Humaneness Voice Large is an expressive text-to-speech model: it turns a script and English performance directions into speech. An optional separate speaker-reference recording provides a voice example; it is not the desired performance of this script.
Compare the same 35 scenes in five languages, each with and without the same reference: 70 takes per model. The words, full GENERAL/SCRIPT prompt, requested frame budget and original case seed are matched. Green players are the released 1.4 Alpha reinforcement-learning adapter, on the frozen final 1.3 SFT base. Earlier 1.2 and 1.3 SFT players are unchanged. Purple players, when present, are the separate subsequent 20/40/40 reward experiment, not a replacement for the 1.4 release.
Each new RL condition has one fixed-seed take: no best-of-N, quality-selected seeds, trimming or time-stretching. Existing 1.2 takes were originally selected by the first duration-valid attempt, not a quality metric; their exact selected seeds are retained here. Matching random seeds does not imply the same trajectory between different model weights.
Content Enjoyment, genuineness, word errors and emotion directions
Every output is included. Values are mean ± sample standard deviation, equally weighted per case (70 overall, 35 in each reference arm). EARS Medium scores the actual streamed 128 kbps mono MP3, and multilingual Parakeet measures ASR word error rate. CE and genuineness are learned estimates, not listener MOS. Higher CE/genuineness and lower WER are preferable. WER is not clipped to 100%.
Model
Condition
N
Content Enjoyment ↑ (≈0–10)
Genuineness ↑ (≈0–6)
WER ↓
Fixed-target top-2 emotion MSE ↓ (raw)
1.2 Alpha · SFT
all
70
6.235 ± 0.155
4.244 ± 0.218
1.133 ± 1.678%
1.838 ± 1.093
1.2 Alpha · SFT
prompt_only
35
6.230 ± 0.160
4.275 ± 0.200
1.237 ± 1.785%
1.826 ± 1.111
1.2 Alpha · SFT
reference
35
6.240 ± 0.151
4.213 ± 0.233
1.029 ± 1.583%
1.851 ± 1.091
1.3 Alpha · SFT
all
70
6.259 ± 0.186
4.200 ± 0.224
3.238 ± 12.006%
1.833 ± 1.007
1.3 Alpha · SFT
prompt_only
35
6.274 ± 0.176
4.257 ± 0.174
4.358 ± 16.170%
1.790 ± 1.004
1.3 Alpha · SFT
reference
35
6.243 ± 0.196
4.143 ± 0.254
2.118 ± 5.335%
1.875 ± 1.023
1.4 Alpha · rank-128 RL
all
70
6.328 ± 0.151
4.299 ± 0.218
2.743 ± 11.612%
1.793 ± 0.890
1.4 Alpha · rank-128 RL
prompt_only
35
6.341 ± 0.137
4.323 ± 0.188
3.689 ± 15.983%
1.714 ± 0.852
1.4 Alpha · rank-128 RL
reference
35
6.314 ± 0.166
4.276 ± 0.245
1.797 ± 4.046%
1.872 ± 0.933
Clips longer than the judge’s 30-second context are split into balanced, non-overlapping scoring windows covering the full MP3. Raw heads are duration-weighted. No audio tail is discarded or edited. All 192 raw heads, all 40 emotions, ASR hypotheses and window scores are retained in metrics.json.
This is a scripted listening/development study, not a certified held-out benchmark. The emotion MSE is a new approximate prompt-following diagnostic, not the audio-target emotion reward used in training. See the fixed targets below. No speaker-identity or reference speaker-similarity measurement is claimed on this page.
Exactly which two emotion targets are compared?
These scenes were written specifically for listening and have no original target performance to score. We therefore fixed two intended emotion heads and approximate scene-wide intensities from each directing prompt. They are identical across all languages, reference conditions and model versions. They were frozen before the new 1.2 scoring and subsequent RL experiment outputs, not selected from whichever emotions a generated sample happens to have.
Important: these are approximate prompt-derived targets, not validated emotional ground truth, and not the source-audio-derived top-two targets used in RL. Each scene changes emotion over time; a full-clip mean cannot verify every sentence’s arc. Raw MSE is the mean squared difference of the two raw predictions and fixed raw targets. Normalized MSE divides by 16 (the square of the nominal 0–4 emotion scale); predictions are not clipped.
Scene
First intended emotion / target
Second intended emotion / target
anger restraint
Anger / 2.5
Impatience and Irritability / 1.5
fear courage
Fear / 2.5
Hope Enthusiasm Optimism / 1.5
grief gratitude
Sadness / 2.0
Thankfulness Gratitude / 2.0
joy disbelief
Elation / 2.5
Thankfulness Gratitude / 2.0
amusement embarrassment
Amusement / 2.5
Embarrassment / 1.5
betrayal resolve
Bitterness / 2.0
Anger / 1.5
awe tenderness
Awe / 2.5
Affection / 2.0
Per-scene, per-head predicted means (all 10 language/reference cases)
The final 1.3 SFT model remains frozen while rank-128 attention adapters in its semantic backbone and Talker learn from groups of sampled performances. A judge scores Content Enjoyment, matching two fixed strongest supported emotions of the training source audio, and genuineness. VESPO sequence importance weighting and separately group-normalized reward components provide the relative policy-gradient update; KL and supervised anchoring regularize it. This is not the flow-matching GROW method used for the separate Voice Converter model.
Released 1.4 Alpha: 33% CE, 33% two-emotion matching and 34% genuineness; 1,500 attempted groups, 1,452 successful updates, eight candidates per reference/no-reference group. Additional experiment: continue that adapter for 2,000 further successful updates, with 20% CE, 40% two-emotion matching and 40% genuineness, sixteen candidates per group on 32 GPUs. It starts from 1.4 adapter weights and retained AdamW moments, with a new schedule, warm-up and rank RNG streams; the changed reward/world/data position means it is not an exact resume of the prior run.
Source data mix real speech and expressive synthetic speech, balanced across English, German, French, Dutch and Spanish. WER, full-support likelihood-parity, source QA and reference verification gates remain enforced. Outputs use the unchanged public deployment sampler (BF16 base with unmerged FP32 adapter residuals; temperature 1, top-p .95, top-k 50), whereas RL sampling/replay uses FP32, full support and temperature .7. The two distributions are deliberately distinguished.
Separate 12-second synthetic reference from the public voice-acting corpus; same input for every model. It does not contain this scene.
German reference
Separate 12-second synthetic reference from the public voice-acting corpus; same input for every model. It does not contain this scene.
French reference
Separate 12-second synthetic reference from the public voice-acting corpus; same input for every model. It does not contain this scene.
Dutch reference
Separate 12-second synthetic reference from the public voice-acting corpus; same input for every model. It does not contain this scene.
Spanish reference
Separate 12-second synthetic reference from the public voice-acting corpus; same input for every model. It does not contain this scene.
The missing safety report · English
Anger, disbelief and a deliberate return to controlled determination.
You told me the inspection was finished, and I let everyone go inside because I trusted your word. Now I am standing here with the unsigned report, while you keep calling this a tiny misunderstanding. A misunderstanding? Those were our friends in there... you do not get to make their safety sound unimportant. No, wait. I am angry, but shouting at you will not help us make this right tonight. Give me the complete file, call the inspector, and stay here until every single person knows the truth.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.72s · seed 20261021
CE 6.44 · genuineness 3.89 · WER 0.0% Fixed-target emotion MSE 0.355 (normalized 0.0222)
1.3 Alpha · SFT
Actual 34.72s · seed 20261021
CE 6.57 · genuineness 4.05 · WER 0.0% Fixed-target emotion MSE 0.379 (normalized 0.0237)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261021
CE 6.47 · genuineness 3.93 · WER 1.1% Fixed-target emotion MSE 0.410 (normalized 0.0256)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.72s · seed 20261021
CE 6.40 · genuineness 3.71 · WER 0.0% Fixed-target emotion MSE 0.597 (normalized 0.0373)
1.3 Alpha · SFT
Actual 34.72s · seed 20261021
CE 6.41 · genuineness 3.68 · WER 0.0% Fixed-target emotion MSE 0.323 (normalized 0.0202)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261021
CE 6.50 · genuineness 3.70 · WER 0.0% Fixed-target emotion MSE 0.252 (normalized 0.0157)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Anger, disbelief and a deliberate return to controlled determination.
SCRIPT:
(quietly furious, clipped but natural) [6.4 seconds duration] You told me the inspection was finished, and I let everyone go inside because I trusted your word.
[0.3 seconds pause]
(anger rising, emphatic and incredulous) [6.5 seconds duration] Now I am standing here with the unsigned report, while you keep calling this a tiny misunderstanding.
[0.3 seconds pause]
(exasperated sigh, 0.4 seconds)
(strongly angry, briefly ranting without losing clarity) [6.8 seconds duration] A misunderstanding? Those were our friends in there... you do not get to make their safety sound unimportant.
[0.3 seconds pause]
(a deliberate breath, regaining control) [6.8 seconds duration] No, wait. I am angry, but shouting at you will not help us make this right tonight.
[0.3 seconds pause]
(quiet, firm and protective rather than threatening) [6.7 seconds duration] Give me the complete file, call the inspector, and stay here until every single person knows the truth.
Anger, disbelief and a deliberate return to controlled determination.
Du hast mir gesagt, die Prüfung sei abgeschlossen, und ich habe alle hineingelassen, weil ich dir vertraut habe. Jetzt stehe ich hier mit dem Bericht ohne Unterschrift, während du immer noch von einem kleinen Missverständnis redest. Ein Missverständnis? Da waren unsere Freunde drin... du kannst doch nicht so tun, als wäre ihre Sicherheit unwichtig. Nein, warte. Ich bin wütend, aber dich anzuschreien hilft uns jetzt auch nicht dabei, das wieder in Ordnung zu bringen. Gib mir die vollständigen Unterlagen, ruf den Prüfer an, und bleib hier, bis jeder die ganze Wahrheit kennt.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.37 · genuineness 4.34 · WER 0.0% Fixed-target emotion MSE 0.448 (normalized 0.0280)
1.3 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.44 · genuineness 4.45 · WER 0.0% Fixed-target emotion MSE 0.281 (normalized 0.0176)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261021
CE 6.46 · genuineness 4.44 · WER 0.0% Fixed-target emotion MSE 0.491 (normalized 0.0307)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.26 · genuineness 4.10 · WER 0.0% Fixed-target emotion MSE 0.398 (normalized 0.0249)
1.3 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.37 · genuineness 4.34 · WER 0.0% Fixed-target emotion MSE 0.456 (normalized 0.0285)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261021
CE 6.50 · genuineness 4.33 · WER 2.2% Fixed-target emotion MSE 0.466 (normalized 0.0291)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Anger, disbelief and a deliberate return to controlled determination.
SCRIPT:
(quietly furious, clipped but natural) [6.4 seconds duration] Du hast mir gesagt, die Prüfung sei abgeschlossen, und ich habe alle hineingelassen, weil ich dir vertraut habe.
[0.3 seconds pause]
(anger rising, emphatic and incredulous) [6.5 seconds duration] Jetzt stehe ich hier mit dem Bericht ohne Unterschrift, während du immer noch von einem kleinen Missverständnis redest.
[0.3 seconds pause]
(exasperated sigh, 0.4 seconds)
(strongly angry, briefly ranting without losing clarity) [6.8 seconds duration] Ein Missverständnis? Da waren unsere Freunde drin... du kannst doch nicht so tun, als wäre ihre Sicherheit unwichtig.
[0.3 seconds pause]
(a deliberate breath, regaining control) [6.8 seconds duration] Nein, warte. Ich bin wütend, aber dich anzuschreien hilft uns jetzt auch nicht dabei, das wieder in Ordnung zu bringen.
[0.3 seconds pause]
(quiet, firm and protective rather than threatening) [6.7 seconds duration] Gib mir die vollständigen Unterlagen, ruf den Prüfer an, und bleib hier, bis jeder die ganze Wahrheit kennt.
Anger, disbelief and a deliberate return to controlled determination.
Je zei dat de inspectie klaar was, en ik heb iedereen binnengelaten omdat ik op jouw woord vertrouwde. Nu sta ik hier met een verslag zonder handtekening, terwijl jij dit nog steeds een klein misverstand noemt. Een misverstand? Onze vrienden waren daarbinnen... je kunt toch niet doen alsof hun veiligheid er niet toe doet. Nee, wacht even. Ik ben boos, maar tegen jou schreeuwen helpt ons niet om dit vanavond goed te maken. Geef me het volledige dossier, bel de inspecteur en blijf hier totdat iedereen precies weet wat er gebeurd is.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.34 · genuineness 3.98 · WER 3.3% Fixed-target emotion MSE 1.164 (normalized 0.0727)
1.3 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.30 · genuineness 3.99 · WER 3.3% Fixed-target emotion MSE 1.827 (normalized 0.1142)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261021
CE 6.40 · genuineness 4.06 · WER 3.3% Fixed-target emotion MSE 1.244 (normalized 0.0777)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.32 · genuineness 3.92 · WER 4.3% Fixed-target emotion MSE 2.263 (normalized 0.1414)
1.3 Alpha · SFT
Actual 34.72s · seed 20261021
CE 6.26 · genuineness 3.75 · WER 4.3% Fixed-target emotion MSE 2.232 (normalized 0.1395)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261021
CE 6.36 · genuineness 3.97 · WER 3.3% Fixed-target emotion MSE 2.498 (normalized 0.1561)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Anger, disbelief and a deliberate return to controlled determination.
SCRIPT:
(quietly furious, clipped but natural) [6.4 seconds duration] Je zei dat de inspectie klaar was, en ik heb iedereen binnengelaten omdat ik op jouw woord vertrouwde.
[0.3 seconds pause]
(anger rising, emphatic and incredulous) [6.5 seconds duration] Nu sta ik hier met een verslag zonder handtekening, terwijl jij dit nog steeds een klein misverstand noemt.
[0.3 seconds pause]
(exasperated sigh, 0.4 seconds)
(strongly angry, briefly ranting without losing clarity) [6.8 seconds duration] Een misverstand? Onze vrienden waren daarbinnen... je kunt toch niet doen alsof hun veiligheid er niet toe doet.
[0.3 seconds pause]
(a deliberate breath, regaining control) [6.8 seconds duration] Nee, wacht even. Ik ben boos, maar tegen jou schreeuwen helpt ons niet om dit vanavond goed te maken.
[0.3 seconds pause]
(quiet, firm and protective rather than threatening) [6.7 seconds duration] Geef me het volledige dossier, bel de inspecteur en blijf hier totdat iedereen precies weet wat er gebeurd is.
Anger, disbelief and a deliberate return to controlled determination.
Tu m'avais dit que l'inspection était terminée, et j'ai laissé entrer tout le monde parce que je te faisais confiance. Maintenant, je tiens un rapport sans signature, pendant que tu continues à parler d'un tout petit malentendu. Un malentendu? Nos amis étaient là-dedans... tu ne peux pas faire comme si leur sécurité était un détail. Non, attends. Je suis en colère, mais te crier dessus ne nous aidera pas à réparer les choses ce soir. Donne-moi le dossier complet, appelle la personne responsable et reste ici jusqu'à ce que tout le monde connaisse la vérité.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.72s · seed 20261021
CE 6.20 · genuineness 4.40 · WER 0.0% Fixed-target emotion MSE 0.852 (normalized 0.0533)
1.3 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.28 · genuineness 4.24 · WER 1.0% Fixed-target emotion MSE 0.967 (normalized 0.0605)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261021
CE 6.33 · genuineness 4.27 · WER 0.0% Fixed-target emotion MSE 2.022 (normalized 0.1263)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.21 · genuineness 4.20 · WER 0.0% Fixed-target emotion MSE 0.323 (normalized 0.0202)
1.3 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.25 · genuineness 4.18 · WER 0.0% Fixed-target emotion MSE 0.389 (normalized 0.0243)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261021
CE 6.28 · genuineness 4.19 · WER 0.0% Fixed-target emotion MSE 1.056 (normalized 0.0660)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Anger, disbelief and a deliberate return to controlled determination.
SCRIPT:
(quietly furious, clipped but natural) [6.4 seconds duration] Tu m'avais dit que l'inspection était terminée, et j'ai laissé entrer tout le monde parce que je te faisais confiance.
[0.3 seconds pause]
(anger rising, emphatic and incredulous) [6.5 seconds duration] Maintenant, je tiens un rapport sans signature, pendant que tu continues à parler d'un tout petit malentendu.
[0.3 seconds pause]
(exasperated sigh, 0.4 seconds)
(strongly angry, briefly ranting without losing clarity) [6.8 seconds duration] Un malentendu? Nos amis étaient là-dedans... tu ne peux pas faire comme si leur sécurité était un détail.
[0.3 seconds pause]
(a deliberate breath, regaining control) [6.8 seconds duration] Non, attends. Je suis en colère, mais te crier dessus ne nous aidera pas à réparer les choses ce soir.
[0.3 seconds pause]
(quiet, firm and protective rather than threatening) [6.7 seconds duration] Donne-moi le dossier complet, appelle la personne responsable et reste ici jusqu'à ce que tout le monde connaisse la vérité.
Anger, disbelief and a deliberate return to controlled determination.
Me dijiste que la inspección había terminado, y dejé entrar a todo el mundo porque confiaba en tu palabra. Ahora tengo delante un informe sin firmar, mientras tú sigues diciendo que todo esto es un pequeño malentendido. ¿Un malentendido? Nuestros amigos estaban ahí dentro... no puedes hablar de su seguridad como si no tuviera importancia. No, espera. Estoy muy enfadado, pero gritarte no nos va a ayudar a arreglar las cosas esta noche. Dame el expediente completo, llama a la persona responsable y quédate aquí hasta que todo el mundo sepa la verdad.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.32 · genuineness 4.49 · WER 0.0% Fixed-target emotion MSE 1.236 (normalized 0.0773)
1.3 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.26 · genuineness 4.27 · WER 0.0% Fixed-target emotion MSE 1.055 (normalized 0.0659)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261021
CE 6.40 · genuineness 4.39 · WER 0.0% Fixed-target emotion MSE 1.843 (normalized 0.1152)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.32 · genuineness 4.17 · WER 0.0% Fixed-target emotion MSE 1.209 (normalized 0.0756)
1.3 Alpha · SFT
Actual 34.80s · seed 20261021
CE 6.32 · genuineness 4.03 · WER 0.0% Fixed-target emotion MSE 0.416 (normalized 0.0260)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261021
CE 6.28 · genuineness 4.21 · WER 0.0% Fixed-target emotion MSE 2.023 (normalized 0.1265)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Anger, disbelief and a deliberate return to controlled determination.
SCRIPT:
(quietly furious, clipped but natural) [6.4 seconds duration] Me dijiste que la inspección había terminado, y dejé entrar a todo el mundo porque confiaba en tu palabra.
[0.3 seconds pause]
(anger rising, emphatic and incredulous) [6.5 seconds duration] Ahora tengo delante un informe sin firmar, mientras tú sigues diciendo que todo esto es un pequeño malentendido.
[0.3 seconds pause]
(exasperated sigh, 0.4 seconds)
(strongly angry, briefly ranting without losing clarity) [6.8 seconds duration] ¿Un malentendido? Nuestros amigos estaban ahí dentro... no puedes hablar de su seguridad como si no tuviera importancia.
[0.3 seconds pause]
(a deliberate breath, regaining control) [6.8 seconds duration] No, espera. Estoy muy enfadado, pero gritarte no nos va a ayudar a arreglar las cosas esta noche.
[0.3 seconds pause]
(quiet, firm and protective rather than threatening) [6.7 seconds duration] Dame el expediente completo, llama a la persona responsable y quédate aquí hasta que todo el mundo sepa la verdad.
Fear and momentary panic become courage and reassuring companionship.
Keep your hand on my shoulder, please; the lights have gone out again, and I cannot see the stairs. Did you hear that behind us? For a second I thought... no, I think it was only the door. I need one moment to breathe, because my heart is racing and I am trying not to frighten you. There, that little green light is still on; we can follow the wall until we reach the exit together. I am scared too, honestly, but I am right beside you, and neither of us is going through this alone.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.29 · genuineness 4.22 · WER 1.0% Fixed-target emotion MSE 0.214 (normalized 0.0133)
1.3 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.22 · genuineness 4.09 · WER 0.0% Fixed-target emotion MSE 0.063 (normalized 0.0040)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261122
CE 6.56 · genuineness 4.23 · WER 0.0% Fixed-target emotion MSE 0.220 (normalized 0.0137)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.72s · seed 20261122
CE 6.27 · genuineness 3.90 · WER 0.0% Fixed-target emotion MSE 0.297 (normalized 0.0186)
1.3 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.33 · genuineness 4.11 · WER 0.0% Fixed-target emotion MSE 0.146 (normalized 0.0091)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261122
CE 6.52 · genuineness 4.22 · WER 0.0% Fixed-target emotion MSE 0.350 (normalized 0.0219)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Fear and momentary panic become courage and reassuring companionship.
SCRIPT:
(hushed and alert, genuinely frightened) [6.4 seconds duration] Keep your hand on my shoulder, please; the lights have gone out again, and I cannot see the stairs.
[0.3 seconds pause]
(panic briefly breaking through, breath catching) [6.5 seconds duration] Did you hear that behind us? For a second I thought... no, I think it was only the door.
[0.3 seconds pause]
(sharp inhale, 0.4 seconds)
(hesitant but trying to steady the breath) [6.8 seconds duration] I need one moment to breathe, because my heart is racing and I am trying not to frighten you.
[0.3 seconds pause]
(gathering courage, clearer and more grounded) [6.8 seconds duration] There, that little green light is still on; we can follow the wall until we reach the exit together.
[0.3 seconds pause]
(gentle reassurance, still afraid but warmly determined) [6.7 seconds duration] I am scared too, honestly, but I am right beside you, and neither of us is going through this alone.
Fear and momentary panic become courage and reassuring companionship.
Lass bitte deine Hand auf meiner Schulter; das Licht ist schon wieder aus, und ich kann die Treppe nicht sehen. Hast du das hinter uns gehört? Einen Moment dachte ich... nein, wahrscheinlich war es wirklich nur die Tür. Ich muss kurz durchatmen, weil mein Herz so rast, und ich möchte dir nicht noch mehr Angst machen. Da vorne leuchtet noch dieses kleine grüne Licht; wir gehen gemeinsam an der Wand entlang bis zum Ausgang. Ich habe auch Angst, ganz ehrlich, aber ich bin direkt neben dir, und niemand von uns muss hier alleine durch.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.19 · genuineness 4.44 · WER 1.1% Fixed-target emotion MSE 0.499 (normalized 0.0312)
1.3 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.28 · genuineness 4.51 · WER 0.0% Fixed-target emotion MSE 0.625 (normalized 0.0391)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261122
CE 6.42 · genuineness 4.67 · WER 0.0% Fixed-target emotion MSE 0.818 (normalized 0.0511)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.72s · seed 20261122
CE 6.31 · genuineness 4.65 · WER 0.0% Fixed-target emotion MSE 0.500 (normalized 0.0313)
1.3 Alpha · SFT
Actual 34.80s · seed 20261122
CE 5.94 · genuineness 4.29 · WER 0.0% Fixed-target emotion MSE 0.340 (normalized 0.0212)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261122
CE 6.40 · genuineness 4.56 · WER 0.0% Fixed-target emotion MSE 0.624 (normalized 0.0390)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Fear and momentary panic become courage and reassuring companionship.
SCRIPT:
(hushed and alert, genuinely frightened) [6.4 seconds duration] Lass bitte deine Hand auf meiner Schulter; das Licht ist schon wieder aus, und ich kann die Treppe nicht sehen.
[0.3 seconds pause]
(panic briefly breaking through, breath catching) [6.5 seconds duration] Hast du das hinter uns gehört? Einen Moment dachte ich... nein, wahrscheinlich war es wirklich nur die Tür.
[0.3 seconds pause]
(sharp inhale, 0.4 seconds)
(hesitant but trying to steady the breath) [6.8 seconds duration] Ich muss kurz durchatmen, weil mein Herz so rast, und ich möchte dir nicht noch mehr Angst machen.
[0.3 seconds pause]
(gathering courage, clearer and more grounded) [6.8 seconds duration] Da vorne leuchtet noch dieses kleine grüne Licht; wir gehen gemeinsam an der Wand entlang bis zum Ausgang.
[0.3 seconds pause]
(gentle reassurance, still afraid but warmly determined) [6.7 seconds duration] Ich habe auch Angst, ganz ehrlich, aber ich bin direkt neben dir, und niemand von uns muss hier alleine durch.
Fear and momentary panic become courage and reassuring companionship.
Houd je hand alsjeblieft op mijn schouder; het licht is weer uitgevallen en ik kan de trap niet meer zien. Hoorde jij dat achter ons? Even dacht ik... nee, volgens mij was het toch alleen maar de deur. Ik moet heel even ademhalen, want mijn hart gaat tekeer en ik wil jou niet nog banger maken. Daar brandt nog een klein groen lampje; we kunnen samen langs de muur lopen totdat we bij de uitgang zijn. Ik ben ook bang, echt waar, maar ik blijf naast je en we hoeven dit geen van beiden alleen te doen.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.16 · genuineness 4.24 · WER 5.2% Fixed-target emotion MSE 1.757 (normalized 0.1098)
1.3 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.15 · genuineness 4.17 · WER 4.1% Fixed-target emotion MSE 1.746 (normalized 0.1092)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261122
CE 6.19 · genuineness 4.17 · WER 3.1% Fixed-target emotion MSE 1.806 (normalized 0.1129)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.06 · genuineness 4.39 · WER 4.1% Fixed-target emotion MSE 2.518 (normalized 0.1574)
1.3 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.24 · genuineness 4.16 · WER 4.1% Fixed-target emotion MSE 1.745 (normalized 0.1091)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261122
CE 6.27 · genuineness 4.36 · WER 4.1% Fixed-target emotion MSE 2.344 (normalized 0.1465)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Fear and momentary panic become courage and reassuring companionship.
SCRIPT:
(hushed and alert, genuinely frightened) [6.4 seconds duration] Houd je hand alsjeblieft op mijn schouder; het licht is weer uitgevallen en ik kan de trap niet meer zien.
[0.3 seconds pause]
(panic briefly breaking through, breath catching) [6.5 seconds duration] Hoorde jij dat achter ons? Even dacht ik... nee, volgens mij was het toch alleen maar de deur.
[0.3 seconds pause]
(sharp inhale, 0.4 seconds)
(hesitant but trying to steady the breath) [6.8 seconds duration] Ik moet heel even ademhalen, want mijn hart gaat tekeer en ik wil jou niet nog banger maken.
[0.3 seconds pause]
(gathering courage, clearer and more grounded) [6.8 seconds duration] Daar brandt nog een klein groen lampje; we kunnen samen langs de muur lopen totdat we bij de uitgang zijn.
[0.3 seconds pause]
(gentle reassurance, still afraid but warmly determined) [6.7 seconds duration] Ik ben ook bang, echt waar, maar ik blijf naast je en we hoeven dit geen van beiden alleen te doen.
Fear and momentary panic become courage and reassuring companionship.
Garde ta main sur mon épaule, s'il te plaît; la lumière s'est encore éteinte et je ne vois plus les marches. Tu as entendu ce bruit derrière nous? Pendant une seconde, j'ai cru... non, c'était probablement juste la porte. Laisse-moi reprendre mon souffle, mon cœur bat trop vite et je ne veux surtout pas te faire encore plus peur. Là-bas, la petite lumière verte fonctionne toujours; nous pouvons longer le mur ensemble jusqu'à la sortie, tout doucement. J'ai peur aussi, vraiment, mais je reste près de toi et personne ne devra traverser ça tout seul.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261122
CE 5.92 · genuineness 4.33 · WER 0.0% Fixed-target emotion MSE 1.188 (normalized 0.0742)
1.3 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.09 · genuineness 4.52 · WER 0.0% Fixed-target emotion MSE 2.124 (normalized 0.1327)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261122
CE 6.18 · genuineness 4.51 · WER 0.0% Fixed-target emotion MSE 1.970 (normalized 0.1231)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.15 · genuineness 4.26 · WER 0.0% Fixed-target emotion MSE 1.113 (normalized 0.0695)
1.3 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.13 · genuineness 4.41 · WER 0.0% Fixed-target emotion MSE 1.713 (normalized 0.1071)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261122
CE 6.17 · genuineness 4.56 · WER 1.0% Fixed-target emotion MSE 2.583 (normalized 0.1614)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Fear and momentary panic become courage and reassuring companionship.
SCRIPT:
(hushed and alert, genuinely frightened) [6.4 seconds duration] Garde ta main sur mon épaule, s'il te plaît; la lumière s'est encore éteinte et je ne vois plus les marches.
[0.3 seconds pause]
(panic briefly breaking through, breath catching) [6.5 seconds duration] Tu as entendu ce bruit derrière nous? Pendant une seconde, j'ai cru... non, c'était probablement juste la porte.
[0.3 seconds pause]
(sharp inhale, 0.4 seconds)
(hesitant but trying to steady the breath) [6.8 seconds duration] Laisse-moi reprendre mon souffle, mon cœur bat trop vite et je ne veux surtout pas te faire encore plus peur.
[0.3 seconds pause]
(gathering courage, clearer and more grounded) [6.8 seconds duration] Là-bas, la petite lumière verte fonctionne toujours; nous pouvons longer le mur ensemble jusqu'à la sortie, tout doucement.
[0.3 seconds pause]
(gentle reassurance, still afraid but warmly determined) [6.7 seconds duration] J'ai peur aussi, vraiment, mais je reste près de toi et personne ne devra traverser ça tout seul.
Fear and momentary panic become courage and reassuring companionship.
Deja la mano en mi hombro, por favor; se han vuelto a apagar las luces y no veo los escalones. ¿Has oído eso detrás de nosotros? Por un momento pensé... no, creo que solamente ha sido la puerta. Necesito un segundo para respirar, porque el corazón me va muy rápido y no quiero asustarte todavía más. Mira, esa pequeña luz verde sigue encendida; podemos avanzar juntos siguiendo la pared hasta que lleguemos a la salida. Yo también tengo miedo, de verdad, pero estoy a tu lado y ninguno de los dos va a pasar por esto solo.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.07 · genuineness 4.17 · WER 0.0% Fixed-target emotion MSE 1.614 (normalized 0.1009)
1.3 Alpha · SFT
Actual 34.72s · seed 20261122
CE 6.08 · genuineness 4.22 · WER 0.0% Fixed-target emotion MSE 1.750 (normalized 0.1094)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261122
CE 6.27 · genuineness 4.40 · WER 0.0% Fixed-target emotion MSE 1.240 (normalized 0.0775)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.14 · genuineness 4.24 · WER 0.0% Fixed-target emotion MSE 1.055 (normalized 0.0660)
1.3 Alpha · SFT
Actual 34.80s · seed 20261122
CE 6.23 · genuineness 4.26 · WER 0.0% Fixed-target emotion MSE 2.142 (normalized 0.1339)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261122
CE 6.29 · genuineness 4.52 · WER 0.0% Fixed-target emotion MSE 2.096 (normalized 0.1310)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Fear and momentary panic become courage and reassuring companionship.
SCRIPT:
(hushed and alert, genuinely frightened) [6.4 seconds duration] Deja la mano en mi hombro, por favor; se han vuelto a apagar las luces y no veo los escalones.
[0.3 seconds pause]
(panic briefly breaking through, breath catching) [6.5 seconds duration] ¿Has oído eso detrás de nosotros? Por un momento pensé... no, creo que solamente ha sido la puerta.
[0.3 seconds pause]
(sharp inhale, 0.4 seconds)
(hesitant but trying to steady the breath) [6.8 seconds duration] Necesito un segundo para respirar, porque el corazón me va muy rápido y no quiero asustarte todavía más.
[0.3 seconds pause]
(gathering courage, clearer and more grounded) [6.8 seconds duration] Mira, esa pequeña luz verde sigue encendida; podemos avanzar juntos siguiendo la pared hasta que lleguemos a la salida.
[0.3 seconds pause]
(gentle reassurance, still afraid but warmly determined) [6.7 seconds duration] Yo también tengo miedo, de verdad, pero estoy a tu lado y ninguno de los dos va a pasar por esto solo.
Grief, a fond memory and unexpectedly tender gratitude.
I nearly threw this old radio away last winter, because every time I touched it I remembered his hands. He used to turn that crooked little dial and insist the music sounded better when the kitchen window was open. And now you have fixed it... I did not realize how much I wanted to hear that silly crackle again. Thank you for keeping the scratches instead of making it look new; they are part of what I missed. Could you stay for a cup of tea while it plays? I think I would like to remember him with someone tonight.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.48 · genuineness 4.28 · WER 0.0% Fixed-target emotion MSE 1.067 (normalized 0.0667)
1.3 Alpha · SFT
Actual 34.72s · seed 20261223
CE 6.41 · genuineness 4.01 · WER 0.0% Fixed-target emotion MSE 1.545 (normalized 0.0965)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261223
CE 6.43 · genuineness 4.20 · WER 0.0% Fixed-target emotion MSE 1.499 (normalized 0.0937)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.43 · genuineness 4.32 · WER 0.0% Fixed-target emotion MSE 1.162 (normalized 0.0727)
1.3 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.49 · genuineness 4.15 · WER 0.0% Fixed-target emotion MSE 1.472 (normalized 0.0920)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261223
CE 6.59 · genuineness 4.25 · WER 0.0% Fixed-target emotion MSE 1.676 (normalized 0.1048)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Grief, a fond memory and unexpectedly tender gratitude.
SCRIPT:
(quiet grief, slightly trembling but intelligible) [6.4 seconds duration] I nearly threw this old radio away last winter, because every time I touched it I remembered his hands.
[0.3 seconds pause]
(a fond memory bringing a fragile smile) [6.5 seconds duration] He used to turn that crooked little dial and insist the music sounded better when the kitchen window was open.
[0.3 seconds pause]
(wistful sigh, 0.4 seconds)
(voice catching, allow a small reflective hesitation) [6.8 seconds duration] And now you have fixed it... I did not realize how much I wanted to hear that silly crackle again.
[0.3 seconds pause]
(surprised gratitude, intimate and sincere) [6.8 seconds duration] Thank you for keeping the scratches instead of making it look new; they are part of what I missed.
[0.3 seconds pause]
(steadier, warm and quietly hopeful) [6.7 seconds duration] Could you stay for a cup of tea while it plays? I think I would like to remember him with someone tonight.
Grief, a fond memory and unexpectedly tender gratitude.
Letzten Winter wollte ich dieses alte Radio fast wegwerfen, weil ich jedes Mal seine Hände vor mir gesehen habe. Er drehte immer an diesem schiefen kleinen Knopf und behauptete, bei offenem Küchenfenster würde die Musik einfach besser klingen. Und jetzt hast du es repariert... mir war gar nicht klar, wie sehr ich dieses alberne Knistern vermisst habe. Danke, dass du die Kratzer nicht wegpoliert hast; genau diese kleinen Spuren gehören zu dem, was mir gefehlt hat. Bleibst du noch auf einen Tee, während es spielt? Ich glaube, heute möchte ich mich nicht ganz alleine an ihn erinnern.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.25 · genuineness 4.53 · WER 0.0% Fixed-target emotion MSE 0.831 (normalized 0.0519)
1.3 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.28 · genuineness 4.53 · WER 0.0% Fixed-target emotion MSE 1.142 (normalized 0.0714)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261223
CE 6.39 · genuineness 4.60 · WER 0.0% Fixed-target emotion MSE 0.999 (normalized 0.0625)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.72s · seed 20261223
CE 6.37 · genuineness 4.47 · WER 0.0% Fixed-target emotion MSE 1.498 (normalized 0.0936)
1.3 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.36 · genuineness 4.40 · WER 0.0% Fixed-target emotion MSE 1.702 (normalized 0.1064)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261223
CE 6.46 · genuineness 4.61 · WER 0.0% Fixed-target emotion MSE 1.472 (normalized 0.0920)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Grief, a fond memory and unexpectedly tender gratitude.
SCRIPT:
(quiet grief, slightly trembling but intelligible) [6.4 seconds duration] Letzten Winter wollte ich dieses alte Radio fast wegwerfen, weil ich jedes Mal seine Hände vor mir gesehen habe.
[0.3 seconds pause]
(a fond memory bringing a fragile smile) [6.5 seconds duration] Er drehte immer an diesem schiefen kleinen Knopf und behauptete, bei offenem Küchenfenster würde die Musik einfach besser klingen.
[0.3 seconds pause]
(wistful sigh, 0.4 seconds)
(voice catching, allow a small reflective hesitation) [6.8 seconds duration] Und jetzt hast du es repariert... mir war gar nicht klar, wie sehr ich dieses alberne Knistern vermisst habe.
[0.3 seconds pause]
(surprised gratitude, intimate and sincere) [6.8 seconds duration] Danke, dass du die Kratzer nicht wegpoliert hast; genau diese kleinen Spuren gehören zu dem, was mir gefehlt hat.
[0.3 seconds pause]
(steadier, warm and quietly hopeful) [6.7 seconds duration] Bleibst du noch auf einen Tee, während es spielt? Ich glaube, heute möchte ich mich nicht ganz alleine an ihn erinnern.
Grief, a fond memory and unexpectedly tender gratitude.
Vorige winter had ik deze oude radio bijna weggegooid, omdat ik bij elke aanraking weer aan zijn handen moest denken. Hij draaide altijd aan dat scheve knopje en beweerde dat de muziek beter klonk als het keukenraam openstond. En nu heb jij hem gerepareerd... ik wist niet eens hoe graag ik dat gekke gekraak weer wilde horen. Dank je dat je de krassen hebt laten zitten in plaats van alles nieuw te maken; die kleine sporen horen erbij. Wil je nog een kop thee blijven drinken terwijl hij speelt? Ik zou vanavond graag samen met iemand aan hem denken.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.21 · genuineness 4.36 · WER 6.1% Fixed-target emotion MSE 2.914 (normalized 0.1821)
1.3 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.36 · genuineness 4.36 · WER 6.1% Fixed-target emotion MSE 2.718 (normalized 0.1699)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261223
CE 6.35 · genuineness 4.33 · WER 9.1% Fixed-target emotion MSE 3.382 (normalized 0.2114)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.25 · genuineness 4.22 · WER 6.1% Fixed-target emotion MSE 2.864 (normalized 0.1790)
1.3 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.19 · genuineness 4.22 · WER 7.1% Fixed-target emotion MSE 3.268 (normalized 0.2043)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261223
CE 6.28 · genuineness 4.16 · WER 5.1% Fixed-target emotion MSE 3.075 (normalized 0.1922)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Grief, a fond memory and unexpectedly tender gratitude.
SCRIPT:
(quiet grief, slightly trembling but intelligible) [6.4 seconds duration] Vorige winter had ik deze oude radio bijna weggegooid, omdat ik bij elke aanraking weer aan zijn handen moest denken.
[0.3 seconds pause]
(a fond memory bringing a fragile smile) [6.5 seconds duration] Hij draaide altijd aan dat scheve knopje en beweerde dat de muziek beter klonk als het keukenraam openstond.
[0.3 seconds pause]
(wistful sigh, 0.4 seconds)
(voice catching, allow a small reflective hesitation) [6.8 seconds duration] En nu heb jij hem gerepareerd... ik wist niet eens hoe graag ik dat gekke gekraak weer wilde horen.
[0.3 seconds pause]
(surprised gratitude, intimate and sincere) [6.8 seconds duration] Dank je dat je de krassen hebt laten zitten in plaats van alles nieuw te maken; die kleine sporen horen erbij.
[0.3 seconds pause]
(steadier, warm and quietly hopeful) [6.7 seconds duration] Wil je nog een kop thee blijven drinken terwijl hij speelt? Ik zou vanavond graag samen met iemand aan hem denken.
Grief, a fond memory and unexpectedly tender gratitude.
L'hiver dernier, j'ai failli jeter cette vieille radio, parce que chaque fois que je la touchais, je revoyais ses mains. Il tournait toujours ce petit bouton de travers et disait que la musique était meilleure quand la fenêtre restait ouverte. Et maintenant, tu l'as réparée... je ne savais pas à quel point ce petit grésillement ridicule me manquait encore. Merci d'avoir gardé les rayures au lieu de tout remettre à neuf; ces petites marques font partie du souvenir. Tu resterais prendre un thé pendant qu'elle joue? Je crois que ce soir, je voudrais penser à lui avec quelqu'un.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.04 · genuineness 4.49 · WER 1.0% Fixed-target emotion MSE 1.002 (normalized 0.0626)
1.3 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.06 · genuineness 4.39 · WER 5.1% Fixed-target emotion MSE 0.619 (normalized 0.0387)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261223
CE 6.15 · genuineness 4.39 · WER 1.0% Fixed-target emotion MSE 0.890 (normalized 0.0556)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261223
CE 5.91 · genuineness 4.39 · WER 3.1% Fixed-target emotion MSE 0.682 (normalized 0.0426)
1.3 Alpha · SFT
Actual 34.72s · seed 20261223
CE 5.52 · genuineness 4.41 · WER 21.4% Fixed-target emotion MSE 1.104 (normalized 0.0690)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261223
CE 5.98 · genuineness 4.47 · WER 4.1% Fixed-target emotion MSE 0.818 (normalized 0.0511)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Grief, a fond memory and unexpectedly tender gratitude.
SCRIPT:
(quiet grief, slightly trembling but intelligible) [6.4 seconds duration] L'hiver dernier, j'ai failli jeter cette vieille radio, parce que chaque fois que je la touchais, je revoyais ses mains.
[0.3 seconds pause]
(a fond memory bringing a fragile smile) [6.5 seconds duration] Il tournait toujours ce petit bouton de travers et disait que la musique était meilleure quand la fenêtre restait ouverte.
[0.3 seconds pause]
(wistful sigh, 0.4 seconds)
(voice catching, allow a small reflective hesitation) [6.8 seconds duration] Et maintenant, tu l'as réparée... je ne savais pas à quel point ce petit grésillement ridicule me manquait encore.
[0.3 seconds pause]
(surprised gratitude, intimate and sincere) [6.8 seconds duration] Merci d'avoir gardé les rayures au lieu de tout remettre à neuf; ces petites marques font partie du souvenir.
[0.3 seconds pause]
(steadier, warm and quietly hopeful) [6.7 seconds duration] Tu resterais prendre un thé pendant qu'elle joue? Je crois que ce soir, je voudrais penser à lui avec quelqu'un.
Grief, a fond memory and unexpectedly tender gratitude.
El invierno pasado casi tiré esta vieja radio, porque cada vez que la tocaba volvía a recordar sus manos. Siempre giraba ese pequeño botón torcido y decía que la música sonaba mejor con la ventana de la cocina abierta. Y ahora la has arreglado... no sabía cuánto necesitaba volver a escuchar ese pequeño crujido que antes me parecía absurdo. Gracias por conservar los arañazos en vez de dejarla como nueva; esas marcas también forman parte de lo que echaba de menos. ¿Te quedarías a tomar un té mientras suena? Creo que esta noche me gustaría recordarlo en compañía de alguien.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.23 · genuineness 4.30 · WER 2.0% Fixed-target emotion MSE 0.646 (normalized 0.0404)
1.3 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.12 · genuineness 4.14 · WER 3.0% Fixed-target emotion MSE 1.036 (normalized 0.0647)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261223
CE 6.26 · genuineness 4.15 · WER 1.0% Fixed-target emotion MSE 1.521 (normalized 0.0951)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.09 · genuineness 4.31 · WER 1.0% Fixed-target emotion MSE 1.159 (normalized 0.0725)
1.3 Alpha · SFT
Actual 34.80s · seed 20261223
CE 6.11 · genuineness 4.10 · WER 1.0% Fixed-target emotion MSE 1.491 (normalized 0.0932)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261223
CE 6.16 · genuineness 4.28 · WER 1.0% Fixed-target emotion MSE 1.198 (normalized 0.0749)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Grief, a fond memory and unexpectedly tender gratitude.
SCRIPT:
(quiet grief, slightly trembling but intelligible) [6.4 seconds duration] El invierno pasado casi tiré esta vieja radio, porque cada vez que la tocaba volvía a recordar sus manos.
[0.3 seconds pause]
(a fond memory bringing a fragile smile) [6.5 seconds duration] Siempre giraba ese pequeño botón torcido y decía que la música sonaba mejor con la ventana de la cocina abierta.
[0.3 seconds pause]
(wistful sigh, 0.4 seconds)
(voice catching, allow a small reflective hesitation) [6.8 seconds duration] Y ahora la has arreglado... no sabía cuánto necesitaba volver a escuchar ese pequeño crujido que antes me parecía absurdo.
[0.3 seconds pause]
(surprised gratitude, intimate and sincere) [6.8 seconds duration] Gracias por conservar los arañazos en vez de dejarla como nueva; esas marcas también forman parte de lo que echaba de menos.
[0.3 seconds pause]
(steadier, warm and quietly hopeful) [6.7 seconds duration] ¿Te quedarías a tomar un té mientras suena? Creo que esta noche me gustaría recordarlo en compañía de alguien.
Disbelief blossoms into exuberant joy and heartfelt gratitude.
Wait, do not say anything yet; I have read the first line three times and I still cannot believe it. They are offering me the place, the actual place, with the little workshop overlooking the sea that we talked about. I got it... after all those applications and all those terrible interviews, somebody finally wants me to come. You stayed up with me through every doubtful evening, and I need you to know that I remember all of it. Come here, please, let me hug you before I call everyone; I want this first happy moment to be ours.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.72s · seed 20261324
CE 6.31 · genuineness 3.71 · WER 0.0% Fixed-target emotion MSE 2.145 (normalized 0.1341)
1.3 Alpha · SFT
Actual 34.72s · seed 20261324
CE 6.48 · genuineness 3.96 · WER 0.0% Fixed-target emotion MSE 1.298 (normalized 0.0811)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261324
CE 6.51 · genuineness 4.20 · WER 0.0% Fixed-target emotion MSE 1.323 (normalized 0.0827)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.48 · genuineness 4.20 · WER 0.0% Fixed-target emotion MSE 1.452 (normalized 0.0908)
1.3 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.43 · genuineness 3.90 · WER 0.0% Fixed-target emotion MSE 0.629 (normalized 0.0393)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261324
CE 6.50 · genuineness 4.01 · WER 0.0% Fixed-target emotion MSE 0.740 (normalized 0.0462)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Disbelief blossoms into exuberant joy and heartfelt gratitude.
SCRIPT:
(very quiet disbelief, rereading the news aloud) [6.4 seconds duration] Wait, do not say anything yet; I have read the first line three times and I still cannot believe it.
[0.3 seconds pause]
(hope rising, slightly breathless with excitement) [6.5 seconds duration] They are offering me the place, the actual place, with the little workshop overlooking the sea that we talked about.
[0.3 seconds pause]
(breathy giggle, 0.4 seconds)
(exuberantly delighted, a spontaneous bright laugh nearby) [6.8 seconds duration] I got it... after all those applications and all those terrible interviews, somebody finally wants me to come.
[0.3 seconds pause]
(overwhelming gratitude, let the smile soften) [6.8 seconds duration] You stayed up with me through every doubtful evening, and I need you to know that I remember all of it.
[0.3 seconds pause]
(warmly elated, sharing rather than announcing) [6.7 seconds duration] Come here, please, let me hug you before I call everyone; I want this first happy moment to be ours.
Disbelief blossoms into exuberant joy and heartfelt gratitude.
Warte, sag bitte noch nichts; ich habe den ersten Satz jetzt dreimal gelesen und kann es immer noch nicht glauben. Sie bieten mir den Platz an, wirklich diesen Platz mit der kleinen Werkstatt am Meer, von der wir gesprochen haben. Ich habe es geschafft... nach all den Bewerbungen und diesen schrecklichen Gesprächen möchte endlich jemand, dass ich komme. Du bist an jedem zweifelnden Abend bei mir geblieben, und ich möchte, dass du weißt, wie viel mir das bedeutet. Komm her, lass dich kurz umarmen, bevor ich alle anrufe; diesen ersten glücklichen Moment möchte ich nur mit dir teilen.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.38 · genuineness 4.43 · WER 2.0% Fixed-target emotion MSE 2.747 (normalized 0.1717)
1.3 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.44 · genuineness 4.27 · WER 0.0% Fixed-target emotion MSE 1.346 (normalized 0.0841)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261324
CE 6.26 · genuineness 4.49 · WER 0.0% Fixed-target emotion MSE 2.075 (normalized 0.1297)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.32 · genuineness 4.62 · WER 0.0% Fixed-target emotion MSE 3.806 (normalized 0.2379)
1.3 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.40 · genuineness 4.49 · WER 2.0% Fixed-target emotion MSE 2.418 (normalized 0.1511)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261324
CE 6.51 · genuineness 4.43 · WER 0.0% Fixed-target emotion MSE 1.902 (normalized 0.1189)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Disbelief blossoms into exuberant joy and heartfelt gratitude.
SCRIPT:
(very quiet disbelief, rereading the news aloud) [6.4 seconds duration] Warte, sag bitte noch nichts; ich habe den ersten Satz jetzt dreimal gelesen und kann es immer noch nicht glauben.
[0.3 seconds pause]
(hope rising, slightly breathless with excitement) [6.5 seconds duration] Sie bieten mir den Platz an, wirklich diesen Platz mit der kleinen Werkstatt am Meer, von der wir gesprochen haben.
[0.3 seconds pause]
(breathy giggle, 0.4 seconds)
(exuberantly delighted, a spontaneous bright laugh nearby) [6.8 seconds duration] Ich habe es geschafft... nach all den Bewerbungen und diesen schrecklichen Gesprächen möchte endlich jemand, dass ich komme.
[0.3 seconds pause]
(overwhelming gratitude, let the smile soften) [6.8 seconds duration] Du bist an jedem zweifelnden Abend bei mir geblieben, und ich möchte, dass du weißt, wie viel mir das bedeutet.
[0.3 seconds pause]
(warmly elated, sharing rather than announcing) [6.7 seconds duration] Komm her, lass dich kurz umarmen, bevor ich alle anrufe; diesen ersten glücklichen Moment möchte ich nur mit dir teilen.
Disbelief blossoms into exuberant joy and heartfelt gratitude.
Wacht, zeg alsjeblieft nog even niets; ik heb die eerste regel nu drie keer gelezen en ik geloof het nog steeds niet. Ze bieden me die plek aan, echt die plek met het kleine atelier aan zee waar we het over hadden. Het is gelukt... na al die sollicitaties en die verschrikkelijke gesprekken wil er eindelijk iemand dat ik kom. Jij bent op al die avonden vol twijfel bij me gebleven, en ik wil dat je weet hoeveel dat betekent. Kom hier, laat me je even vasthouden voordat ik iedereen bel; dit eerste gelukkige moment wil ik met jou delen.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.19 · genuineness 4.13 · WER 1.0% Fixed-target emotion MSE 3.245 (normalized 0.2028)
1.3 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.37 · genuineness 4.39 · WER 0.0% Fixed-target emotion MSE 3.009 (normalized 0.1881)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261324
CE 6.32 · genuineness 4.18 · WER 0.0% Fixed-target emotion MSE 2.974 (normalized 0.1859)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.24 · genuineness 4.00 · WER 1.0% Fixed-target emotion MSE 4.059 (normalized 0.2537)
1.3 Alpha · SFT
Actual 34.72s · seed 20261324
CE 6.26 · genuineness 4.02 · WER 0.0% Fixed-target emotion MSE 2.498 (normalized 0.1561)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261324
CE 6.23 · genuineness 4.08 · WER 2.0% Fixed-target emotion MSE 2.620 (normalized 0.1638)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Disbelief blossoms into exuberant joy and heartfelt gratitude.
SCRIPT:
(very quiet disbelief, rereading the news aloud) [6.4 seconds duration] Wacht, zeg alsjeblieft nog even niets; ik heb die eerste regel nu drie keer gelezen en ik geloof het nog steeds niet.
[0.3 seconds pause]
(hope rising, slightly breathless with excitement) [6.5 seconds duration] Ze bieden me die plek aan, echt die plek met het kleine atelier aan zee waar we het over hadden.
[0.3 seconds pause]
(breathy giggle, 0.4 seconds)
(exuberantly delighted, a spontaneous bright laugh nearby) [6.8 seconds duration] Het is gelukt... na al die sollicitaties en die verschrikkelijke gesprekken wil er eindelijk iemand dat ik kom.
[0.3 seconds pause]
(overwhelming gratitude, let the smile soften) [6.8 seconds duration] Jij bent op al die avonden vol twijfel bij me gebleven, en ik wil dat je weet hoeveel dat betekent.
[0.3 seconds pause]
(warmly elated, sharing rather than announcing) [6.7 seconds duration] Kom hier, laat me je even vasthouden voordat ik iedereen bel; dit eerste gelukkige moment wil ik met jou delen.
Disbelief blossoms into exuberant joy and heartfelt gratitude.
Attends, ne dis rien tout de suite; j'ai relu la première phrase trois fois et je n'arrive toujours pas à y croire. Ils me proposent la place, vraiment cette place avec le petit atelier face à la mer dont nous avions parlé. Ça a marché... après toutes ces candidatures et ces entretiens affreux, quelqu'un veut enfin que je vienne travailler là-bas. Tu es resté près de moi pendant toutes ces soirées de doute, et je veux que tu saches que je n'oublie rien. Viens, laisse-moi te serrer dans mes bras avant d'appeler tout le monde; ce premier moment de bonheur, je le veux avec toi.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.21 · genuineness 4.28 · WER 2.8% Fixed-target emotion MSE 3.813 (normalized 0.2383)
1.3 Alpha · SFT
Actual 34.88s · seed 20261324
CE 5.69 · genuineness 4.34 · WER 24.3% Fixed-target emotion MSE 3.875 (normalized 0.2422)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261324
CE 6.40 · genuineness 4.48 · WER 2.8% Fixed-target emotion MSE 2.187 (normalized 0.1367)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.28 · genuineness 4.36 · WER 1.9% Fixed-target emotion MSE 3.782 (normalized 0.2364)
1.3 Alpha · SFT
Actual 34.96s · seed 20261324
CE 5.83 · genuineness 4.42 · WER 23.4% Fixed-target emotion MSE 4.049 (normalized 0.2531)
1.4 Alpha · rank-128 RL
Actual 34.96s · seed 20261324
CE 5.83 · genuineness 4.57 · WER 23.4% Fixed-target emotion MSE 3.483 (normalized 0.2177)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Disbelief blossoms into exuberant joy and heartfelt gratitude.
SCRIPT:
(very quiet disbelief, rereading the news aloud) [6.4 seconds duration] Attends, ne dis rien tout de suite; j'ai relu la première phrase trois fois et je n'arrive toujours pas à y croire.
[0.3 seconds pause]
(hope rising, slightly breathless with excitement) [6.5 seconds duration] Ils me proposent la place, vraiment cette place avec le petit atelier face à la mer dont nous avions parlé.
[0.3 seconds pause]
(breathy giggle, 0.4 seconds)
(exuberantly delighted, a spontaneous bright laugh nearby) [6.8 seconds duration] Ça a marché... après toutes ces candidatures et ces entretiens affreux, quelqu'un veut enfin que je vienne travailler là-bas.
[0.3 seconds pause]
(overwhelming gratitude, let the smile soften) [6.8 seconds duration] Tu es resté près de moi pendant toutes ces soirées de doute, et je veux que tu saches que je n'oublie rien.
[0.3 seconds pause]
(warmly elated, sharing rather than announcing) [6.7 seconds duration] Viens, laisse-moi te serrer dans mes bras avant d'appeler tout le monde; ce premier moment de bonheur, je le veux avec toi.
Disbelief blossoms into exuberant joy and heartfelt gratitude.
Espera, no digas nada todavía; he leído la primera frase tres veces y sigo sin poder creer lo que pone. Me ofrecen la plaza, de verdad, esa plaza con el pequeño taller frente al mar del que tanto habíamos hablado. Lo he conseguido... después de tantas solicitudes y tantas entrevistas horribles, por fin hay alguien que quiere que vaya. Tú te quedaste conmigo todas esas noches llenas de dudas, y necesito que sepas que me acuerdo de cada una. Ven, déjame abrazarte antes de llamar a los demás; quiero que este primer momento de felicidad sea solamente nuestro.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.21 · genuineness 4.40 · WER 0.0% Fixed-target emotion MSE 3.972 (normalized 0.2482)
1.3 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.14 · genuineness 4.03 · WER 0.0% Fixed-target emotion MSE 2.881 (normalized 0.1800)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261324
CE 6.26 · genuineness 4.36 · WER 0.0% Fixed-target emotion MSE 1.598 (normalized 0.0999)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.23 · genuineness 4.12 · WER 0.0% Fixed-target emotion MSE 3.572 (normalized 0.2233)
1.3 Alpha · SFT
Actual 34.80s · seed 20261324
CE 6.21 · genuineness 3.94 · WER 0.0% Fixed-target emotion MSE 2.434 (normalized 0.1521)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261324
CE 6.18 · genuineness 4.13 · WER 1.0% Fixed-target emotion MSE 1.162 (normalized 0.0726)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Disbelief blossoms into exuberant joy and heartfelt gratitude.
SCRIPT:
(very quiet disbelief, rereading the news aloud) [6.4 seconds duration] Espera, no digas nada todavía; he leído la primera frase tres veces y sigo sin poder creer lo que pone.
[0.3 seconds pause]
(hope rising, slightly breathless with excitement) [6.5 seconds duration] Me ofrecen la plaza, de verdad, esa plaza con el pequeño taller frente al mar del que tanto habíamos hablado.
[0.3 seconds pause]
(breathy giggle, 0.4 seconds)
(exuberantly delighted, a spontaneous bright laugh nearby) [6.8 seconds duration] Lo he conseguido... después de tantas solicitudes y tantas entrevistas horribles, por fin hay alguien que quiere que vaya.
[0.3 seconds pause]
(overwhelming gratitude, let the smile soften) [6.8 seconds duration] Tú te quedaste conmigo todas esas noches llenas de dudas, y necesito que sepas que me acuerdo de cada una.
[0.3 seconds pause]
(warmly elated, sharing rather than announcing) [6.7 seconds duration] Ven, déjame abrazarte antes de llamar a los demás; quiero que este primer momento de felicidad sea solamente nuestro.
Conversational storytelling turns into embarrassed laughter and affectionate amusement.
So I walked into the meeting feeling very professional, carrying my notes and the expensive coffee I had bought to look organized. Everyone went quiet when I connected the laptop, and I honestly thought they were impressed by how prepared I was. Then I turned around... my entire presentation was the photograph of your cat wearing that enormous orange birthday hat. I tried to explain, but the director asked whether our new strategy involved more naps, and that was the end of me. Anyway, your cat is apparently our unofficial consultant now, so please tell him I expect a very reasonable invoice.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.72s · seed 20261425
CE 6.47 · genuineness 4.23 · WER 0.0% Fixed-target emotion MSE 1.489 (normalized 0.0930)
1.3 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.54 · genuineness 4.25 · WER 2.9% Fixed-target emotion MSE 0.956 (normalized 0.0597)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261425
CE 6.56 · genuineness 4.29 · WER 0.0% Fixed-target emotion MSE 1.500 (normalized 0.0938)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.72s · seed 20261425
CE 6.39 · genuineness 3.85 · WER 0.0% Fixed-target emotion MSE 1.205 (normalized 0.0753)
1.3 Alpha · SFT
Actual 34.72s · seed 20261425
CE 6.39 · genuineness 3.76 · WER 0.0% Fixed-target emotion MSE 1.743 (normalized 0.1090)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261425
CE 6.49 · genuineness 4.17 · WER 0.0% Fixed-target emotion MSE 0.887 (normalized 0.0555)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Conversational storytelling turns into embarrassed laughter and affectionate amusement.
SCRIPT:
(casual storytelling, confidently amused) [6.4 seconds duration] So I walked into the meeting feeling very professional, carrying my notes and the expensive coffee I had bought to look organized.
[0.3 seconds pause]
(playful suspense, trying to remain serious) [6.5 seconds duration] Everyone went quiet when I connected the laptop, and I honestly thought they were impressed by how prepared I was.
[0.3 seconds pause]
(chuckle, 0.4 seconds)
(embarrassed realization, a tiny hesitation) [6.8 seconds duration] Then I turned around... my entire presentation was the photograph of your cat wearing that enormous orange birthday hat.
[0.3 seconds pause]
(warm unrestrained amusement, laughing through the memory) [6.8 seconds duration] I tried to explain, but the director asked whether our new strategy involved more naps, and that was the end of me.
[0.3 seconds pause]
(affectionate and self-deprecating, winding down naturally) [6.7 seconds duration] Anyway, your cat is apparently our unofficial consultant now, so please tell him I expect a very reasonable invoice.
Conversational storytelling turns into embarrassed laughter and affectionate amusement.
Also, ich kam ziemlich selbstbewusst in die Besprechung, mit meinen Notizen und diesem teuren Kaffee, damit alles schön organisiert aussieht. Als ich den Laptop anschloss, wurden plötzlich alle still, und ich dachte tatsächlich, sie wären von meiner Vorbereitung beeindruckt. Dann habe ich mich umgedreht... auf der ganzen Leinwand war dein Kater mit diesem riesigen orangefarbenen Geburtstagshut zu sehen. Ich wollte es erklären, aber der Chef fragte, ob unsere neue Strategie mehr Mittagsschlaf vorsieht, und da war es vorbei. Jedenfalls ist dein Kater jetzt unser inoffizieller Berater; sag ihm bitte, dass ich eine wirklich faire Rechnung erwarte.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.72s · seed 20261425
CE 6.25 · genuineness 4.46 · WER 3.1% Fixed-target emotion MSE 1.753 (normalized 0.1096)
1.3 Alpha · SFT
Actual 34.72s · seed 20261425
CE 6.45 · genuineness 4.39 · WER 0.0% Fixed-target emotion MSE 1.431 (normalized 0.0894)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261425
CE 6.43 · genuineness 4.19 · WER 2.1% Fixed-target emotion MSE 1.848 (normalized 0.1155)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.72s · seed 20261425
CE 6.44 · genuineness 4.34 · WER 1.0% Fixed-target emotion MSE 1.235 (normalized 0.0772)
1.3 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.44 · genuineness 4.26 · WER 0.0% Fixed-target emotion MSE 1.374 (normalized 0.0859)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261425
CE 6.41 · genuineness 4.49 · WER 4.2% Fixed-target emotion MSE 2.125 (normalized 0.1328)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Conversational storytelling turns into embarrassed laughter and affectionate amusement.
SCRIPT:
(casual storytelling, confidently amused) [6.4 seconds duration] Also, ich kam ziemlich selbstbewusst in die Besprechung, mit meinen Notizen und diesem teuren Kaffee, damit alles schön organisiert aussieht.
[0.3 seconds pause]
(playful suspense, trying to remain serious) [6.5 seconds duration] Als ich den Laptop anschloss, wurden plötzlich alle still, und ich dachte tatsächlich, sie wären von meiner Vorbereitung beeindruckt.
[0.3 seconds pause]
(chuckle, 0.4 seconds)
(embarrassed realization, a tiny hesitation) [6.8 seconds duration] Dann habe ich mich umgedreht... auf der ganzen Leinwand war dein Kater mit diesem riesigen orangefarbenen Geburtstagshut zu sehen.
[0.3 seconds pause]
(warm unrestrained amusement, laughing through the memory) [6.8 seconds duration] Ich wollte es erklären, aber der Chef fragte, ob unsere neue Strategie mehr Mittagsschlaf vorsieht, und da war es vorbei.
[0.3 seconds pause]
(affectionate and self-deprecating, winding down naturally) [6.7 seconds duration] Jedenfalls ist dein Kater jetzt unser inoffizieller Berater; sag ihm bitte, dass ich eine wirklich faire Rechnung erwarte.
Conversational storytelling turns into embarrassed laughter and affectionate amusement.
Dus ik liep heel zelfverzekerd die vergadering binnen, met mijn notities en dure koffie, zodat ik lekker georganiseerd zou lijken. Toen ik mijn laptop aansloot, werd iedereen ineens stil, en ik dacht werkelijk dat ze onder de indruk waren. Toen draaide ik me om... jouw kat stond levensgroot op het scherm, met die belachelijk grote oranje verjaardagshoed op. Ik wilde het uitleggen, maar de directeur vroeg of onze nieuwe strategie meer middagdutjes inhield, en toen hield ik het niet meer. Hoe dan ook, jouw kat is nu onze onofficiële adviseur, dus zeg hem alsjeblieft dat ik een redelijke rekening verwacht.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.19 · genuineness 4.24 · WER 3.0% Fixed-target emotion MSE 1.927 (normalized 0.1204)
1.3 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.28 · genuineness 4.16 · WER 3.0% Fixed-target emotion MSE 1.807 (normalized 0.1129)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261425
CE 6.33 · genuineness 4.21 · WER 2.0% Fixed-target emotion MSE 1.168 (normalized 0.0730)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.13 · genuineness 4.20 · WER 3.0% Fixed-target emotion MSE 1.810 (normalized 0.1131)
1.3 Alpha · SFT
Actual 34.72s · seed 20261425
CE 6.19 · genuineness 3.67 · WER 3.0% Fixed-target emotion MSE 2.920 (normalized 0.1825)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261425
CE 6.26 · genuineness 3.91 · WER 3.0% Fixed-target emotion MSE 1.934 (normalized 0.1208)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Conversational storytelling turns into embarrassed laughter and affectionate amusement.
SCRIPT:
(casual storytelling, confidently amused) [6.4 seconds duration] Dus ik liep heel zelfverzekerd die vergadering binnen, met mijn notities en dure koffie, zodat ik lekker georganiseerd zou lijken.
[0.3 seconds pause]
(playful suspense, trying to remain serious) [6.5 seconds duration] Toen ik mijn laptop aansloot, werd iedereen ineens stil, en ik dacht werkelijk dat ze onder de indruk waren.
[0.3 seconds pause]
(chuckle, 0.4 seconds)
(embarrassed realization, a tiny hesitation) [6.8 seconds duration] Toen draaide ik me om... jouw kat stond levensgroot op het scherm, met die belachelijk grote oranje verjaardagshoed op.
[0.3 seconds pause]
(warm unrestrained amusement, laughing through the memory) [6.8 seconds duration] Ik wilde het uitleggen, maar de directeur vroeg of onze nieuwe strategie meer middagdutjes inhield, en toen hield ik het niet meer.
[0.3 seconds pause]
(affectionate and self-deprecating, winding down naturally) [6.7 seconds duration] Hoe dan ook, jouw kat is nu onze onofficiële adviseur, dus zeg hem alsjeblieft dat ik een redelijke rekening verwacht.
Conversational storytelling turns into embarrassed laughter and affectionate amusement.
Alors, je suis arrivé à la réunion avec mes notes et un café beaucoup trop cher pour avoir l'air parfaitement organisé. Tout le monde s'est tu quand j'ai branché mon ordinateur, et j'ai vraiment cru que ma préparation les impressionnait. Puis je me suis retourné... sur tout l'écran, il y avait ton chat avec cet immense chapeau d'anniversaire orange. J'ai voulu expliquer, mais la directrice a demandé si notre nouvelle stratégie prévoyait davantage de siestes, et là, j'ai craqué. Bref, ton chat est devenu notre conseiller officieux; dis-lui quand même que j'espère recevoir une facture tout à fait raisonnable.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.72s · seed 20261425
CE 6.23 · genuineness 4.36 · WER 6.0% Fixed-target emotion MSE 2.908 (normalized 0.1818)
1.3 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.16 · genuineness 4.07 · WER 94.0% Fixed-target emotion MSE 2.431 (normalized 0.1520)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261425
CE 6.23 · genuineness 4.32 · WER 95.0% Fixed-target emotion MSE 1.486 (normalized 0.0929)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.72s · seed 20261425
CE 5.92 · genuineness 4.29 · WER 3.0% Fixed-target emotion MSE 2.817 (normalized 0.1761)
1.3 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.19 · genuineness 4.25 · WER 4.0% Fixed-target emotion MSE 2.530 (normalized 0.1581)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261425
CE 6.14 · genuineness 4.52 · WER 3.0% Fixed-target emotion MSE 2.699 (normalized 0.1687)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Conversational storytelling turns into embarrassed laughter and affectionate amusement.
SCRIPT:
(casual storytelling, confidently amused) [6.4 seconds duration] Alors, je suis arrivé à la réunion avec mes notes et un café beaucoup trop cher pour avoir l'air parfaitement organisé.
[0.3 seconds pause]
(playful suspense, trying to remain serious) [6.5 seconds duration] Tout le monde s'est tu quand j'ai branché mon ordinateur, et j'ai vraiment cru que ma préparation les impressionnait.
[0.3 seconds pause]
(chuckle, 0.4 seconds)
(embarrassed realization, a tiny hesitation) [6.8 seconds duration] Puis je me suis retourné... sur tout l'écran, il y avait ton chat avec cet immense chapeau d'anniversaire orange.
[0.3 seconds pause]
(warm unrestrained amusement, laughing through the memory) [6.8 seconds duration] J'ai voulu expliquer, mais la directrice a demandé si notre nouvelle stratégie prévoyait davantage de siestes, et là, j'ai craqué.
[0.3 seconds pause]
(affectionate and self-deprecating, winding down naturally) [6.7 seconds duration] Bref, ton chat est devenu notre conseiller officieux; dis-lui quand même que j'espère recevoir une facture tout à fait raisonnable.
Conversational storytelling turns into embarrassed laughter and affectionate amusement.
Pues entré en la reunión sintiéndome muy profesional, con mis notas y un café carísimo para parecer una persona organizada. Todos se quedaron callados cuando conecté el portátil, y yo pensé de verdad que estaban impresionados con mi preparación. Entonces me di la vuelta... toda la pantalla mostraba a tu gato llevando aquel enorme sombrero naranja de cumpleaños. Intenté explicarlo, pero la directora preguntó si nuestra nueva estrategia incluía más siestas, y ahí ya no pude aguantarme. En fin, tu gato es ahora nuestro asesor extraoficial; dile, por favor, que espero una factura bastante razonable por sus servicios.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.04 · genuineness 4.23 · WER 0.0% Fixed-target emotion MSE 2.515 (normalized 0.1572)
1.3 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.25 · genuineness 4.05 · WER 1.0% Fixed-target emotion MSE 2.551 (normalized 0.1594)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261425
CE 6.17 · genuineness 3.89 · WER 1.0% Fixed-target emotion MSE 1.358 (normalized 0.0849)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.72s · seed 20261425
CE 6.12 · genuineness 3.73 · WER 0.0% Fixed-target emotion MSE 1.977 (normalized 0.1236)
1.3 Alpha · SFT
Actual 34.80s · seed 20261425
CE 6.06 · genuineness 3.67 · WER 0.0% Fixed-target emotion MSE 2.370 (normalized 0.1482)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261425
CE 6.26 · genuineness 3.94 · WER 1.0% Fixed-target emotion MSE 1.340 (normalized 0.0837)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Conversational storytelling turns into embarrassed laughter and affectionate amusement.
SCRIPT:
(casual storytelling, confidently amused) [6.4 seconds duration] Pues entré en la reunión sintiéndome muy profesional, con mis notas y un café carísimo para parecer una persona organizada.
[0.3 seconds pause]
(playful suspense, trying to remain serious) [6.5 seconds duration] Todos se quedaron callados cuando conecté el portátil, y yo pensé de verdad que estaban impresionados con mi preparación.
[0.3 seconds pause]
(chuckle, 0.4 seconds)
(embarrassed realization, a tiny hesitation) [6.8 seconds duration] Entonces me di la vuelta... toda la pantalla mostraba a tu gato llevando aquel enorme sombrero naranja de cumpleaños.
[0.3 seconds pause]
(warm unrestrained amusement, laughing through the memory) [6.8 seconds duration] Intenté explicarlo, pero la directora preguntó si nuestra nueva estrategia incluía más siestas, y ahí ya no pude aguantarme.
[0.3 seconds pause]
(affectionate and self-deprecating, winding down naturally) [6.7 seconds duration] En fin, tu gato es ahora nuestro asesor extraoficial; dile, por favor, que espero una factura bastante razonable por sus servicios.
Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
I saw the announcement this morning, and at first I thought there must be another page with the rest of our names. But there was only yours, underneath the idea we stayed here building together while everyone else had already gone home. You even used the little phrase I wrote on the window... that is almost funny, in a really miserable sort of way. I am not asking you to apologize beautifully; I am asking you to correct the announcement and give the team its credit. After that, we can talk about working together again, but not before you understand why I cannot simply smile this away.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.72s · seed 20261526
CE 6.55 · genuineness 4.19 · WER 0.0% Fixed-target emotion MSE 2.400 (normalized 0.1500)
1.3 Alpha · SFT
Actual 34.72s · seed 20261526
CE 6.50 · genuineness 4.15 · WER 0.0% Fixed-target emotion MSE 2.386 (normalized 0.1491)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261526
CE 6.47 · genuineness 4.15 · WER 0.0% Fixed-target emotion MSE 2.763 (normalized 0.1727)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.42 · genuineness 3.97 · WER 0.0% Fixed-target emotion MSE 2.723 (normalized 0.1702)
1.3 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.46 · genuineness 3.90 · WER 0.9% Fixed-target emotion MSE 2.694 (normalized 0.1683)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261526
CE 6.40 · genuineness 3.90 · WER 0.9% Fixed-target emotion MSE 2.591 (normalized 0.1619)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
SCRIPT:
(carefully polite, hurt held beneath the surface) [6.4 seconds duration] I saw the announcement this morning, and at first I thought there must be another page with the rest of our names.
[0.3 seconds pause]
(disbelief, a little colder and slower) [6.5 seconds duration] But there was only yours, underneath the idea we stayed here building together while everyone else had already gone home.
[0.3 seconds pause]
(low bitter chuckle, 0.4 seconds)
(bitter and wounded, not villainous) [6.8 seconds duration] You even used the little phrase I wrote on the window... that is almost funny, in a really miserable sort of way.
[0.3 seconds pause]
(firm, controlled anger, clearly articulated) [6.8 seconds duration] I am not asking you to apologize beautifully; I am asking you to correct the announcement and give the team its credit.
[0.3 seconds pause]
(quiet dignity, a boundary stated without shouting) [6.7 seconds duration] After that, we can talk about working together again, but not before you understand why I cannot simply smile this away.
[('emo_Bitterness', 2.0), ('emo_Anger', 1.5)]
The name on the project · German
Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
Ich habe heute Morgen die Ankündigung gesehen und dachte zuerst, es gäbe noch eine zweite Seite mit unseren anderen Namen. Aber da stand nur deiner, direkt unter der Idee, an der wir hier gemeinsam gearbeitet haben, als längst alle anderen zu Hause waren. Du hast sogar den kleinen Satz übernommen, den ich ans Fenster geschrieben habe... irgendwie ist das beinahe schon komisch. Ich verlange keine wunderschöne Entschuldigung; ich möchte, dass du die Ankündigung korrigierst und die Arbeit des ganzen Teams anerkennst. Danach können wir darüber reden, ob wir wieder zusammenarbeiten, aber erst musst du verstehen, warum ich das nicht einfach weglächeln kann.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.33 · genuineness 4.41 · WER 0.0% Fixed-target emotion MSE 1.784 (normalized 0.1115)
1.3 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.45 · genuineness 4.36 · WER 1.0% Fixed-target emotion MSE 2.339 (normalized 0.1462)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261526
CE 6.42 · genuineness 4.37 · WER 1.0% Fixed-target emotion MSE 2.860 (normalized 0.1787)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.49 · genuineness 4.50 · WER 2.0% Fixed-target emotion MSE 2.456 (normalized 0.1535)
1.3 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.45 · genuineness 4.46 · WER 1.0% Fixed-target emotion MSE 2.451 (normalized 0.1532)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261526
CE 6.47 · genuineness 4.33 · WER 1.0% Fixed-target emotion MSE 2.819 (normalized 0.1762)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
SCRIPT:
(carefully polite, hurt held beneath the surface) [6.4 seconds duration] Ich habe heute Morgen die Ankündigung gesehen und dachte zuerst, es gäbe noch eine zweite Seite mit unseren anderen Namen.
[0.3 seconds pause]
(disbelief, a little colder and slower) [6.5 seconds duration] Aber da stand nur deiner, direkt unter der Idee, an der wir hier gemeinsam gearbeitet haben, als längst alle anderen zu Hause waren.
[0.3 seconds pause]
(low bitter chuckle, 0.4 seconds)
(bitter and wounded, not villainous) [6.8 seconds duration] Du hast sogar den kleinen Satz übernommen, den ich ans Fenster geschrieben habe... irgendwie ist das beinahe schon komisch.
[0.3 seconds pause]
(firm, controlled anger, clearly articulated) [6.8 seconds duration] Ich verlange keine wunderschöne Entschuldigung; ich möchte, dass du die Ankündigung korrigierst und die Arbeit des ganzen Teams anerkennst.
[0.3 seconds pause]
(quiet dignity, a boundary stated without shouting) [6.7 seconds duration] Danach können wir darüber reden, ob wir wieder zusammenarbeiten, aber erst musst du verstehen, warum ich das nicht einfach weglächeln kann.
[('emo_Bitterness', 2.0), ('emo_Anger', 1.5)]
The name on the project · Dutch
Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
Ik zag de aankondiging vanochtend en dacht eerst dat er vast nog een tweede pagina met onze andere namen was. Maar alleen jouw naam stond er, onder het idee waar wij samen aan werkten terwijl iedereen al lang naar huis was. Je hebt zelfs dat zinnetje gebruikt dat ik op het raam schreef... op een heel ellendige manier is dat bijna grappig. Ik vraag niet om een prachtig excuus; ik wil dat je de aankondiging verbetert en het hele team de eer geeft. Daarna kunnen we praten over opnieuw samenwerken, maar eerst moet je begrijpen waarom ik dit niet zomaar kan weglachen.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.30 · genuineness 4.22 · WER 2.9% Fixed-target emotion MSE 1.781 (normalized 0.1113)
1.3 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.31 · genuineness 4.31 · WER 0.0% Fixed-target emotion MSE 2.438 (normalized 0.1524)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261526
CE 6.22 · genuineness 4.31 · WER 2.9% Fixed-target emotion MSE 2.795 (normalized 0.1747)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.28 · genuineness 4.17 · WER 2.0% Fixed-target emotion MSE 2.679 (normalized 0.1674)
1.3 Alpha · SFT
Actual 34.72s · seed 20261526
CE 6.31 · genuineness 4.14 · WER 0.0% Fixed-target emotion MSE 2.757 (normalized 0.1723)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261526
CE 6.28 · genuineness 4.35 · WER 0.0% Fixed-target emotion MSE 2.104 (normalized 0.1315)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
SCRIPT:
(carefully polite, hurt held beneath the surface) [6.4 seconds duration] Ik zag de aankondiging vanochtend en dacht eerst dat er vast nog een tweede pagina met onze andere namen was.
[0.3 seconds pause]
(disbelief, a little colder and slower) [6.5 seconds duration] Maar alleen jouw naam stond er, onder het idee waar wij samen aan werkten terwijl iedereen al lang naar huis was.
[0.3 seconds pause]
(low bitter chuckle, 0.4 seconds)
(bitter and wounded, not villainous) [6.8 seconds duration] Je hebt zelfs dat zinnetje gebruikt dat ik op het raam schreef... op een heel ellendige manier is dat bijna grappig.
[0.3 seconds pause]
(firm, controlled anger, clearly articulated) [6.8 seconds duration] Ik vraag niet om een prachtig excuus; ik wil dat je de aankondiging verbetert en het hele team de eer geeft.
[0.3 seconds pause]
(quiet dignity, a boundary stated without shouting) [6.7 seconds duration] Daarna kunnen we praten over opnieuw samenwerken, maar eerst moet je begrijpen waarom ik dit niet zomaar kan weglachen.
[('emo_Bitterness', 2.0), ('emo_Anger', 1.5)]
The name on the project · French
Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
J'ai vu l'annonce ce matin, et j'ai d'abord pensé qu'il devait y avoir une autre page avec les noms de notre équipe. Mais il n'y avait que le tien, sous cette idée que nous avons construite ensemble quand tout le monde était déjà parti. Tu as même repris la petite phrase que j'avais écrite sur la fenêtre... c'est presque drôle, d'une façon vraiment triste. Je ne te demande pas de belles excuses; je te demande de corriger l'annonce et de reconnaître le travail de toute l'équipe. Ensuite, nous pourrons parler de travailler à nouveau ensemble, mais comprends d'abord pourquoi je ne peux pas simplement sourire et oublier.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.20 · genuineness 4.33 · WER 0.0% Fixed-target emotion MSE 1.891 (normalized 0.1182)
1.3 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.21 · genuineness 4.38 · WER 0.0% Fixed-target emotion MSE 2.305 (normalized 0.1441)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261526
CE 6.01 · genuineness 4.39 · WER 0.9% Fixed-target emotion MSE 2.239 (normalized 0.1400)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.13 · genuineness 4.39 · WER 0.0% Fixed-target emotion MSE 2.327 (normalized 0.1455)
1.3 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.20 · genuineness 4.28 · WER 0.0% Fixed-target emotion MSE 2.430 (normalized 0.1519)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261526
CE 6.17 · genuineness 4.24 · WER 0.0% Fixed-target emotion MSE 2.751 (normalized 0.1719)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
SCRIPT:
(carefully polite, hurt held beneath the surface) [6.4 seconds duration] J'ai vu l'annonce ce matin, et j'ai d'abord pensé qu'il devait y avoir une autre page avec les noms de notre équipe.
[0.3 seconds pause]
(disbelief, a little colder and slower) [6.5 seconds duration] Mais il n'y avait que le tien, sous cette idée que nous avons construite ensemble quand tout le monde était déjà parti.
[0.3 seconds pause]
(low bitter chuckle, 0.4 seconds)
(bitter and wounded, not villainous) [6.8 seconds duration] Tu as même repris la petite phrase que j'avais écrite sur la fenêtre... c'est presque drôle, d'une façon vraiment triste.
[0.3 seconds pause]
(firm, controlled anger, clearly articulated) [6.8 seconds duration] Je ne te demande pas de belles excuses; je te demande de corriger l'annonce et de reconnaître le travail de toute l'équipe.
[0.3 seconds pause]
(quiet dignity, a boundary stated without shouting) [6.7 seconds duration] Ensuite, nous pourrons parler de travailler à nouveau ensemble, mais comprends d'abord pourquoi je ne peux pas simplement sourire et oublier.
[('emo_Bitterness', 2.0), ('emo_Anger', 1.5)]
The name on the project · Spanish
Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
Vi el anuncio esta mañana y al principio pensé que habría otra página con los nombres de las demás personas del equipo. Pero solamente estaba el tuyo, debajo de la idea que construimos juntos cuando todos los demás ya se habían marchado a casa. Hasta utilizaste aquella frase que escribí en la ventana... casi tiene gracia, pero de una manera bastante triste, la verdad. No te estoy pidiendo una disculpa preciosa; te estoy pidiendo que corrijas el anuncio y reconozcas el trabajo de todo el equipo. Después podremos hablar de volver a trabajar juntos, pero primero tienes que entender por qué no puedo sonreír y olvidarlo sin más.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261526
CE 5.86 · genuineness 3.97 · WER 0.0% Fixed-target emotion MSE 1.134 (normalized 0.0709)
1.3 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.21 · genuineness 4.20 · WER 0.0% Fixed-target emotion MSE 1.420 (normalized 0.0887)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261526
CE 6.27 · genuineness 4.30 · WER 0.0% Fixed-target emotion MSE 1.922 (normalized 0.1201)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.13 · genuineness 3.90 · WER 0.0% Fixed-target emotion MSE 1.215 (normalized 0.0759)
1.3 Alpha · SFT
Actual 34.80s · seed 20261526
CE 6.25 · genuineness 3.91 · WER 0.9% Fixed-target emotion MSE 2.054 (normalized 0.1284)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261526
CE 6.21 · genuineness 4.17 · WER 0.0% Fixed-target emotion MSE 2.250 (normalized 0.1406)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Polite restraint fractures into hurt and bitterness, then becomes self-respecting resolve.
SCRIPT:
(carefully polite, hurt held beneath the surface) [6.4 seconds duration] Vi el anuncio esta mañana y al principio pensé que habría otra página con los nombres de las demás personas del equipo.
[0.3 seconds pause]
(disbelief, a little colder and slower) [6.5 seconds duration] Pero solamente estaba el tuyo, debajo de la idea que construimos juntos cuando todos los demás ya se habían marchado a casa.
[0.3 seconds pause]
(low bitter chuckle, 0.4 seconds)
(bitter and wounded, not villainous) [6.8 seconds duration] Hasta utilizaste aquella frase que escribí en la ventana... casi tiene gracia, pero de una manera bastante triste, la verdad.
[0.3 seconds pause]
(firm, controlled anger, clearly articulated) [6.8 seconds duration] No te estoy pidiendo una disculpa preciosa; te estoy pidiendo que corrijas el anuncio y reconozcas el trabajo de todo el equipo.
[0.3 seconds pause]
(quiet dignity, a boundary stated without shouting) [6.7 seconds duration] Después podremos hablar de volver a trabajar juntos, pero primero tienes que entender por qué no puedo sonreír y olvidarlo sin más.
[('emo_Bitterness', 2.0), ('emo_Anger', 1.5)]
Sunrise above the clouds · English
Breathless awe, a little vulnerability and quiet, shared tenderness.
Look at the edge of the clouds, just there; I thought the sunrise would be bright, but I did not expect that color. All those months of planning, and now the whole world seems to be waking up beneath this little window beside us. I keep trying to think of something clever to say... but honestly, I feel very small, and a little overwhelmed. You do not have to say anything either; put your hand here and let us just watch it change for a moment. When we get home, I will forget the exact shape of the clouds, but I do not think I will forget being here with you.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.38 · genuineness 4.20 · WER 0.0% Fixed-target emotion MSE 0.760 (normalized 0.0475)
1.3 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.46 · genuineness 4.33 · WER 0.0% Fixed-target emotion MSE 0.549 (normalized 0.0343)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261627
CE 6.59 · genuineness 4.60 · WER 0.0% Fixed-target emotion MSE 0.857 (normalized 0.0536)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.41 · genuineness 4.30 · WER 0.0% Fixed-target emotion MSE 0.625 (normalized 0.0391)
1.3 Alpha · SFT
Actual 34.72s · seed 20261627
CE 6.48 · genuineness 4.36 · WER 0.0% Fixed-target emotion MSE 0.433 (normalized 0.0271)
1.4 Alpha · rank-128 RL
Actual 34.72s · seed 20261627
CE 6.52 · genuineness 4.31 · WER 0.0% Fixed-target emotion MSE 0.585 (normalized 0.0366)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Breathless awe, a little vulnerability and quiet, shared tenderness.
SCRIPT:
(breathless wonder, almost a whisper) [6.4 seconds duration] Look at the edge of the clouds, just there; I thought the sunrise would be bright, but I did not expect that color.
[0.3 seconds pause]
(expansive awe, gradually opening the voice) [6.5 seconds duration] All those months of planning, and now the whole world seems to be waking up beneath this little window beside us.
[0.3 seconds pause]
(soft amazed exhale, 0.4 seconds)
(vulnerable honesty, a tiny searching pause) [6.8 seconds duration] I keep trying to think of something clever to say... but honestly, I feel very small, and a little overwhelmed.
[0.3 seconds pause]
(soft affection and reassurance) [6.8 seconds duration] You do not have to say anything either; put your hand here and let us just watch it change for a moment.
[0.3 seconds pause]
(quiet inspiration, intimate rather than grandiose) [6.7 seconds duration] When we get home, I will forget the exact shape of the clouds, but I do not think I will forget being here with you.
[('emo_Awe', 2.5), ('emo_Affection', 2.0)]
Sunrise above the clouds · German
Breathless awe, a little vulnerability and quiet, shared tenderness.
Schau mal an den Rand der Wolken, genau dort; ich wusste, der Sonnenaufgang wird hell, aber mit dieser Farbe habe ich nicht gerechnet. Nach all diesen Monaten der Planung scheint jetzt die ganze Welt unter unserem kleinen Fenster langsam wieder aufzuwachen. Ich versuche die ganze Zeit, etwas Kluges zu sagen... aber ehrlich gesagt fühle ich mich gerade sehr klein und ziemlich überwältigt. Du musst auch nichts sagen; leg einfach deine Hand hierher, und lass uns einen Moment gemeinsam zusehen, wie sich alles verändert. Zu Hause werde ich die genaue Form der Wolken vergessen, aber nicht, wie es war, diesen Augenblick mit dir zu erleben.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.25 · genuineness 4.64 · WER 0.0% Fixed-target emotion MSE 1.290 (normalized 0.0806)
1.3 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.37 · genuineness 4.66 · WER 1.0% Fixed-target emotion MSE 1.445 (normalized 0.0903)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261627
CE 6.53 · genuineness 4.69 · WER 0.0% Fixed-target emotion MSE 0.559 (normalized 0.0350)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.21 · genuineness 4.50 · WER 0.0% Fixed-target emotion MSE 1.243 (normalized 0.0777)
1.3 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.28 · genuineness 4.52 · WER 0.0% Fixed-target emotion MSE 1.980 (normalized 0.1237)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261627
CE 6.43 · genuineness 4.75 · WER 0.0% Fixed-target emotion MSE 1.548 (normalized 0.0967)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Breathless awe, a little vulnerability and quiet, shared tenderness.
SCRIPT:
(breathless wonder, almost a whisper) [6.4 seconds duration] Schau mal an den Rand der Wolken, genau dort; ich wusste, der Sonnenaufgang wird hell, aber mit dieser Farbe habe ich nicht gerechnet.
[0.3 seconds pause]
(expansive awe, gradually opening the voice) [6.5 seconds duration] Nach all diesen Monaten der Planung scheint jetzt die ganze Welt unter unserem kleinen Fenster langsam wieder aufzuwachen.
[0.3 seconds pause]
(soft amazed exhale, 0.4 seconds)
(vulnerable honesty, a tiny searching pause) [6.8 seconds duration] Ich versuche die ganze Zeit, etwas Kluges zu sagen... aber ehrlich gesagt fühle ich mich gerade sehr klein und ziemlich überwältigt.
[0.3 seconds pause]
(soft affection and reassurance) [6.8 seconds duration] Du musst auch nichts sagen; leg einfach deine Hand hierher, und lass uns einen Moment gemeinsam zusehen, wie sich alles verändert.
[0.3 seconds pause]
(quiet inspiration, intimate rather than grandiose) [6.7 seconds duration] Zu Hause werde ich die genaue Form der Wolken vergessen, aber nicht, wie es war, diesen Augenblick mit dir zu erleben.
[('emo_Awe', 2.5), ('emo_Affection', 2.0)]
Sunrise above the clouds · Dutch
Breathless awe, a little vulnerability and quiet, shared tenderness.
Kijk naar de rand van de wolken, precies daar; ik wist dat de zonsopgang fel zou zijn, maar die kleur had ik niet verwacht. Na al die maanden plannen lijkt nu de hele wereld langzaam wakker te worden onder dit kleine raam naast ons. Ik probeer steeds iets verstandigs te bedenken om te zeggen... maar eerlijk gezegd voel ik me heel klein en een beetje overweldigd. Jij hoeft ook niets te zeggen; leg je hand maar hier en laten we gewoon even samen kijken hoe alles verandert. Thuis zal ik de precieze vorm van de wolken vergeten, maar volgens mij vergeet ik nooit hoe het was om hier met jou te zijn.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.72s · seed 20261627
CE 6.18 · genuineness 4.00 · WER 0.0% Fixed-target emotion MSE 4.733 (normalized 0.2958)
1.3 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.11 · genuineness 4.15 · WER 0.0% Fixed-target emotion MSE 3.609 (normalized 0.2256)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261627
CE 6.26 · genuineness 4.20 · WER 0.9% Fixed-target emotion MSE 3.509 (normalized 0.2193)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.00 · genuineness 4.22 · WER 2.7% Fixed-target emotion MSE 3.376 (normalized 0.2110)
1.3 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.24 · genuineness 4.24 · WER 0.0% Fixed-target emotion MSE 3.750 (normalized 0.2344)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261627
CE 6.30 · genuineness 4.21 · WER 1.8% Fixed-target emotion MSE 3.021 (normalized 0.1888)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Breathless awe, a little vulnerability and quiet, shared tenderness.
SCRIPT:
(breathless wonder, almost a whisper) [6.4 seconds duration] Kijk naar de rand van de wolken, precies daar; ik wist dat de zonsopgang fel zou zijn, maar die kleur had ik niet verwacht.
[0.3 seconds pause]
(expansive awe, gradually opening the voice) [6.5 seconds duration] Na al die maanden plannen lijkt nu de hele wereld langzaam wakker te worden onder dit kleine raam naast ons.
[0.3 seconds pause]
(soft amazed exhale, 0.4 seconds)
(vulnerable honesty, a tiny searching pause) [6.8 seconds duration] Ik probeer steeds iets verstandigs te bedenken om te zeggen... maar eerlijk gezegd voel ik me heel klein en een beetje overweldigd.
[0.3 seconds pause]
(soft affection and reassurance) [6.8 seconds duration] Jij hoeft ook niets te zeggen; leg je hand maar hier en laten we gewoon even samen kijken hoe alles verandert.
[0.3 seconds pause]
(quiet inspiration, intimate rather than grandiose) [6.7 seconds duration] Thuis zal ik de precieze vorm van de wolken vergeten, maar volgens mij vergeet ik nooit hoe het was om hier met jou te zijn.
[('emo_Awe', 2.5), ('emo_Affection', 2.0)]
Sunrise above the clouds · French
Breathless awe, a little vulnerability and quiet, shared tenderness.
Regarde le bord des nuages, juste là; je savais que le lever du soleil serait lumineux, mais je n'imaginais pas cette couleur. Après tous ces mois de préparation, on dirait que le monde entier se réveille doucement sous cette petite fenêtre près de nous. Je cherche quelque chose d'intelligent à dire... mais en vérité, je me sens tout petit et un peu dépassé par ce moment. Tu n'as pas besoin de parler non plus; pose ta main ici et regardons simplement la lumière changer pendant quelques instants. En rentrant, j'oublierai sans doute la forme exacte des nuages, mais je crois que je n'oublierai jamais d'avoir été ici avec toi.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.17 · genuineness 4.60 · WER 1.8% Fixed-target emotion MSE 2.616 (normalized 0.1635)
1.3 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.10 · genuineness 4.43 · WER 2.8% Fixed-target emotion MSE 2.866 (normalized 0.1791)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261627
CE 6.21 · genuineness 4.58 · WER 0.9% Fixed-target emotion MSE 1.515 (normalized 0.0947)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.08 · genuineness 4.45 · WER 0.9% Fixed-target emotion MSE 2.323 (normalized 0.1452)
1.3 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.02 · genuineness 4.46 · WER 0.9% Fixed-target emotion MSE 2.197 (normalized 0.1373)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261627
CE 6.22 · genuineness 4.68 · WER 0.9% Fixed-target emotion MSE 3.251 (normalized 0.2032)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Breathless awe, a little vulnerability and quiet, shared tenderness.
SCRIPT:
(breathless wonder, almost a whisper) [6.4 seconds duration] Regarde le bord des nuages, juste là; je savais que le lever du soleil serait lumineux, mais je n'imaginais pas cette couleur.
[0.3 seconds pause]
(expansive awe, gradually opening the voice) [6.5 seconds duration] Après tous ces mois de préparation, on dirait que le monde entier se réveille doucement sous cette petite fenêtre près de nous.
[0.3 seconds pause]
(soft amazed exhale, 0.4 seconds)
(vulnerable honesty, a tiny searching pause) [6.8 seconds duration] Je cherche quelque chose d'intelligent à dire... mais en vérité, je me sens tout petit et un peu dépassé par ce moment.
[0.3 seconds pause]
(soft affection and reassurance) [6.8 seconds duration] Tu n'as pas besoin de parler non plus; pose ta main ici et regardons simplement la lumière changer pendant quelques instants.
[0.3 seconds pause]
(quiet inspiration, intimate rather than grandiose) [6.7 seconds duration] En rentrant, j'oublierai sans doute la forme exacte des nuages, mais je crois que je n'oublierai jamais d'avoir été ici avec toi.
[('emo_Awe', 2.5), ('emo_Affection', 2.0)]
Sunrise above the clouds · Spanish
Breathless awe, a little vulnerability and quiet, shared tenderness.
Mira el borde de las nubes, justo ahí; sabía que el amanecer sería luminoso, pero no me esperaba ese color. Después de tantos meses de preparativos, parece que el mundo entero está despertando despacio debajo de esta pequeña ventana junto a nosotros. Llevo un rato intentando pensar en algo inteligente que decir... pero sinceramente, me siento muy pequeño y un poco sobrepasado. Tú tampoco tienes que decir nada; pon la mano aquí y vamos a mirar juntos cómo cambia la luz durante un momento. Cuando volvamos a casa olvidaré la forma exacta de las nubes, pero creo que nunca olvidaré haber compartido este instante contigo.
No reference — the performance prompt only
1.2 Alpha · SFT
Actual 34.80s · seed 20261627
CE 5.84 · genuineness 4.14 · WER 1.0% Fixed-target emotion MSE 3.209 (normalized 0.2005)
1.3 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.19 · genuineness 4.18 · WER 0.0% Fixed-target emotion MSE 3.840 (normalized 0.2400)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261627
CE 6.20 · genuineness 4.36 · WER 1.0% Fixed-target emotion MSE 3.086 (normalized 0.1929)
Same performance prompt + separate speaker reference
1.2 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.22 · genuineness 4.08 · WER 0.0% Fixed-target emotion MSE 2.454 (normalized 0.1534)
1.3 Alpha · SFT
Actual 34.80s · seed 20261627
CE 6.26 · genuineness 3.87 · WER 0.0% Fixed-target emotion MSE 2.976 (normalized 0.1860)
1.4 Alpha · rank-128 RL
Actual 34.80s · seed 20261627
CE 6.16 · genuineness 4.04 · WER 0.0% Fixed-target emotion MSE 3.187 (normalized 0.1992)
Exact performance prompt, script and fixed emotion targets
GENERAL: Clean, close-miked studio speech with one speaker. Natural, organic, spontaneous acting, not an advertisement or a formal announcement. Let the emotion evolve rather than staying fixed; keep small human hesitations and breaths. Breathless awe, a little vulnerability and quiet, shared tenderness.
SCRIPT:
(breathless wonder, almost a whisper) [6.4 seconds duration] Mira el borde de las nubes, justo ahí; sabía que el amanecer sería luminoso, pero no me esperaba ese color.
[0.3 seconds pause]
(expansive awe, gradually opening the voice) [6.5 seconds duration] Después de tantos meses de preparativos, parece que el mundo entero está despertando despacio debajo de esta pequeña ventana junto a nosotros.
[0.3 seconds pause]
(soft amazed exhale, 0.4 seconds)
(vulnerable honesty, a tiny searching pause) [6.8 seconds duration] Llevo un rato intentando pensar en algo inteligente que decir... pero sinceramente, me siento muy pequeño y un poco sobrepasado.
[0.3 seconds pause]
(soft affection and reassurance) [6.8 seconds duration] Tú tampoco tienes que decir nada; pon la mano aquí y vamos a mirar juntos cómo cambia la luz durante un momento.
[0.3 seconds pause]
(quiet inspiration, intimate rather than grandiose) [6.7 seconds duration] Cuando volvamos a casa olvidaré la forma exacta de las nubes, pero creo que nunca olvidaré haber compartido este instante contigo.