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 2.5; 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.
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
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 2.5
canonical-v1-explorationCANONICAL
Current ranking: EARS Steering v2
Emotion/style only. Content Enjoyment: 0%. Genuineness: 0%. These remain visible diagnostics, not reward terms. WER and speaker identity are separate qualification gates, not terms in this score.
Observed head means below refer to repair-canonical_both 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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.902
6.393
3.986
0.09%
0.824
100.00%
-0.031
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
Three local directions
1. Deliver the first line with icy, deliberate contempt. Control the breath meticulously.
2. Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.
3. The conclusion must be a statement of absolute, unwavering indignation, utterly serious.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
SCRIPT:
(Deliver the first line with icy, deliberate contempt. Control the breath meticulously.) [3.8 seconds duration] You promised me you would tell the truth this time.
(Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.) [4.2 seconds duration] And now... you are standing here pretending none of it happened.
(The conclusion must be a statement of absolute, unwavering indignation, utterly serious.) [5.0 seconds duration] Do not ask me to calm down when you will not even listen.
Literal words in the separate text field:
You promised me you would tell the truth this time. And now... you are standing here pretending none of it happened. Do not ask me to calm down when you will not even listen.
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] You promised me you would tell the truth this time.
(Natural clear speech.) [4.2 seconds duration] And now... you are standing here pretending none of it happened.
(Natural clear speech.) [5.0 seconds duration] Do not ask me to calm down when you will not even listen.
Reference versus no-reference breakdown
reference
50
0.907
6.367
3.954
0.17%
0.608
none
50
0.896
6.419
4.018
0.00%
n/a
heldout-selected · CFG 2.5
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 2.5; 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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.541
6.375
4.125
0.37%
0.617
100.00%
-0.033
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
Three local directions
1. Deliver the first line with icy, deliberate contempt. Control the breath meticulously.
2. Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.
3. The conclusion must be a statement of absolute, unwavering indignation, utterly serious.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
SCRIPT:
(Deliver the first line with icy, deliberate contempt. Control the breath meticulously.) [3.8 seconds duration] I was certain that I knew what would happen next.
(Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.) [3.1 seconds duration] And then... everything felt different for a second.
(The conclusion must be a statement of absolute, unwavering indignation, utterly serious.) [4.2 seconds duration] Look at me and tell me you understand what I mean.
Literal words in the separate text field:
I was certain that I knew what would happen next. And then... everything felt different for a second. Look at me and tell me you understand what I mean.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [3.8 seconds duration] I was certain that I knew what would happen next.
(Natural clear speech.) [3.1 seconds duration] And then... everything felt different for a second.
(Natural clear speech.) [4.2 seconds duration] Look at me and tell me you understand what I mean.
Reference versus no-reference breakdown
reference
50
0.556
6.359
4.069
0.47%
0.583
none
50
0.525
6.391
4.181
0.28%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.332
6.402
4.277
0.51%
0.511
100.00%
0.000
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
Three local directions
1. Deliver the first line with icy, deliberate contempt. Control the breath meticulously.
2. Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.
3. The conclusion must be a statement of absolute, unwavering indignation, utterly serious.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
SCRIPT:
(Deliver the first line with icy, deliberate contempt. Control the breath meticulously.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(The conclusion must be a statement of absolute, unwavering indignation, utterly serious.) [3.5 seconds duration] Stay with me until I can say this properly.
Literal words in the separate text field:
I thought you would be here before the lights went out. Wait... I need a moment to understand what just happened. Stay with me until I can say this properly.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech.) [4.2 seconds duration] I thought you would be here before the lights went out.
(Natural clear speech.) [3.8 seconds duration] Wait... I need a moment to understand what just happened.
(Natural clear speech.) [3.5 seconds duration] Stay with me until I can say this properly.
Reference versus no-reference breakdown
reference
50
0.339
6.380
4.246
0.69%
0.616
none
50
0.326
6.425
4.308
0.34%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.981
6.055
3.785
1.39%
0.798
99.00%
-0.022
Prompting strategy: containment contrast.
Reusable direction recipe
Stored GENERAL content
Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Start with the feeling held back, then let it escape through the voice.
Three local directions
1. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Start with the feeling held back, then let it escape through the voice.
2. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Start with the feeling held back, then let it escape through the voice.
3. Bitter indignation remains audible as you insist on being taken seriously. 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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Start with the feeling held back, then let it escape through the voice.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Start with the feeling held back, then let it escape through the voice. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Start with the feeling held back, then let it escape through the voice. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Bitter indignation remains audible as you insist on being taken seriously. Start with the feeling held back, then let it escape through the voice. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
Literal words in the separate text field:
I have spent the entire morning fixing what you broke. You knew exactly what you were doing... and you did it anyway. For once in your life, take responsibility for the mess you made.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(Natural clear speech. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Natural clear speech. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Start with the feeling held back, then let it escape through the voice.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Start with the feeling held back, then let it escape through the voice.) [3.8 seconds duration] I have spent the entire morning fixing what you broke.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Start with the feeling held back, then let it escape through the voice.) [4.6 seconds duration] You knew exactly what you were doing... and you did it anyway.
(Bitter indignation remains audible as you insist on being taken seriously. Start with the feeling held back, then let it escape through the voice.) [4.6 seconds duration] For once in your life, take responsibility for the mess you made.
Reference versus no-reference breakdown
reference
50
0.983
6.020
3.767
0.69%
0.617
none
50
0.979
6.091
3.803
2.08%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.969
6.143
3.844
1.00%
0.806
98.00%
-0.016
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Build an attack, a sustained emotional middle, and a clear release.
Three local directions
1. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Build an attack, a sustained emotional middle, and a clear release.
2. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Build an attack, a sustained emotional middle, and a clear release.
3. Bitter indignation remains audible as you insist on being taken seriously. 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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Build an attack, a sustained emotional middle, and a clear release.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Build an attack, a sustained emotional middle, and a clear release. [4.6 seconds duration]) I asked for one simple thing, and you laughed in my face.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Build an attack, a sustained emotional middle, and a clear release. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Bitter indignation remains audible as you insist on being taken seriously. Build an attack, a sustained emotional middle, and a clear release. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
Literal words in the separate text field:
I asked for one simple thing, and you laughed in my face. No... this is not another little misunderstanding you can wave away. Look at me and explain why you thought that was acceptable.
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]) I asked for one simple thing, and you laughed in my face.
(Natural clear speech. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Natural clear speech. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Build an attack, a sustained emotional middle, and a clear release.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Build an attack, a sustained emotional middle, and a clear release.) [4.6 seconds duration] I asked for one simple thing, and you laughed in my face.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Build an attack, a sustained emotional middle, and a clear release.) [4.2 seconds duration] No... this is not another little misunderstanding you can wave away.
(Bitter indignation remains audible as you insist on being taken seriously. Build an attack, a sustained emotional middle, and a clear release.) [4.2 seconds duration] Look at me and explain why you thought that was acceptable.
Reference versus no-reference breakdown
reference
50
0.976
6.120
3.793
0.81%
0.623
none
50
0.962
6.166
3.895
1.20%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.952
6.098
3.894
4.47%
0.778
90.00%
-0.040
Prompting strategy: inline-only delivery cues.
Reusable direction recipe
Stored GENERAL content — NOT forwarded in inline-onlyP09
Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Let the local directions alone carry the changing emotion.
Three local directions
1. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Let the local directions alone carry the changing emotion.
2. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Let the local directions alone carry the changing emotion.
3. Bitter indignation remains audible as you insist on being taken seriously. 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:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Let the local directions alone carry the changing emotion. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Let the local directions alone carry the changing emotion. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Bitter indignation remains audible as you insist on being taken seriously. Let the local directions alone carry the changing emotion. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
Literal words in the separate text field:
I have spent the entire morning fixing what you broke. You knew exactly what you were doing... and you did it anyway. For once in your life, take responsibility for the mess you made.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(Natural clear speech. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Natural clear speech. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
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:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Let the local directions alone carry the changing emotion.) [3.8 seconds duration] I have spent the entire morning fixing what you broke.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Let the local directions alone carry the changing emotion.) [4.6 seconds duration] You knew exactly what you were doing... and you did it anyway.
(Bitter indignation remains audible as you insist on being taken seriously. Let the local directions alone carry the changing emotion.) [4.6 seconds duration] For once in your life, take responsibility for the mess you made.
Reference versus no-reference breakdown
reference
50
0.956
6.089
3.859
3.58%
0.599
none
50
0.948
6.107
3.929
5.37%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.938
6.246
3.944
0.74%
0.814
99.00%
-0.001
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Speak organically and spontaneously, thinking while speaking, never reading a list.
Three local directions
1. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Speak organically and spontaneously, thinking while speaking, never reading a list.
2. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Speak organically and spontaneously, thinking while speaking, never reading a list.
3. Bitter indignation remains audible as you insist on being taken seriously. 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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Speak organically and spontaneously, thinking while speaking, never reading a list.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Speak organically and spontaneously, thinking while speaking, never reading a list. [3.8 seconds duration]) You promised me you would tell the truth this time.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Speak organically and spontaneously, thinking while speaking, never reading a list. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(Bitter indignation remains audible as you insist on being taken seriously. Speak organically and spontaneously, thinking while speaking, never reading a list. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
Literal words in the separate text field:
You promised me you would tell the truth this time. And now... you are standing here pretending none of it happened. Do not ask me to calm down when you will not even listen.
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]) You promised me you would tell the truth this time.
(Natural clear speech. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(Natural clear speech. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Speak organically and spontaneously, thinking while speaking, never reading a list.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Speak organically and spontaneously, thinking while speaking, never reading a list.) [3.8 seconds duration] You promised me you would tell the truth this time.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Speak organically and spontaneously, thinking while speaking, never reading a list.) [4.2 seconds duration] And now... you are standing here pretending none of it happened.
(Bitter indignation remains audible as you insist on being taken seriously. Speak organically and spontaneously, thinking while speaking, never reading a list.) [5.0 seconds duration] Do not ask me to calm down when you will not even listen.
Reference versus no-reference breakdown
reference
50
0.947
6.201
3.890
0.96%
0.638
none
50
0.930
6.290
3.998
0.53%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.918
6.230
4.080
3.19%
0.799
94.00%
-0.044
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
Start with a sudden burst of raw, physical anger. Project the frustration from your diaphragm, sharp and clipped. Let the breath catch before the outburst.
Three local directions
1. Physicality of sudden explosive rage. Sharp attack on consonants. Audible strained inhale before speaking.
2. Sustained high-intensity vocal tension. Every word pushed forward with physical force, like throwing something.
3. A ragged, desperate quality beneath the anger. The insistence must sound physically tiring.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Start with a sudden burst of raw, physical anger. Project the frustration from your diaphragm, sharp and clipped. Let the breath catch before the outburst.
SCRIPT:
(Physicality of sudden explosive rage. Sharp attack on consonants. Audible strained inhale before speaking. [3.8 seconds duration]) You promised me you would tell the truth this time.
(Sustained high-intensity vocal tension. Every word pushed forward with physical force, like throwing something. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(A ragged, desperate quality beneath the anger. The insistence must sound physically tiring. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
Literal words in the separate text field:
You promised me you would tell the truth this time. And now... you are standing here pretending none of it happened. Do not ask me to calm down when you will not even listen.
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]) You promised me you would tell the truth this time.
(Natural clear speech. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(Natural clear speech. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
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 with a sudden burst of raw, physical anger. Project the frustration from your diaphragm, sharp and clipped. Let the breath catch before the outburst.
SCRIPT:
(Physicality of sudden explosive rage. Sharp attack on consonants. Audible strained inhale before speaking.) [3.8 seconds duration] You promised me you would tell the truth this time.
(Sustained high-intensity vocal tension. Every word pushed forward with physical force, like throwing something.) [4.2 seconds duration] And now... you are standing here pretending none of it happened.
(A ragged, desperate quality beneath the anger. The insistence must sound physically tiring.) [5.0 seconds duration] Do not ask me to calm down when you will not even listen.
Reference versus no-reference breakdown
reference
50
0.916
6.210
4.060
3.97%
0.595
none
50
0.919
6.250
4.099
2.41%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.917
6.266
3.980
2.44%
0.801
96.00%
-0.025
Prompting strategy: GENERAL-only delivery cues.
Reusable direction recipe
Stored GENERAL content
Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Carry the complete intended performance through the global direction.
Three local directions — NOT forwarded in GENERAL-onlyP08; neutral speech cues remain
1. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Carry the complete intended performance through the global direction.
2. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Carry the complete intended performance through the global direction.
3. Bitter indignation remains audible as you insist on being taken seriously. 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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Carry the complete intended performance through the global direction.
SCRIPT:
(Natural clear speech. [4.6 seconds duration]) I asked for one simple thing, and you laughed in my face.
(Natural clear speech. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Natural clear speech. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
Literal words in the separate text field:
I asked for one simple thing, and you laughed in my face. No... this is not another little misunderstanding you can wave away. Look at me and explain why you thought that was acceptable.
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]) I asked for one simple thing, and you laughed in my face.
(Natural clear speech. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Natural clear speech. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Carry the complete intended performance through the global direction.
SCRIPT:
(Natural clear speech.) [4.6 seconds duration] I asked for one simple thing, and you laughed in my face.
(Natural clear speech.) [4.2 seconds duration] No... this is not another little misunderstanding you can wave away.
(Natural clear speech.) [4.2 seconds duration] Look at me and explain why you thought that was acceptable.
Reference versus no-reference breakdown
reference
50
0.929
6.234
3.910
1.26%
0.614
none
50
0.905
6.299
4.049
3.63%
n/a
g1-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 g1-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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.900
6.247
4.072
5.77%
0.772
89.00%
0.005
Prompting strategy: reacting to an unheard partner.
Reusable direction recipe
Stored GENERAL content
A sudden burst of explosive anger that cracks the air, sharp intakes of breath punctuating each stressed word.
Three local directions
1. Start with a sudden, sharp intake of breath before the first word
2. Increase vocal volume sharply on key stressed consonants
3. End the rant with a ragged, forceful exhale
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A sudden burst of explosive anger that cracks the air, sharp intakes of breath punctuating each stressed word.
SCRIPT:
(Start with a sudden, sharp intake of breath before the first word [3.8 seconds duration]) You promised me you would tell the truth this time.
(Increase vocal volume sharply on key stressed consonants [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(End the rant with a ragged, forceful exhale [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
Literal words in the separate text field:
You promised me you would tell the truth this time. And now... you are standing here pretending none of it happened. Do not ask me to calm down when you will not even listen.
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]) You promised me you would tell the truth this time.
(Natural clear speech. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(Natural clear speech. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
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 sudden burst of explosive anger that cracks the air, sharp intakes of breath punctuating each stressed word.
SCRIPT:
(Start with a sudden, sharp intake of breath before the first word) [3.8 seconds duration] You promised me you would tell the truth this time.
(Increase vocal volume sharply on key stressed consonants) [4.2 seconds duration] And now... you are standing here pretending none of it happened.
(End the rant with a ragged, forceful exhale) [5.0 seconds duration] Do not ask me to calm down when you will not even listen.
Reference versus no-reference breakdown
reference
50
0.915
6.213
4.023
3.26%
0.644
none
50
0.885
6.282
4.121
8.28%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.888
6.296
4.050
1.55%
0.811
97.00%
0.008
Prompting strategy: reacting to an unheard partner.
Reusable direction recipe
Stored GENERAL content
Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. React to a silent person right beside you; generate only one voice.
Three local directions
1. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. React to a silent person right beside you; generate only one voice.
2. The anger escalates into a controlled furious confrontation, not an unintelligible scream. React to a silent person right beside you; generate only one voice.
3. Bitter indignation remains audible as you insist on being taken seriously. 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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. React to a silent person right beside you; generate only one voice.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. React to a silent person right beside you; generate only one voice. [3.8 seconds duration]) You promised me you would tell the truth this time.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. React to a silent person right beside you; generate only one voice. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(Bitter indignation remains audible as you insist on being taken seriously. React to a silent person right beside you; generate only one voice. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
Literal words in the separate text field:
You promised me you would tell the truth this time. And now... you are standing here pretending none of it happened. Do not ask me to calm down when you will not even listen.
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]) You promised me you would tell the truth this time.
(Natural clear speech. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(Natural clear speech. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. React to a silent person right beside you; generate only one voice.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. React to a silent person right beside you; generate only one voice.) [3.8 seconds duration] You promised me you would tell the truth this time.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. React to a silent person right beside you; generate only one voice.) [4.2 seconds duration] And now... you are standing here pretending none of it happened.
(Bitter indignation remains audible as you insist on being taken seriously. React to a silent person right beside you; generate only one voice.) [5.0 seconds duration] Do not ask me to calm down when you will not even listen.
Reference versus no-reference breakdown
reference
50
0.909
6.273
3.991
1.47%
0.647
none
50
0.868
6.318
4.110
1.63%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.886
6.279
4.063
2.88%
0.798
92.00%
-0.002
Prompting strategy: situation and intention.
Reusable direction recipe
Stored GENERAL content
Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. You urgently need an unheard listener to understand what just happened.
Three local directions
1. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. You urgently need an unheard listener to understand what just happened.
2. The anger escalates into a controlled furious confrontation, not an unintelligible scream. You urgently need an unheard listener to understand what just happened.
3. Bitter indignation remains audible as you insist on being taken seriously. 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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. You urgently need an unheard listener to understand what just happened.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. You urgently need an unheard listener to understand what just happened. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. You urgently need an unheard listener to understand what just happened. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Bitter indignation remains audible as you insist on being taken seriously. You urgently need an unheard listener to understand what just happened. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
Literal words in the separate text field:
I have spent the entire morning fixing what you broke. You knew exactly what you were doing... and you did it anyway. For once in your life, take responsibility for the mess you made.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(Natural clear speech. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Natural clear speech. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. You urgently need an unheard listener to understand what just happened.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. You urgently need an unheard listener to understand what just happened.) [3.8 seconds duration] I have spent the entire morning fixing what you broke.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. You urgently need an unheard listener to understand what just happened.) [4.6 seconds duration] You knew exactly what you were doing... and you did it anyway.
(Bitter indignation remains audible as you insist on being taken seriously. You urgently need an unheard listener to understand what just happened.) [4.6 seconds duration] For once in your life, take responsibility for the mess you made.
Reference versus no-reference breakdown
reference
50
0.906
6.231
3.991
2.12%
0.637
none
50
0.866
6.326
4.136
3.63%
n/a
g1-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 g1-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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.877
6.355
4.118
0.55%
0.822
100.00%
-0.024
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
Three local directions
1. Deliver the first line with icy, deliberate contempt. Control the breath meticulously.
2. Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.
3. The conclusion must be a statement of absolute, unwavering indignation, utterly serious.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
SCRIPT:
(Deliver the first line with icy, deliberate contempt. Control the breath meticulously. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(Escalate the pace and pitch slightly, injecting venom into each syllable without losing control. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(The conclusion must be a statement of absolute, unwavering indignation, utterly serious. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
Literal words in the separate text field:
I have spent the entire morning fixing what you broke. You knew exactly what you were doing... and you did it anyway. For once in your life, take responsibility for the mess you made.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(Natural clear speech. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Natural clear speech. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
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 like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
SCRIPT:
(Deliver the first line with icy, deliberate contempt. Control the breath meticulously.) [3.8 seconds duration] I have spent the entire morning fixing what you broke.
(Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.) [4.6 seconds duration] You knew exactly what you were doing... and you did it anyway.
(The conclusion must be a statement of absolute, unwavering indignation, utterly serious.) [4.6 seconds duration] For once in your life, take responsibility for the mess you made.
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 2.5; 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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.877
6.355
4.118
0.55%
0.822
100.00%
-0.024
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
Three local directions
1. Deliver the first line with icy, deliberate contempt. Control the breath meticulously.
2. Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.
3. The conclusion must be a statement of absolute, unwavering indignation, utterly serious.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Treat this like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
SCRIPT:
(Deliver the first line with icy, deliberate contempt. Control the breath meticulously. [4.6 seconds duration]) I asked for one simple thing, and you laughed in my face.
(Escalate the pace and pitch slightly, injecting venom into each syllable without losing control. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(The conclusion must be a statement of absolute, unwavering indignation, utterly serious. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
Literal words in the separate text field:
I asked for one simple thing, and you laughed in my face. No... this is not another little misunderstanding you can wave away. Look at me and explain why you thought that was acceptable.
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]) I asked for one simple thing, and you laughed in my face.
(Natural clear speech. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Natural clear speech. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
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 like a calculated verbal assault. Every word must be aimed, precise, and heavy with contempt. The anger is intellectualized rage, delivered with absolute conviction.
SCRIPT:
(Deliver the first line with icy, deliberate contempt. Control the breath meticulously.) [4.6 seconds duration] I asked for one simple thing, and you laughed in my face.
(Escalate the pace and pitch slightly, injecting venom into each syllable without losing control.) [4.2 seconds duration] No... this is not another little misunderstanding you can wave away.
(The conclusion must be a statement of absolute, unwavering indignation, utterly serious.) [4.2 seconds duration] Look at me and explain why you thought that was acceptable.
Reference versus no-reference breakdown
reference
50
0.887
6.326
4.081
0.30%
0.615
none
50
0.866
6.385
4.156
0.80%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.854
6.319
4.131
2.12%
0.804
95.00%
0.013
Prompting strategy: audible mechanics.
Reusable direction recipe
Stored GENERAL content
Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Make breath pressure, pitch motion, onset and release physically audible.
Three local directions
1. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Make breath pressure, pitch motion, onset and release physically audible.
2. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Make breath pressure, pitch motion, onset and release physically audible.
3. Bitter indignation remains audible as you insist on being taken seriously. 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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Make breath pressure, pitch motion, onset and release physically audible.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Make breath pressure, pitch motion, onset and release physically audible. [4.6 seconds duration]) I asked for one simple thing, and you laughed in my face.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Make breath pressure, pitch motion, onset and release physically audible. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Bitter indignation remains audible as you insist on being taken seriously. Make breath pressure, pitch motion, onset and release physically audible. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
Literal words in the separate text field:
I asked for one simple thing, and you laughed in my face. No... this is not another little misunderstanding you can wave away. Look at me and explain why you thought that was acceptable.
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]) I asked for one simple thing, and you laughed in my face.
(Natural clear speech. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Natural clear speech. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Make breath pressure, pitch motion, onset and release physically audible.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Make breath pressure, pitch motion, onset and release physically audible.) [4.6 seconds duration] I asked for one simple thing, and you laughed in my face.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Make breath pressure, pitch motion, onset and release physically audible.) [4.2 seconds duration] No... this is not another little misunderstanding you can wave away.
(Bitter indignation remains audible as you insist on being taken seriously. Make breath pressure, pitch motion, onset and release physically audible.) [4.2 seconds duration] Look at me and explain why you thought that was acceptable.
Reference versus no-reference breakdown
reference
50
0.870
6.285
4.073
1.96%
0.652
none
50
0.839
6.354
4.189
2.28%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.828
6.372
4.153
1.40%
0.803
98.00%
0.000
Prompting strategy: concise trained baseline.
Reusable direction recipe
Stored GENERAL content
Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Deliver the intended emotion unmistakably, without adding words.
Three local directions
1. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Deliver the intended emotion unmistakably, without adding words.
2. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Deliver the intended emotion unmistakably, without adding words.
3. Bitter indignation remains audible as you insist on being taken seriously. 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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Deliver the intended emotion unmistakably, without adding words.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Deliver the intended emotion unmistakably, without adding words. [3.8 seconds duration]) You promised me you would tell the truth this time.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Deliver the intended emotion unmistakably, without adding words. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(Bitter indignation remains audible as you insist on being taken seriously. Deliver the intended emotion unmistakably, without adding words. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
Literal words in the separate text field:
You promised me you would tell the truth this time. And now... you are standing here pretending none of it happened. Do not ask me to calm down when you will not even listen.
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]) You promised me you would tell the truth this time.
(Natural clear speech. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(Natural clear speech. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
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. Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. The anger escalates into a controlled furious confrontation, not an unintelligible scream. Bitter indignation remains audible as you insist on being taken seriously. Deliver the intended emotion unmistakably, without adding words.
SCRIPT:
(Intense anger, loud frustrated ranting, sharply stressed words and forceful breath. Deliver the intended emotion unmistakably, without adding words.) [3.8 seconds duration] You promised me you would tell the truth this time.
(The anger escalates into a controlled furious confrontation, not an unintelligible scream. Deliver the intended emotion unmistakably, without adding words.) [4.2 seconds duration] And now... you are standing here pretending none of it happened.
(Bitter indignation remains audible as you insist on being taken seriously. Deliver the intended emotion unmistakably, without adding words.) [5.0 seconds duration] Do not ask me to calm down when you will not even listen.
Reference versus no-reference breakdown
reference
50
0.844
6.363
4.100
2.21%
0.639
none
50
0.812
6.381
4.207
0.59%
n/a
g1-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 g1-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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.822
6.349
4.255
1.85%
0.807
95.00%
-0.034
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
Start with a sudden burst of raw, unfiltered rage. Project from the diaphragm with sharp, clipped consonants. Let the frustration physically shake your upper body on the first phrase.
Three local directions
1. Sudden eruption of raw, unfiltered rage. Project from the diaphragm.
2. Maintain high vocal tension as the anger becomes focused and dangerous.
3. End with a low, controlled rumble of sustained, bitter resentment.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Start with a sudden burst of raw, unfiltered rage. Project from the diaphragm with sharp, clipped consonants. Let the frustration physically shake your upper body on the first phrase.
SCRIPT:
(Sudden eruption of raw, unfiltered rage. Project from the diaphragm. [3.8 seconds duration]) You promised me you would tell the truth this time.
(Maintain high vocal tension as the anger becomes focused and dangerous. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(End with a low, controlled rumble of sustained, bitter resentment. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
Literal words in the separate text field:
You promised me you would tell the truth this time. And now... you are standing here pretending none of it happened. Do not ask me to calm down when you will not even listen.
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]) You promised me you would tell the truth this time.
(Natural clear speech. [4.2 seconds duration]) And now... you are standing here pretending none of it happened.
(Natural clear speech. [5.0 seconds duration]) Do not ask me to calm down when you will not even listen.
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 with a sudden burst of raw, unfiltered rage. Project from the diaphragm with sharp, clipped consonants. Let the frustration physically shake your upper body on the first phrase.
SCRIPT:
(Sudden eruption of raw, unfiltered rage. Project from the diaphragm.) [3.8 seconds duration] You promised me you would tell the truth this time.
(Maintain high vocal tension as the anger becomes focused and dangerous.) [4.2 seconds duration] And now... you are standing here pretending none of it happened.
(End with a low, controlled rumble of sustained, bitter resentment.) [5.0 seconds duration] Do not ask me to calm down when you will not even listen.
Reference versus no-reference breakdown
reference
50
0.823
6.345
4.207
2.64%
0.605
none
50
0.820
6.354
4.302
1.06%
n/a
g1-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 g1-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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.809
6.305
4.171
1.94%
0.803
98.00%
-0.017
Prompting strategy: reacting to an unheard partner.
Reusable direction recipe
Stored GENERAL content
Shift the anger into a simmering, dangerous intensity; controlled fury delivered with deliberate, heavy weight to every syllable.
Three local directions
1. Lower the pitch slightly but maintain high tension in the throat
2. Speak with immense restraint allowing the threat to build beneath the surface
3. Hold the final word with a sustained, vibrating resonance
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Shift the anger into a simmering, dangerous intensity; controlled fury delivered with deliberate, heavy weight to every syllable.
SCRIPT:
(Lower the pitch slightly but maintain high tension in the throat [4.6 seconds duration]) I asked for one simple thing, and you laughed in my face.
(Speak with immense restraint allowing the threat to build beneath the surface [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Hold the final word with a sustained, vibrating resonance [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
Literal words in the separate text field:
I asked for one simple thing, and you laughed in my face. No... this is not another little misunderstanding you can wave away. Look at me and explain why you thought that was acceptable.
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]) I asked for one simple thing, and you laughed in my face.
(Natural clear speech. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Natural clear speech. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
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. Shift the anger into a simmering, dangerous intensity; controlled fury delivered with deliberate, heavy weight to every syllable.
SCRIPT:
(Lower the pitch slightly but maintain high tension in the throat) [4.6 seconds duration] I asked for one simple thing, and you laughed in my face.
(Speak with immense restraint allowing the threat to build beneath the surface) [4.2 seconds duration] No... this is not another little misunderstanding you can wave away.
(Hold the final word with a sustained, vibrating resonance) [4.2 seconds duration] Look at me and explain why you thought that was acceptable.
Reference versus no-reference breakdown
reference
50
0.833
6.289
4.115
1.69%
0.622
none
50
0.785
6.321
4.227
2.18%
n/a
g1-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 g1-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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.798
6.363
4.293
0.84%
0.812
99.00%
-0.029
Prompting strategy: attack, sustain and release.
Reusable direction recipe
Stored GENERAL content
Approach this with mounting pressure. The initial anger should be simmering, boiling just beneath the surface. Increase volume steadily, aiming for controlled intensity rather than shouting.
Three local directions
1. Begin with tightly constrained, simmering fury. Keep the voice lower initially.
2. Let the intensity rise like steam escaping a pressurized vessel. Focus on vocal strain.
3. The final delivery must sound weary yet stubbornly defiant, a low burn.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Approach this with mounting pressure. The initial anger should be simmering, boiling just beneath the surface. Increase volume steadily, aiming for controlled intensity rather than shouting.
SCRIPT:
(Begin with tightly constrained, simmering fury. Keep the voice lower initially. [4.6 seconds duration]) I asked for one simple thing, and you laughed in my face.
(Let the intensity rise like steam escaping a pressurized vessel. Focus on vocal strain. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(The final delivery must sound weary yet stubbornly defiant, a low burn. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
Literal words in the separate text field:
I asked for one simple thing, and you laughed in my face. No... this is not another little misunderstanding you can wave away. Look at me and explain why you thought that was acceptable.
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]) I asked for one simple thing, and you laughed in my face.
(Natural clear speech. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Natural clear speech. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
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. Approach this with mounting pressure. The initial anger should be simmering, boiling just beneath the surface. Increase volume steadily, aiming for controlled intensity rather than shouting.
SCRIPT:
(Begin with tightly constrained, simmering fury. Keep the voice lower initially.) [4.6 seconds duration] I asked for one simple thing, and you laughed in my face.
(Let the intensity rise like steam escaping a pressurized vessel. Focus on vocal strain.) [4.2 seconds duration] No... this is not another little misunderstanding you can wave away.
(The final delivery must sound weary yet stubbornly defiant, a low burn.) [4.2 seconds duration] Look at me and explain why you thought that was acceptable.
Reference versus no-reference breakdown
reference
50
0.793
6.350
4.245
0.44%
0.610
none
50
0.803
6.377
4.342
1.25%
n/a
g1-p06 · 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-p06 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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.754
6.368
4.228
2.91%
0.779
95.00%
-0.033
Prompting strategy: reacting to an unheard partner.
Reusable direction recipe
Stored GENERAL content
A deep, grinding indignation; the anger is less about shouting and more about the absolute refusal to yield or be ignored.
Three local directions
1. Infuse every word with a low guttural texture of resentment
2. Maintain a steady, unwavering pace despite the fury
3. Deliver the final statement with absolute, cold conviction
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. A deep, grinding indignation; the anger is less about shouting and more about the absolute refusal to yield or be ignored.
SCRIPT:
(Infuse every word with a low guttural texture of resentment [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(Maintain a steady, unwavering pace despite the fury [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Deliver the final statement with absolute, cold conviction [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
Literal words in the separate text field:
I have spent the entire morning fixing what you broke. You knew exactly what you were doing... and you did it anyway. For once in your life, take responsibility for the mess you made.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(Natural clear speech. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Natural clear speech. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
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 deep, grinding indignation; the anger is less about shouting and more about the absolute refusal to yield or be ignored.
SCRIPT:
(Infuse every word with a low guttural texture of resentment) [3.8 seconds duration] I have spent the entire morning fixing what you broke.
(Maintain a steady, unwavering pace despite the fury) [4.6 seconds duration] You knew exactly what you were doing... and you did it anyway.
(Deliver the final statement with absolute, cold conviction) [4.6 seconds duration] For once in your life, take responsibility for the mess you made.
Reference versus no-reference breakdown
reference
50
0.788
6.384
4.193
1.39%
0.606
none
50
0.720
6.352
4.264
4.42%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.746
6.344
4.279
1.59%
0.794
98.00%
-0.013
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
Transition to a controlled, simmering fury. Lower the volume slightly but increase the density of the threat in your tone. Maintain absolute control over your breathing.
Three local directions
1. Controlled, contained fury. The volume dips but the vocal texture becomes incredibly harsh and grating.
2. Pacing becomes deliberate and heavy. Each syllable is weighed down, like stone dropping.
3. Focus on the intellectual contempt behind the anger. It is a deliberate dismissal, not a tantrum.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. Transition to a controlled, simmering fury. Lower the volume slightly but increase the density of the threat in your tone. Maintain absolute control over your breathing.
SCRIPT:
(Controlled, contained fury. The volume dips but the vocal texture becomes incredibly harsh and grating. [4.6 seconds duration]) I asked for one simple thing, and you laughed in my face.
(Pacing becomes deliberate and heavy. Each syllable is weighed down, like stone dropping. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Focus on the intellectual contempt behind the anger. It is a deliberate dismissal, not a tantrum. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
Literal words in the separate text field:
I asked for one simple thing, and you laughed in my face. No... this is not another little misunderstanding you can wave away. Look at me and explain why you thought that was acceptable.
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]) I asked for one simple thing, and you laughed in my face.
(Natural clear speech. [4.2 seconds duration]) No... this is not another little misunderstanding you can wave away.
(Natural clear speech. [4.2 seconds duration]) Look at me and explain why you thought that was acceptable.
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. Transition to a controlled, simmering fury. Lower the volume slightly but increase the density of the threat in your tone. Maintain absolute control over your breathing.
SCRIPT:
(Controlled, contained fury. The volume dips but the vocal texture becomes incredibly harsh and grating.) [4.6 seconds duration] I asked for one simple thing, and you laughed in my face.
(Pacing becomes deliberate and heavy. Each syllable is weighed down, like stone dropping.) [4.2 seconds duration] No... this is not another little misunderstanding you can wave away.
(Focus on the intellectual contempt behind the anger. It is a deliberate dismissal, not a tantrum.) [4.2 seconds duration] Look at me and explain why you thought that was acceptable.
Reference versus no-reference breakdown
reference
50
0.751
6.320
4.246
0.64%
0.626
none
50
0.742
6.368
4.311
2.54%
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 emo_Anger. VoiceCLAP and absolute speaker cosine are diagnostics here, not extra weighted rewards.
Fixed target normalization: z = (raw score − 0.361) / 0.580; 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.
0.702
6.420
4.329
1.41%
0.777
98.00%
0.013
Prompting strategy: organic conversation.
Reusable direction recipe
Stored GENERAL content
End with a tone of exhausted, wounded indignation. The volume should drop to a forceful, low register. The anger is now rooted in deep personal offense.
Three local directions
1. Low, resonant indignation. The anger is banked, simmering just beneath the surface of weary disbelief.
2. Speak slowly, emphasizing key words with sheer, unwavering conviction. No shouting, only pressure.
3. A dry, bitter edge to the voice. You are utterly done with this, but you refuse to concede.
Exact tested prompt — fixed EN example, seed1701, no reference
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects. End with a tone of exhausted, wounded indignation. The volume should drop to a forceful, low register. The anger is now rooted in deep personal offense.
SCRIPT:
(Low, resonant indignation. The anger is banked, simmering just beneath the surface of weary disbelief. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(Speak slowly, emphasizing key words with sheer, unwavering conviction. No shouting, only pressure. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(A dry, bitter edge to the voice. You are utterly done with this, but you refuse to concede. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
Literal words in the separate text field:
I have spent the entire morning fixing what you broke. You knew exactly what you were doing... and you did it anyway. For once in your life, take responsibility for the mess you made.
Exact neutral CFG branch
GENERAL: One adult speaker, clean high-quality recording, no music or sound effects.
SCRIPT:
(Natural clear speech. [3.8 seconds duration]) I have spent the entire morning fixing what you broke.
(Natural clear speech. [4.6 seconds duration]) You knew exactly what you were doing... and you did it anyway.
(Natural clear speech. [4.6 seconds duration]) For once in your life, take responsibility for the mess you made.
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. End with a tone of exhausted, wounded indignation. The volume should drop to a forceful, low register. The anger is now rooted in deep personal offense.
SCRIPT:
(Low, resonant indignation. The anger is banked, simmering just beneath the surface of weary disbelief.) [3.8 seconds duration] I have spent the entire morning fixing what you broke.
(Speak slowly, emphasizing key words with sheer, unwavering conviction. No shouting, only pressure.) [4.6 seconds duration] You knew exactly what you were doing... and you did it anyway.
(A dry, bitter edge to the voice. You are utterly done with this, but you refuse to concede.) [4.6 seconds duration] For once in your life, take responsibility for the mess you made.