to a silent person right beside you; generate only one voice.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. React to a silent person right beside you; generate only one voice.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. React to a silent person right beside you; generate only one voice.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. React to a silent person right beside you; generate only one voice. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. React to a silent person right beside you; generate only one voice. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. React to a silent person right beside you; generate only one voice. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. React to a silent person right beside you; generate only one voice.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. React to a silent person right beside you; generate only one voice.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. React to a silent person right beside you; generate only one voice.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. React to a silent person right beside you; generate only one voice.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.547
6.148
4.469
1.23%
0.567
none
50
0.539
6.125
4.472
1.53%
n/a
g1-p04 · CFG 1.25
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 g1-p04 at CFG 1.25; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.516
6.182
4.523
1.53%
0.592
99.00%
0.006
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper focusing on the physical closeness of the breath. The delivery should feel like a secret shared directly into the ear canal.
Three local directions
1. An extremely quiet intimate whisper with audible intake of breath. No projection.
2. Maintain the near-silent intimacy. Consonants must be barely breathed out.
3. Sustain the barely audible presence. Emotional weight carried entirely by the breath texture.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper focusing on the physical closeness of the breath. The delivery should feel like a secret shared directly into the ear canal.
SCRIPT:
(An extremely quiet intimate whisper with audible intake of breath. No projection. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Maintain the near-silent intimacy. Consonants must be barely breathed out. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Sustain the barely audible presence. Emotional weight carried entirely by the breath texture. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper focusing on the physical closeness of the breath. The delivery should feel like a secret shared directly into the ear canal.
SCRIPT:
(An extremely quiet intimate whisper with audible intake of breath. No projection.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Maintain the near-silent intimacy. Consonants must be barely breathed out.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Sustain the barely audible presence. Emotional weight carried entirely by the breath texture.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.526
6.197
4.509
1.34%
0.592
none
50
0.507
6.167
4.536
1.72%
n/a
g0-p02 · CFG 1.25
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.508
6.170
4.488
1.57%
0.586
100.00%
0.016
Prompting strategy: audible mechanics.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Make breath pressure, pitch motion, onset and release physically audible.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Make breath pressure, pitch motion, onset and release physically audible.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Make breath pressure, pitch motion, onset and release physically audible.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Make breath pressure, pitch motion, onset and release physically audible.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Make breath pressure, pitch motion, onset and release physically audible.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Make breath pressure, pitch motion, onset and release physically audible. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Make breath pressure, pitch motion, onset and release physically audible. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Make breath pressure, pitch motion, onset and release physically audible. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Make breath pressure, pitch motion, onset and release physically audible.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Make breath pressure, pitch motion, onset and release physically audible.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Make breath pressure, pitch motion, onset and release physically audible.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Make breath pressure, pitch motion, onset and release physically audible.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.529
6.151
4.441
1.21%
0.602
none
50
0.486
6.189
4.535
1.94%
n/a
g1-p03 · CFG 1.25
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 g1-p03 at CFG 1.25; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.468
6.262
4.543
1.92%
0.569
99.00%
0.012
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
Remain a near-silent whisper the emotional presence must be palpable without increasing volume maintain absolute control.
Three local directions
1. Every syllable must feel heavy with meaning despite the lack of volume.
2. Hold the intimacy steady do not let the whisper drift into mere breathiness.
3. Deliver the final words with complete emotional weight but zero volume increase.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Remain a near-silent whisper the emotional presence must be palpable without increasing volume maintain absolute control.
SCRIPT:
(Every syllable must feel heavy with meaning despite the lack of volume. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Hold the intimacy steady do not let the whisper drift into mere breathiness. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Deliver the final words with complete emotional weight but zero volume increase. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Remain a near-silent whisper the emotional presence must be palpable without increasing volume maintain absolute control.
SCRIPT:
(Every syllable must feel heavy with meaning despite the lack of volume.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Hold the intimacy steady do not let the whisper drift into mere breathiness.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Deliver the final words with complete emotional weight but zero volume increase.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.478
6.246
4.511
2.14%
0.598
none
50
0.458
6.278
4.575
1.70%
n/a
g1-p05 · CFG 1.5
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 g1-p05 at CFG 1.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.459
6.256
4.555
1.33%
0.569
99.00%
0.025
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
A confidential whisper with a very delicate, almost hesitant rhythm. Focus on making the consonants feel fragile and intentional.
Three local directions
1. Whisper with a soft, staccato rhythm. Each word is a careful release of air.
2. Confidential tone. Consonants are soft puffs, not distinct sounds. Spontaneity in timing.
3. Intelligibility maintained through subtle pitch shifts, not volume. A quiet, knowing presence.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A confidential whisper with a very delicate, almost hesitant rhythm. Focus on making the consonants feel fragile and intentional.
SCRIPT:
(Whisper with a soft, staccato rhythm. Each word is a careful release of air. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Confidential tone. Consonants are soft puffs, not distinct sounds. Spontaneity in timing. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Intelligibility maintained through subtle pitch shifts, not volume. A quiet, knowing presence. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A confidential whisper with a very delicate, almost hesitant rhythm. Focus on making the consonants feel fragile and intentional.
SCRIPT:
(Whisper with a soft, staccato rhythm. Each word is a careful release of air.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Confidential tone. Consonants are soft puffs, not distinct sounds. Spontaneity in timing.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Intelligibility maintained through subtle pitch shifts, not volume. A quiet, knowing presence.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.476
6.227
4.533
1.05%
0.611
none
50
0.442
6.285
4.576
1.60%
n/a
g0-p01 · CFG 1.0
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 1.0; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.425
6.290
4.519
1.13%
0.552
100.00%
0.000
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Deliver the intended emotion unmistakably, without adding words.
Three local directions
1. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Deliver the intended emotion unmistakably, without adding words.
2. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Deliver the intended emotion unmistakably, without adding words.
3. Remain a near-silent whisper, intelligible and emotionally present, never louder. Deliver the intended emotion unmistakably, without adding words.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Deliver the intended emotion unmistakably, without adding words.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Deliver the intended emotion unmistakably, without adding words. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Deliver the intended emotion unmistakably, without adding words. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Deliver the intended emotion unmistakably, without adding words. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. An extremely quiet intimate whisper, audible breath and no projected speaking voice. Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Remain a near-silent whisper, intelligible and emotionally present, never louder. Deliver the intended emotion unmistakably, without adding words.
SCRIPT:
(An extremely quiet intimate whisper, audible breath and no projected speaking voice. Deliver the intended emotion unmistakably, without adding words.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Continue whispering clearly, with soft consonants and a confidential spontaneous rhythm. Deliver the intended emotion unmistakably, without adding words.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Remain a near-silent whisper, intelligible and emotionally present, never louder. Deliver the intended emotion unmistakably, without adding words.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.436
6.278
4.495
1.11%
0.586
none
50
0.414
6.302
4.544
1.15%
n/a
g1-p02 · CFG 1.5
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 g1-p02 at CFG 1.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.410
6.311
4.563
1.12%
0.547
100.00%
0.020
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
Continue whispering with pronounced soft consonants a delicate conversational cadence as if sharing a fleeting thought.
Three local directions
1. Roll off the hard sounds make the S and T sounds airy and soft.
2. Maintain a spontaneous rhythm like a natural thought bubble not rehearsed.
3. The whisper must sound like a private confidence a shared moment between two people.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Continue whispering with pronounced soft consonants a delicate conversational cadence as if sharing a fleeting thought.
SCRIPT:
(Roll off the hard sounds make the S and T sounds airy and soft. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Maintain a spontaneous rhythm like a natural thought bubble not rehearsed. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(The whisper must sound like a private confidence a shared moment between two people. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Continue whispering with pronounced soft consonants a delicate conversational cadence as if sharing a fleeting thought.
SCRIPT:
(Roll off the hard sounds make the S and T sounds airy and soft.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Maintain a spontaneous rhythm like a natural thought bubble not rehearsed.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(The whisper must sound like a private confidence a shared moment between two people.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.409
6.316
4.530
0.75%
0.606
none
50
0.412
6.306
4.596
1.48%
n/a
g1-p08 · CFG 1.5
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 g1-p08 at CFG 1.5; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.336
6.335
4.533
2.02%
0.505
98.00%
0.021
Prompting strategy: GENERAL-only delivery cues.
Reusable direction recipe
Stored GENERAL content
Deliver the lines in a conspiratorial, hushed tone. Emphasize the secrets in the phrasing. The rhythm should feel like a shared confidence.
Three local directions — NOT forwarded in GENERAL-onlyP08; neutral speech cues remain
1. A very quiet whisper conveying deep intimacy and shared knowledge.
2. Let the rhythm be slightly hesitant like someone afraid to be overheard.
3. Keep the vocal energy extremely low a near secret shared between two people.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Deliver the lines in a conspiratorial, hushed tone. Emphasize the secrets in the phrasing. The rhythm should feel like a shared confidence.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Deliver the lines in a conspiratorial, hushed tone. Emphasize the secrets in the phrasing. The rhythm should feel like a shared confidence.
SCRIPT:
(Natural clear speech.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Natural clear speech.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Natural clear speech.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.
Reference versus no-reference breakdown
reference
50
0.341
6.310
4.470
2.47%
0.607
none
50
0.331
6.360
4.595
1.57%
n/a
g1-p09 · CFG 1.25
pilot-v1HISTORICAL_NONCANONICAL_TIMING_SYNTAX
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 g1-p09 at CFG 1.25; every recipe also shows its own values below. These are recipe means, not one hand-picked audio.
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_WHIS_reg. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 1.767) / 1.599; 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.
Passes legacy speech/identity gates only; burst reliability is not established. N=100 across both languages, both reference arms and all five seeds. Joint prompt+CFG candidate, not an isolated optimal-CFG estimate.
0.235
6.370
4.469
1.64%
0.456
99.00%
0.030
Prompting strategy: GENERAL-only delivery cues.
Reusable direction recipe
Stored GENERAL content
Treat this as a near-silent thought spoken aloud. The focus is on clarity despite the minimal volume. Intention must override projection.
Three local directions — NOT forwarded in GENERAL-onlyP08; neutral speech cues remain
1. Speak at the lowest possible whisper level ensuring every word is discernible.
2. Use precise but gentle articulation for a clear confidential delivery.
3. Hold the emotional resonance without ever increasing the physical volume.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Treat this as a near-silent thought spoken aloud. The focus is on clarity despite the minimal volume. Intention must override projection.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Literal words in the separate text field:
Come a little closer, because I do not want to wake anyone. I think... I finally understand what you were trying to tell me. Let us sit here quietly for another minute before we go.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) Come a little closer, because I do not want to wake anyone.
(Natural clear speech. [4.6 seconds duration]) I think... I finally understand what you were trying to tell me.
(Natural clear speech. [4.2 seconds duration]) Let us sit here quietly for another minute before we go.
Canonical duration-layout rewrite — not itself an observed legacy result
This fixes the bracket placement only. It is not evidence that the corrected prompt achieves the legacy scores. See the separately scored schema-repair experiment.
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Treat this as a near-silent thought spoken aloud. The focus is on clarity despite the minimal volume. Intention must override projection.
SCRIPT:
(Natural clear speech.) [4.6 seconds duration] Come a little closer, because I do not want to wake anyone.
(Natural clear speech.) [4.6 seconds duration] I think... I finally understand what you were trying to tell me.
(Natural clear speech.) [4.2 seconds duration] Let us sit here quietly for another minute before we go.