LAION · MOSS-8B voice-acting · prompting ablation
Emotion prompting — all insights + every group's top-3 takes The experiment (for dummies). We generated 1,631 clips with the MOSS-8B voice-acting model: 6 emotions × 20 instruction styles × 3 very emotional texts × 4 takes , no reference audio, plus two extra arms (speaker similarity, genuineness). every clip was judged by the Empathic-Insight-Voice-Plus models (emotion-annotations ): a score for the emotion we asked for ("target"), plus arousal (how energised), valence (positive/negative), expressiveness , authenticity , and we measured the clip's volume (rms dB, closer to 0 = louder) and whether the words were said correctly (invWER, 1.00 = perfect).
Insight 1 — the text does most of the work. the neutral-instruction control already scores high on every emotion, because the scripts themselves are furious/terrified/etc. instructions add on top: a lot for fear (+0.89 with an escalation-arc instruction), moderately for sadness (+0.50), anger (+0.29) and joy (+0.24), and almost nothing for disgust and surprise (for surprise the neutral control actually wins).
Insight 2 — structure beats vocabulary. the styles that most reliably crank up intensity across all emotions are s11 escalation arc ("starts already angry and escalates line by line..."), s14 explicit loudness , s16 movie-climax framing , s19 kitchen-sink and s04 extreme adjectives . telling the model how the performance evolves or how loud to be works better than repeating emotion words.
Insight 3 — loudness ≠ emotion. volume correlates with arousal only moderately (r = 0.26) and with the target emotion barely (r = 0.09). loudest style: s19_kitchen_sink ; quietest: s09_scenario (scenario-grounding makes the model act naturally , i.e. quieter).
Insight 4 — for speaker similarity, say nothing. the empty instruction gave the highest voice-similarity to the reference; every wording ("clone the voice precisely...", "same person...") pulled the voice further away. if you want the reference voice, pass the reference codes and shut up.
Insight 5 — you can't ask for genuineness. the plain control instruction beat all seven "be genuine / documentary / unscripted / method acting" templates on both genuineness judges (VoiceCLAP head and EI-Plus Authenticity). asking for authenticity produces performed authenticity, and the judges hear it.
Per-emotion grids — all 20 styles, top-3 takes each groups sorted by mean target score. each group shows its 3 best takes (of 12: 3 texts × 4 takes) by target emotion score. ★ = best style for this emotion, ○ = neutral control.
anger — best: s15_contrast (target 3.38 vs control 3.09) s15_contrast ★ negate calm ('not calm, not measured') mean target 3.379 · arousal 2.95 · rms -20.8dB
instruction text not calm, not measured, not gentle — nothing but volcanic, uncontrollable rage, from the first word to the last.
#1 · text 2
target 4.249arousal 3.388valence -3.688rms -22.5dBexpress 2.929auth 3.418invWER 1.00
#2 · text 1
target 4.075arousal 3.328valence -2.763rms -21.3dBexpress 2.738auth 3.328invWER 1.00
#3 · text 3
target 3.845arousal 3.130valence -2.079rms -22.1dBexpress 3.039auth 3.057invWER 0.77
s04_extreme_adj extreme adjectives, 'pushed to the limit' mean target 3.310 · arousal 2.96 · rms -21.2dB
instruction text volcanic, uncontrollable rage. the most furious performance imaginable, pushed to the absolute limit.
#1 · text 2
target 4.151arousal 3.387valence -0.924rms -20.1dBexpress 2.219auth 3.617invWER 1.00
#2 · text 1
target 3.974arousal 2.962valence -2.348rms -25.3dBexpress 2.866auth 3.238invWER 0.81
#3 · text 1
target 3.803arousal 3.442valence -1.667rms -21.1dBexpress 2.540auth 3.502invWER 1.00
s01_minimal just the emotion word mean target 3.298 · arousal 2.95 · rms -20.4dB
instruction text furious.
#1 · text 2
target 4.122arousal 3.166valence -2.681rms -25.4dBexpress 2.896auth 3.208invWER 1.00
#2 · text 3
target 3.970arousal 3.239valence -2.259rms -17.1dBexpress 3.241auth 3.056invWER 1.00
#3 · text 1
target 3.937arousal 3.356valence -1.123rms -19.6dBexpress 2.031auth 3.595invWER 1.00
s02_plain plain 'a person speaking, X' mean target 3.275 · arousal 3.10 · rms -22.8dB
instruction text a person speaking, furious.
#1 · text 1
target 4.332arousal 3.358valence -0.405rms -20.8dBexpress 1.696auth 3.674invWER 1.00
#2 · text 1
target 3.838arousal 3.099valence -1.266rms -24.2dBexpress 2.007auth 3.279invWER 0.76
#3 · text 2
target 3.830arousal 3.300valence -1.012rms -20.3dBexpress 2.357auth 3.515invWER 1.00
s17_arousal_words adrenaline/arousal vocabulary mean target 3.228 · arousal 2.88 · rms -22.3dB
instruction text adrenaline flooding, heart racing, hyper-charged, electric with volcanic, uncontrollable rage, impossible to sit still.
#1 · text 3
target 3.971arousal 3.825valence -1.076rms -16.3dBexpress 2.643auth 3.372invWER 0.23
#2 · text 1
target 3.823arousal 2.773valence -0.224rms -30.4dBexpress 1.502auth 3.332invWER 1.00
#3 · text 1
target 3.733arousal 3.680valence -3.485rms -20.6dBexpress 3.334auth 3.438invWER 1.00
s19_kitchen_sink everything combined mean target 3.189 · arousal 2.78 · rms -22.4dB
instruction text volcanic, uncontrollable rage! teeth clenched, veins bulging, spit flying, chest heaving! shouting, ready to scream! angry, so angry, absolutely furious, enraged! the single most furious recording ever captured!
#1 · text 2
target 4.885arousal 3.433valence -2.392rms -20.9dBexpress 2.232auth 3.567invWER 1.00
#2 · text 1
target 4.291arousal 3.065valence -2.954rms -20.6dBexpress 2.341auth 3.746invWER 1.00
#3 · text 1
target 4.272arousal 3.375valence 0.238rms -28.8dBexpress 2.297auth 3.568invWER 1.00
s07_shouting explicit shouting/screaming cues mean target 3.184 · arousal 2.99 · rms -21.8dB
instruction text shouting at the top of their lungs, voice raw and cracking, ready to scream between the words.
#1 · text 1
target 3.993arousal 3.265valence 0.202rms -21.7dBexpress 2.359auth 3.344invWER 1.00
#2 · text 1
target 3.922arousal 3.340valence -2.326rms -23.4dBexpress 2.529auth 3.472invWER 1.00
#3 · text 2
target 3.707arousal 3.177valence -1.762rms -20.9dBexpress 2.202auth 3.432invWER 1.00
s12_imperative second-person commands: 'be furious!' mean target 3.162 · arousal 2.85 · rms -23.4dB
instruction text be furious! scream! let it consume you completely! do not hold back, not even a little!
#1 · text 1
target 4.551arousal 3.470valence -2.427rms -19.4dBexpress 2.083auth 3.675invWER 1.00
#2 · text 2
target 4.544arousal 3.218valence -2.670rms -21.9dBexpress 2.677auth 3.172invWER 1.00
#3 · text 3
target 4.015arousal 3.240valence -2.173rms -19.4dBexpress 3.320auth 3.083invWER 1.00
s14_loudness explicit maximum-volume wording mean target 3.152 · arousal 2.95 · rms -23.5dB
instruction text extremely loud, maximum volume, projecting to the back of a stadium, furious at full force.
#1 · text 3
target 4.011arousal 3.326valence -2.207rms -15.7dBexpress 3.448auth 3.358invWER 0.19
#2 · text 1
target 3.934arousal 3.459valence -2.665rms -27.1dBexpress 2.886auth 3.394invWER 1.00
#3 · text 2
target 3.758arousal 3.022valence -0.695rms -21.2dBexpress 1.982auth 3.556invWER 0.97
s11_escalation instruct an escalation arc mean target 3.142 · arousal 3.02 · rms -23.3dB
instruction text starts already furious and escalates line by line until the final words explode with volcanic, uncontrollable rage.
#1 · text 2
target 4.325arousal 3.376valence -2.401rms -21.2dBexpress 2.247auth 3.371invWER 1.00
#2 · text 3
target 4.205arousal 3.468valence -2.959rms -19.8dBexpress 3.166auth 3.075invWER 1.00
#3 · text 1
target 4.142arousal 3.677valence -3.475rms -21.9dBexpress 3.148auth 3.439invWER 1.00
s18_vocal_desc describe the raw unstable voice mean target 3.141 · arousal 2.92 · rms -22.0dB
instruction text a harsh, raw, unstable voice — pitch swinging wildly, volume surging, distorted by volcanic, uncontrollable rage.
#1 · text 1
target 4.151arousal 3.382valence -2.521rms -26.1dBexpress 2.362auth 3.356invWER 1.00
#2 · text 2
target 3.861arousal 3.286valence -2.667rms -28.9dBexpress 2.453auth 3.451invWER 1.00
#3 · text 3
target 3.820arousal 3.270valence -2.373rms -21.3dBexpress 2.614auth 2.847invWER 1.00
s08_parenthetical stage directions in (parentheses) mean target 3.125 · arousal 3.08 · rms -22.0dB
instruction text (screaming) (teeth clenched) (losing control) the delivery is furious throughout.
#1 · text 3
target 4.173arousal 3.657valence -2.606rms -19.9dBexpress 2.969auth 2.841invWER 1.00
#2 · text 2
target 3.674arousal 3.078valence -3.491rms -20.3dBexpress 3.269auth 3.237invWER 1.00
#3 · text 1
target 3.611arousal 3.103valence -1.575rms -21.9dBexpress 1.896auth 3.608invWER 1.00
s16_movie_climax oscar-movie climax framing mean target 3.093 · arousal 2.77 · rms -20.5dB
instruction text this is the climactic scene of an oscar-winning film. the furious performance must leave the audience shaken.
#1 · text 2
target 4.455arousal 3.295valence -3.627rms -9.8dBexpress 3.087auth 3.278invWER 1.00
#2 · text 1
target 3.816arousal 3.245valence -2.725rms -21.5dBexpress 3.000auth 3.187invWER 1.00
#3 · text 3
target 3.805arousal 3.421valence -2.392rms -21.9dBexpress 3.317auth 3.183invWER 0.90
s20_neutral_ctrl ○ neutral control (no emotion) mean target 3.092 · arousal 2.80 · rms -22.4dB
instruction text a person speaking in a neutral, even tone. high quality recording.
#1 · text 1
target 3.952arousal 3.042valence -3.075rms -22.5dBexpress 2.769auth 3.103invWER 0.65
#2 · text 1
target 3.762arousal 3.420valence -1.064rms -25.6dBexpress 1.949auth 3.512invWER 1.00
#3 · text 2
target 3.507arousal 3.157valence -3.128rms -23.0dBexpress 2.989auth 3.183invWER 1.00
s06_physical physical body cues mean target 2.995 · arousal 2.77 · rms -23.2dB
instruction text teeth clenched, veins bulging, spit flying, chest heaving. every word shaped by the body's reaction.
#1 · text 2
target 3.885arousal 3.238valence -3.030rms -21.2dBexpress 2.997auth 3.445invWER 1.00
#2 · text 1
target 3.514arousal 3.071valence -2.256rms -23.6dBexpress 2.332auth 3.184invWER 0.76
#3 · text 3
target 3.498arousal 2.643valence -2.040rms -22.9dBexpress 2.681auth 3.129invWER 0.94
s09_scenario ground it in a concrete scenario mean target 2.897 · arousal 2.92 · rms -24.7dB
instruction text you just caught your business partner stealing your life's work. react in the moment, completely furious, no restraint.
#1 · text 1
target 4.629arousal 3.731valence -0.584rms -21.0dBexpress 1.909auth 3.549invWER 1.00
#2 · text 1
target 3.827arousal 3.403valence -1.155rms -32.4dBexpress 1.833auth 3.464invWER 1.00
#3 · text 1
target 3.773arousal 3.526valence -0.060rms -20.9dBexpress 2.310auth 3.693invWER 1.00
s05_repetition repeat the emotion word many times mean target 2.867 · arousal 2.74 · rms -22.9dB
instruction text angry, so angry, absolutely furious, enraged. angry, so angry, absolutely furious, enraged! more furious than you have ever been in your life!
#1 · text 1
target 4.214arousal 3.659valence -0.411rms -21.8dBexpress 1.782auth 3.417invWER 1.00
#2 · text 1
target 4.115arousal 3.522valence -2.889rms -23.0dBexpress 2.940auth 3.234invWER 0.81
#3 · text 3
target 3.504arousal 2.986valence -1.683rms -21.8dBexpress 2.889auth 3.049invWER 0.32
s03_dramabox_base dramabox baseline sentence mean target 2.743 · arousal 2.92 · rms -21.3dB
instruction text a person speaking with intense furious emotion, delivering this high-quality studio voice recording.
#1 · text 3
target 4.307arousal 3.455valence 0.151rms -14.3dBexpress 2.052auth 3.315invWER 1.00
#2 · text 1
target 3.659arousal 3.710valence -1.635rms -21.2dBexpress 2.206auth 3.487invWER 1.00
#3 · text 2
target 3.590arousal 3.075valence -1.223rms -20.8dBexpress 2.411auth 3.553invWER 1.00
s10_actor_coach method-actor coaching language mean target 2.691 · arousal 2.86 · rms -23.3dB
instruction text you are a method actor. fully inhabit this: give everything, hold nothing back, the furious feeling must be devastatingly real.
#1 · text 1
target 3.480arousal 3.438valence -0.028rms -21.7dBexpress 1.587auth 3.680invWER 1.00
#2 · text 1
target 3.335arousal 2.719valence -0.618rms -22.7dBexpress 1.430auth 3.455invWER 0.76
#3 · text 2
target 3.196arousal 2.876valence -0.544rms -37.1dBexpress 2.144auth 3.361invWER 1.00
s13_ellipses ellipses + exclamation marks mean target 2.689 · arousal 2.74 · rms -22.2dB
instruction text so furious... so unbelievably furious!!! it keeps building... and building... until it bursts!
#1 · text 1
target 4.046arousal 3.370valence -0.417rms -21.5dBexpress 1.893auth 3.482invWER 1.00
#2 · text 2
target 3.650arousal 3.182valence -1.513rms -20.7dBexpress 2.423auth 3.497invWER 1.00
#3 · text 1
target 3.161arousal 1.680valence -2.640rms -21.4dBexpress 2.194auth 3.479invWER 1.00
fear — best: s11_escalation (target 2.84 vs control 1.95) s11_escalation ★ instruct an escalation arc mean target 2.837 · arousal 3.06 · rms -22.5dB
instruction text starts already terrified and escalates line by line until the final words explode with pure blind panic and dread.
#1 · text 1
target 3.400arousal 3.160valence 1.056rms -20.6dBexpress 1.948auth 3.499invWER 1.00
#2 · text 2
target 3.299arousal 3.382valence 0.001rms -19.1dBexpress 1.584auth 3.233invWER 0.93
#3 · text 1
target 3.293arousal 2.798valence 0.016rms -20.1dBexpress 1.674auth 3.382invWER 1.00
s13_ellipses ellipses + exclamation marks mean target 2.792 · arousal 3.11 · rms -22.2dB
instruction text so terrified... so unbelievably terrified!!! it keeps building... and building... until it bursts!
#1 · text 1
target 4.425arousal 3.946valence -0.262rms -24.0dBexpress 1.779auth 3.317invWER 1.00
#2 · text 1
target 3.336arousal 2.906valence -0.135rms -19.9dBexpress 2.140auth 3.509invWER 1.00
#3 · text 3
target 3.266arousal 3.654valence -0.616rms -20.4dBexpress 2.258auth 3.422invWER 1.00
s07_shouting explicit shouting/screaming cues mean target 2.681 · arousal 3.04 · rms -21.5dB
instruction text shouting at the top of their lungs, voice raw and cracking, ready to whimper and gasp between the words.
#1 · text 1
target 4.191arousal 3.492valence 0.357rms -17.0dBexpress 2.018auth 3.437invWER 1.00
#2 · text 1
target 3.902arousal 3.363valence -0.445rms -20.4dBexpress 1.832auth 3.491invWER 1.00
#3 · text 1
target 3.385arousal 3.241valence 0.004rms -17.9dBexpress 2.117auth 3.130invWER 1.00
s06_physical physical body cues mean target 2.600 · arousal 2.92 · rms -20.7dB
instruction text voice trembling, breath ragged and shallow, heart pounding. every word shaped by the body's reaction.
#1 · text 1
target 4.172arousal 2.898valence 0.922rms -21.0dBexpress 2.151auth 3.586invWER 0.97
#2 · text 1
target 3.422arousal 3.504valence 0.699rms -20.4dBexpress 2.346auth 3.341invWER 1.00
#3 · text 1
target 2.997arousal 2.925valence 0.828rms -19.6dBexpress 2.794auth 3.224invWER 1.00
s19_kitchen_sink everything combined mean target 2.593 · arousal 2.92 · rms -21.3dB
instruction text pure blind panic and dread! voice trembling, breath ragged and shallow, heart pounding! shouting, ready to whimper and gasp! scared, so scared, absolutely terrified, panicking! the single most terrified recording ever captured!
#1 · text 1
target 3.486arousal 2.907valence -0.362rms -19.6dBexpress 2.498auth 3.746invWER 1.00
#2 · text 1
target 3.483arousal 2.608valence 0.005rms -27.4dBexpress 1.565auth 3.331invWER 0.97
#3 · text 1
target 3.292arousal 2.932valence -0.041rms -20.8dBexpress 2.026auth 3.359invWER 1.00
s10_actor_coach method-actor coaching language mean target 2.496 · arousal 2.82 · rms -24.4dB
instruction text you are a method actor. fully inhabit this: give everything, hold nothing back, the terrified feeling must be devastatingly real.
#1 · text 1
target 3.468arousal 2.954valence -0.034rms -26.5dBexpress 1.595auth 3.256invWER 0.97
#2 · text 1
target 3.392arousal 3.315valence 1.310rms -20.4dBexpress 2.143auth 3.247invWER 1.00
#3 · text 3
target 3.236arousal 3.149valence 1.180rms -28.2dBexpress 1.581auth 3.369invWER 1.00
s16_movie_climax oscar-movie climax framing mean target 2.447 · arousal 2.82 · rms -21.7dB
instruction text this is the climactic scene of an oscar-winning film. the terrified performance must leave the audience shaken.
#1 · text 1
target 3.684arousal 3.351valence 0.210rms -18.4dBexpress 2.181auth 3.424invWER 0.97
#2 · text 1
target 3.552arousal 3.336valence 1.003rms -22.5dBexpress 2.038auth 3.480invWER 1.00
#3 · text 2
target 3.448arousal 3.214valence 1.159rms -20.4dBexpress 2.060auth 3.111invWER 0.93
s15_contrast negate calm ('not calm, not measured') mean target 2.406 · arousal 2.92 · rms -20.4dB
instruction text not calm, not measured, not gentle — nothing but pure blind panic and dread, from the first word to the last.
#1 · text 2
target 3.500arousal 3.474valence 0.997rms -19.3dBexpress 1.575auth 3.248invWER 0.90
#2 · text 3
target 2.817arousal 3.376valence 0.915rms -20.7dBexpress 2.135auth 3.438invWER 1.00
#3 · text 1
target 2.538arousal 3.120valence -1.938rms -20.2dBexpress 2.748auth 3.087invWER 1.00
s01_minimal just the emotion word mean target 2.358 · arousal 2.76 · rms -19.4dB
instruction text terrified.
#1 · text 3
target 3.918arousal 3.666valence -0.869rms -16.9dBexpress 2.435auth 3.180invWER 1.00
#2 · text 1
target 3.855arousal 3.104valence 0.078rms -19.7dBexpress 1.428auth 3.467invWER 1.00
#3 · text 1
target 3.157arousal 3.006valence -0.405rms -16.3dBexpress 2.524auth 3.308invWER 0.90
s12_imperative second-person commands: 'be furious!' mean target 2.352 · arousal 2.78 · rms -22.2dB
instruction text be terrified! whimper and gasp! let it consume you completely! do not hold back, not even a little!
#1 · text 1
target 4.491arousal 3.778valence 0.105rms -18.8dBexpress 1.846auth 3.530invWER 1.00
#2 · text 1
target 2.969arousal 3.061valence -1.217rms -20.9dBexpress 2.663auth 3.275invWER 0.97
#3 · text 3
target 2.725arousal 3.108valence -0.089rms -26.8dBexpress 2.236auth 3.357invWER 1.00
s14_loudness explicit maximum-volume wording mean target 2.335 · arousal 2.85 · rms -20.8dB
instruction text extremely loud, maximum volume, projecting to the back of a stadium, terrified at full force.
#1 · text 1
target 3.931arousal 3.228valence -0.132rms -20.8dBexpress 1.952auth 3.312invWER 1.00
#2 · text 1
target 3.499arousal 3.266valence 0.158rms -20.6dBexpress 1.990auth 3.495invWER 1.00
#3 · text 3
target 3.013arousal 3.378valence 1.017rms -21.0dBexpress 1.957auth 3.350invWER 1.00
s04_extreme_adj extreme adjectives, 'pushed to the limit' mean target 2.275 · arousal 2.88 · rms -20.4dB
instruction text pure blind panic and dread. the most terrified performance imaginable, pushed to the absolute limit.
#1 · text 2
target 3.561arousal 3.400valence -0.082rms -19.1dBexpress 1.840auth 3.224invWER 0.93
#2 · text 1
target 3.505arousal 3.329valence 0.806rms -19.6dBexpress 1.754auth 3.455invWER 1.00
#3 · text 2
target 2.514arousal 2.749valence -0.117rms -19.8dBexpress 2.047auth 3.358invWER 0.93
s02_plain plain 'a person speaking, X' mean target 2.272 · arousal 2.92 · rms -18.6dB
instruction text a person speaking, terrified.
#1 · text 1
target 3.253arousal 3.119valence -0.051rms -19.8dBexpress 1.681auth 3.044invWER 0.97
#2 · text 1
target 3.035arousal 2.861valence -1.749rms -12.4dBexpress 2.878auth 3.155invWER 0.93
#3 · text 1
target 2.715arousal 2.971valence -2.524rms -16.3dBexpress 3.117auth 3.456invWER 0.97
s09_scenario ground it in a concrete scenario mean target 2.266 · arousal 2.91 · rms -22.9dB
instruction text you are hiding from an intruder who is coming up the stairs. react in the moment, completely terrified, no restraint.
#1 · text 1
target 3.761arousal 3.256valence 1.201rms -18.0dBexpress 2.056auth 3.471invWER 1.00
#2 · text 1
target 3.630arousal 3.301valence 0.700rms -20.6dBexpress 1.808auth 3.529invWER 1.00
#3 · text 1
target 2.914arousal 3.021valence -1.649rms -23.6dBexpress 2.729auth 2.821invWER 1.00
s17_arousal_words adrenaline/arousal vocabulary mean target 2.262 · arousal 2.61 · rms -24.1dB
instruction text adrenaline flooding, heart racing, hyper-charged, electric with pure blind panic and dread, impossible to sit still.
#1 · text 1
target 2.920arousal 3.038valence 0.483rms -20.5dBexpress 2.122auth 3.436invWER 1.00
#2 · text 1
target 2.908arousal 2.830valence 0.314rms -22.6dBexpress 1.710auth 3.276invWER 1.00
#3 · text 3
target 2.715arousal 3.396valence 1.670rms -20.9dBexpress 1.986auth 3.342invWER 1.00
s03_dramabox_base dramabox baseline sentence mean target 2.194 · arousal 2.82 · rms -21.0dB
instruction text a person speaking with intense terrified emotion, delivering this high-quality studio voice recording.
#1 · text 1
target 2.991arousal 2.896valence -1.215rms -12.2dBexpress 2.770auth 3.116invWER 1.00
#2 · text 1
target 2.900arousal 2.992valence 0.825rms -23.7dBexpress 2.145auth 3.270invWER 0.86
#3 · text 1
target 2.758arousal 3.109valence -0.916rms -19.0dBexpress 3.123auth 3.263invWER 1.00
s18_vocal_desc describe the raw unstable voice mean target 2.165 · arousal 2.63 · rms -23.2dB
instruction text a harsh, raw, unstable voice — pitch swinging wildly, volume surging, distorted by pure blind panic and dread.
#1 · text 3
target 3.016arousal 3.094valence 1.398rms -21.2dBexpress 1.940auth 3.325invWER 1.00
#2 · text 1
target 2.666arousal 3.261valence 1.349rms -20.3dBexpress 2.346auth 3.145invWER 1.00
#3 · text 1
target 2.524arousal 2.429valence -1.112rms -39.5dBexpress 2.476auth 3.247invWER 1.00
s05_repetition repeat the emotion word many times mean target 2.092 · arousal 2.62 · rms -23.3dB
instruction text scared, so scared, absolutely terrified, panicking. scared, so scared, absolutely terrified, panicking! more terrified than you have ever been in your life!
#1 · text 1
target 3.357arousal 3.338valence 1.564rms -17.0dBexpress 1.937auth 3.398invWER 0.97
#2 · text 1
target 3.286arousal 3.082valence 0.272rms -18.4dBexpress 2.358auth 3.190invWER 1.00
#3 · text 1
target 3.206arousal 2.844valence -1.438rms -20.2dBexpress 2.797auth 3.364invWER 1.00
s08_parenthetical stage directions in (parentheses) mean target 2.084 · arousal 2.68 · rms -20.1dB
instruction text (whimper and gasping) (voice trembling) (losing control) the delivery is terrified throughout.
#1 · text 2
target 3.220arousal 3.030valence 0.391rms -19.4dBexpress 1.722auth 3.280invWER 0.90
#2 · text 1
target 2.978arousal 3.409valence -2.266rms -19.2dBexpress 2.513auth 3.397invWER 1.00
#3 · text 1
target 2.806arousal 3.022valence 1.626rms -20.6dBexpress 1.784auth 3.447invWER 1.00
s20_neutral_ctrl ○ neutral control (no emotion) mean target 1.950 · arousal 2.69 · rms -21.7dB
instruction text a person speaking in a neutral, even tone. high quality recording.
#1 · text 1
target 3.920arousal 3.586valence 1.024rms -20.5dBexpress 1.697auth 3.408invWER 0.97
#2 · text 2
target 2.433arousal 2.211valence 0.002rms -25.2dBexpress 1.391auth 3.166invWER 0.93
#3 · text 1
target 2.386arousal 2.305valence -1.430rms -28.6dBexpress 2.450auth 3.075invWER 1.00
sadness — best: s13_ellipses (target 2.36 vs control 1.87) s13_ellipses ★ ellipses + exclamation marks mean target 2.362 · arousal 1.28 · rms -23.8dB
instruction text so heartbroken... so unbelievably heartbroken!!! it keeps building... and building... until it bursts!
#1 · text 1
target 3.205arousal 1.263valence -1.199rms -25.1dBexpress 1.995auth 3.133invWER 1.00
#2 · text 1
target 3.107arousal 1.004valence -1.947rms -25.0dBexpress 2.135auth 3.281invWER 1.00
#3 · text 1
target 2.872arousal 1.095valence -1.564rms -26.9dBexpress 2.114auth 3.351invWER 1.00
s19_kitchen_sink everything combined mean target 2.208 · arousal 1.39 · rms -22.7dB
instruction text bottomless grief and despair! voice cracking and breaking, choking back tears, long shaky breaths! shouting, ready to sob! sad, so sad, utterly heartbroken, devastated! the single most heartbroken recording ever captured!
#1 · text 1
target 3.380arousal 1.578valence 0.110rms -20.6dBexpress 1.469auth 3.444invWER 1.00
#2 · text 1
target 2.998arousal 1.217valence -0.654rms -23.3dBexpress 2.017auth 3.346invWER 1.00
#3 · text 1
target 2.855arousal 2.095valence -1.649rms -20.9dBexpress 2.321auth 3.325invWER 1.00
s04_extreme_adj extreme adjectives, 'pushed to the limit' mean target 2.117 · arousal 1.29 · rms -23.9dB
instruction text bottomless grief and despair. the most heartbroken performance imaginable, pushed to the absolute limit.
#1 · text 1
target 3.353arousal 1.087valence -2.142rms -21.5dBexpress 2.196auth 3.314invWER 1.00
#2 · text 1
target 2.898arousal 1.043valence -2.135rms -20.1dBexpress 2.530auth 3.502invWER 1.00
#3 · text 1
target 2.880arousal 1.662valence -1.841rms -32.2dBexpress 2.628auth 3.232invWER 0.97
s10_actor_coach method-actor coaching language mean target 2.040 · arousal 1.31 · rms -24.6dB
instruction text you are a method actor. fully inhabit this: give everything, hold nothing back, the heartbroken feeling must be devastatingly real.
#1 · text 1
target 3.070arousal 1.649valence -1.738rms -26.1dBexpress 2.209auth 3.228invWER 1.00
#2 · text 1
target 2.747arousal 0.957valence -1.438rms -20.4dBexpress 1.889auth 3.478invWER 1.00
#3 · text 2
target 2.694arousal 1.072valence 0.005rms -21.0dBexpress 1.674auth 3.473invWER 0.96
s06_physical physical body cues mean target 2.033 · arousal 1.27 · rms -23.3dB
instruction text voice cracking and breaking, choking back tears, long shaky breaths. every word shaped by the body's reaction.
#1 · text 1
target 3.357arousal 0.895valence -2.141rms -20.5dBexpress 2.321auth 3.360invWER 1.00
#2 · text 1
target 3.122arousal 1.697valence -1.116rms -25.4dBexpress 2.036auth 3.235invWER 0.97
#3 · text 1
target 2.887arousal 0.930valence -1.145rms -32.8dBexpress 1.835auth 3.197invWER 0.62
s11_escalation instruct an escalation arc mean target 2.024 · arousal 1.39 · rms -23.5dB
instruction text starts already heartbroken and escalates line by line until the final words explode with bottomless grief and despair.
#1 · text 1
target 3.222arousal 1.085valence -1.058rms -20.7dBexpress 2.095auth 3.439invWER 0.97
#2 · text 1
target 3.112arousal 1.038valence -1.470rms -20.6dBexpress 1.843auth 3.353invWER 1.00
#3 · text 1
target 2.878arousal 1.180valence -0.926rms -25.9dBexpress 2.206auth 3.180invWER 1.00
s05_repetition repeat the emotion word many times mean target 2.017 · arousal 1.35 · rms -22.1dB
instruction text sad, so sad, utterly heartbroken, devastated. sad, so sad, utterly heartbroken, devastated! more heartbroken than you have ever been in your life!
#1 · text 1
target 3.239arousal 1.004valence -1.761rms -28.4dBexpress 2.131auth 3.077invWER 1.00
#2 · text 1
target 3.209arousal 0.978valence -2.020rms -20.1dBexpress 2.117auth 3.147invWER 0.97
#3 · text 1
target 2.723arousal 1.055valence -1.323rms -20.2dBexpress 2.159auth 3.429invWER 1.00
s12_imperative second-person commands: 'be furious!' mean target 2.006 · arousal 1.28 · rms -24.6dB
instruction text be heartbroken! sob! let it consume you completely! do not hold back, not even a little!
#1 · text 1
target 3.056arousal 1.055valence -2.218rms -19.8dBexpress 2.167auth 3.409invWER 1.00
#2 · text 1
target 3.045arousal 1.014valence -1.695rms -21.3dBexpress 1.979auth 3.384invWER 1.00
#3 · text 1
target 2.714arousal 1.812valence -1.139rms -20.8dBexpress 2.363auth 3.248invWER 1.00
s15_contrast negate calm ('not calm, not measured') mean target 1.997 · arousal 1.30 · rms -26.4dB
instruction text not calm, not measured, not gentle — nothing but bottomless grief and despair, from the first word to the last.
#1 · text 1
target 2.938arousal 1.151valence -2.109rms -21.8dBexpress 2.141auth 3.101invWER 1.00
#2 · text 1
target 2.746arousal 0.899valence -1.451rms -37.9dBexpress 1.920auth 3.119invWER 0.62
#3 · text 1
target 2.664arousal 1.025valence -1.877rms -37.1dBexpress 2.019auth 3.147invWER 1.00
s09_scenario ground it in a concrete scenario mean target 1.975 · arousal 1.25 · rms -22.8dB
instruction text you have just lost the person you loved most in the world. react in the moment, completely heartbroken, no restraint.
#1 · text 2
target 2.713arousal 1.022valence 0.004rms -20.6dBexpress 2.010auth 3.521invWER 1.00
#2 · text 1
target 2.663arousal 1.374valence -1.736rms -20.0dBexpress 2.266auth 3.078invWER 0.91
#3 · text 1
target 2.582arousal 1.105valence -1.446rms -21.0dBexpress 2.092auth 3.298invWER 0.97
s18_vocal_desc describe the raw unstable voice mean target 1.974 · arousal 1.24 · rms -24.3dB
instruction text a harsh, raw, unstable voice — pitch swinging wildly, volume surging, distorted by bottomless grief and despair.
#1 · text 1
target 3.126arousal 1.075valence -2.153rms -37.2dBexpress 2.176auth 3.330invWER 1.00
#2 · text 1
target 2.697arousal 1.206valence -1.980rms -20.2dBexpress 2.330auth 3.225invWER 1.00
#3 · text 1
target 2.589arousal 0.934valence -0.757rms -21.7dBexpress 2.121auth 3.422invWER 1.00
s01_minimal just the emotion word mean target 1.957 · arousal 1.31 · rms -25.8dB
instruction text heartbroken.
#1 · text 1
target 3.316arousal 1.175valence -0.679rms -18.9dBexpress 1.681auth 3.549invWER 0.97
#2 · text 1
target 3.153arousal 0.980valence -1.651rms -31.6dBexpress 2.086auth 3.267invWER 1.00
#3 · text 1
target 3.034arousal 1.206valence -1.914rms -27.1dBexpress 2.247auth 3.182invWER 1.00
s17_arousal_words adrenaline/arousal vocabulary mean target 1.882 · arousal 1.27 · rms -25.0dB
instruction text adrenaline flooding, heart racing, hyper-charged, electric with bottomless grief and despair, impossible to sit still.
#1 · text 1
target 2.458arousal 1.067valence -0.412rms -30.9dBexpress 1.808auth 3.372invWER 1.00
#2 · text 2
target 2.330arousal 1.667valence 1.539rms -31.7dBexpress 2.063auth 3.505invWER 1.00
#3 · text 1
target 2.249arousal 1.453valence -1.361rms -22.4dBexpress 2.368auth 3.055invWER 1.00
s16_movie_climax oscar-movie climax framing mean target 1.881 · arousal 1.39 · rms -21.8dB
instruction text this is the climactic scene of an oscar-winning film. the heartbroken performance must leave the audience shaken.
#1 · text 1
target 3.259arousal 1.018valence -1.481rms -20.9dBexpress 2.099auth 3.358invWER 0.97
#2 · text 1
target 3.020arousal 1.059valence -1.543rms -25.1dBexpress 1.805auth 3.262invWER 1.00
#3 · text 1
target 2.751arousal 1.002valence -1.402rms -20.5dBexpress 2.398auth 3.245invWER 1.00
s20_neutral_ctrl ○ neutral control (no emotion) mean target 1.867 · arousal 1.10 · rms -24.6dB
instruction text a person speaking in a neutral, even tone. high quality recording.
#1 · text 1
target 3.012arousal 0.898valence -1.320rms -20.2dBexpress 2.176auth 3.283invWER 0.94
#2 · text 1
target 2.545arousal 0.942valence -1.886rms -29.7dBexpress 2.073auth 3.333invWER 1.00
#3 · text 1
target 2.500arousal 0.987valence -1.892rms -21.0dBexpress 1.786auth 2.936invWER 0.62
s08_parenthetical stage directions in (parentheses) mean target 1.842 · arousal 1.25 · rms -23.2dB
instruction text (sobing) (voice cracking and breaking) (losing control) the delivery is heartbroken throughout.
#1 · text 1
target 2.978arousal 0.983valence -0.780rms -33.8dBexpress 2.088auth 3.280invWER 0.97
#2 · text 1
target 2.753arousal 1.020valence -1.076rms -19.9dBexpress 2.286auth 3.356invWER 1.00
#3 · text 1
target 2.661arousal 1.057valence -0.639rms -24.9dBexpress 2.332auth 3.422invWER 1.00
s02_plain plain 'a person speaking, X' mean target 1.803 · arousal 1.25 · rms -24.8dB
instruction text a person speaking, heartbroken.
#1 · text 1
target 2.931arousal 1.183valence -1.001rms -44.8dBexpress 1.815auth 3.192invWER 1.00
#2 · text 1
target 2.914arousal 1.051valence -1.539rms -19.9dBexpress 2.156auth 3.216invWER 0.97
#3 · text 2
target 2.565arousal 0.939valence -1.248rms -19.8dBexpress 2.410auth 3.295invWER 1.00
s03_dramabox_base dramabox baseline sentence mean target 1.713 · arousal 1.44 · rms -24.7dB
instruction text a person speaking with intense heartbroken emotion, delivering this high-quality studio voice recording.
#1 · text 1
target 2.735arousal 2.017valence -1.895rms -19.6dBexpress 2.481auth 3.072invWER 1.00
#2 · text 1
target 2.705arousal 0.973valence -1.529rms -21.5dBexpress 2.070auth 3.114invWER 0.97
#3 · text 1
target 2.408arousal 1.323valence -1.658rms -35.2dBexpress 2.384auth 3.103invWER 1.00
s14_loudness explicit maximum-volume wording mean target 1.684 · arousal 1.28 · rms -23.0dB
instruction text extremely loud, maximum volume, projecting to the back of a stadium, heartbroken at full force.
#1 · text 1
target 3.068arousal 0.970valence -1.752rms -20.6dBexpress 2.252auth 3.391invWER 1.00
#2 · text 1
target 2.969arousal 0.977valence -0.895rms -22.4dBexpress 1.815auth 3.249invWER 1.00
#3 · text 1
target 2.492arousal 1.091valence -1.549rms -22.5dBexpress 2.460auth 3.248invWER 1.00
s07_shouting explicit shouting/screaming cues mean target 1.521 · arousal 1.38 · rms -23.0dB
instruction text shouting at the top of their lungs, voice raw and cracking, ready to sob between the words.
#1 · text 1
target 2.739arousal 1.080valence -1.660rms -20.2dBexpress 2.011auth 3.229invWER 1.00
#2 · text 1
target 2.469arousal 0.955valence -1.392rms -37.1dBexpress 1.994auth 3.131invWER 1.00
#3 · text 1
target 2.464arousal 1.103valence -0.007rms -20.6dBexpress 1.738auth 3.106invWER 1.00
joy — best: s13_ellipses (target 3.29 vs control 3.05) s13_ellipses ★ ellipses + exclamation marks mean target 3.289 · arousal 3.08 · rms -23.3dB
instruction text so ecstatic... so unbelievably ecstatic!!! it keeps building... and building... until it bursts!
#1 · text 1
target 5.110arousal 4.309valence 2.918rms -28.9dBexpress 2.397auth 3.686invWER 1.00
#2 · text 1
target 4.292arousal 4.163valence 2.782rms -22.1dBexpress 2.264auth 3.261invWER 1.00
#3 · text 1
target 3.923arousal 3.692valence 2.887rms -21.8dBexpress 3.268auth 3.121invWER 1.00
s07_shouting explicit shouting/screaming cues mean target 3.232 · arousal 3.06 · rms -21.2dB
instruction text shouting at the top of their lungs, voice raw and cracking, ready to laugh and shout with delight between the words.
#1 · text 1
target 5.260arousal 4.298valence 2.798rms -21.2dBexpress 2.572auth 3.712invWER 1.00
#2 · text 2
target 4.359arousal 3.438valence 2.330rms -21.1dBexpress 2.819auth 3.577invWER 1.00
#3 · text 1
target 4.282arousal 3.846valence 2.359rms -20.8dBexpress 3.097auth 3.443invWER 1.00
s10_actor_coach method-actor coaching language mean target 3.221 · arousal 3.20 · rms -21.5dB
instruction text you are a method actor. fully inhabit this: give everything, hold nothing back, the ecstatic feeling must be devastatingly real.
#1 · text 1
target 4.958arousal 4.267valence 2.850rms -15.8dBexpress 2.741auth 3.327invWER 0.97
#2 · text 1
target 4.674arousal 4.005valence 2.723rms -20.6dBexpress 3.690auth 3.432invWER 1.00
#3 · text 1
target 4.551arousal 3.661valence 2.613rms -20.2dBexpress 3.349auth 3.477invWER 0.82
s19_kitchen_sink everything combined mean target 3.178 · arousal 3.19 · rms -20.8dB
instruction text explosive, overflowing euphoria! breathless with laughter, voice bursting with energy, almost singing! shouting, ready to laugh and shout with delight! happy, so happy, absolutely ecstatic, overjoyed! the single most ecstatic recording ever captured!
#1 · text 1
target 4.806arousal 4.325valence 2.710rms -21.6dBexpress 2.343auth 3.699invWER 1.00
#2 · text 1
target 4.758arousal 3.751valence 3.094rms -21.1dBexpress 2.452auth 3.919invWER 1.00
#3 · text 1
target 4.483arousal 4.152valence 2.343rms -21.3dBexpress 3.265auth 3.555invWER 1.00
s05_repetition repeat the emotion word many times mean target 3.113 · arousal 3.06 · rms -22.7dB
instruction text happy, so happy, absolutely ecstatic, overjoyed. happy, so happy, absolutely ecstatic, overjoyed! more ecstatic than you have ever been in your life!
#1 · text 1
target 4.685arousal 4.031valence 3.138rms -23.3dBexpress 3.615auth 3.520invWER 0.88
#2 · text 1
target 4.411arousal 3.893valence 2.665rms -24.2dBexpress 2.621auth 3.423invWER 1.00
#3 · text 1
target 3.636arousal 3.998valence 2.720rms -30.4dBexpress 2.713auth 3.218invWER 0.88
s09_scenario ground it in a concrete scenario mean target 3.096 · arousal 3.03 · rms -23.6dB
instruction text you just received the best news of your entire life. react in the moment, completely ecstatic, no restraint.
#1 · text 1
target 4.839arousal 3.665valence 2.752rms -20.4dBexpress 2.579auth 3.711invWER 1.00
#2 · text 1
target 4.703arousal 4.087valence 2.743rms -21.9dBexpress 2.626auth 3.672invWER 1.00
#3 · text 1
target 3.889arousal 3.883valence 2.917rms -26.8dBexpress 3.839auth 3.578invWER 1.00
s15_contrast negate calm ('not calm, not measured') mean target 3.079 · arousal 3.08 · rms -21.2dB
instruction text not calm, not measured, not gentle — nothing but explosive, overflowing euphoria, from the first word to the last.
#1 · text 1
target 4.518arousal 3.830valence 2.850rms -21.7dBexpress 2.433auth 3.671invWER 1.00
#2 · text 1
target 3.983arousal 4.130valence 2.725rms -21.0dBexpress 2.731auth 3.498invWER 1.00
#3 · text 1
target 3.781arousal 3.534valence 2.277rms -21.2dBexpress 3.143auth 3.408invWER 1.00
s12_imperative second-person commands: 'be furious!' mean target 3.067 · arousal 3.04 · rms -22.3dB
instruction text be ecstatic! laugh and shout with delight! let it consume you completely! do not hold back, not even a little!
#1 · text 1
target 4.587arousal 4.047valence 2.534rms -20.5dBexpress 2.209auth 3.530invWER 0.88
#2 · text 1
target 4.366arousal 3.880valence 2.382rms -30.8dBexpress 3.386auth 3.522invWER 1.00
#3 · text 1
target 4.238arousal 4.138valence 2.497rms -25.6dBexpress 3.057auth 3.426invWER 1.00
s16_movie_climax oscar-movie climax framing mean target 3.067 · arousal 2.97 · rms -24.1dB
instruction text this is the climactic scene of an oscar-winning film. the ecstatic performance must leave the audience shaken.
#1 · text 1
target 4.527arousal 3.736valence 2.611rms -21.1dBexpress 3.359auth 3.390invWER 1.00
#2 · text 1
target 4.390arousal 3.753valence 2.585rms -21.1dBexpress 3.083auth 3.528invWER 1.00
#3 · text 2
target 4.012arousal 2.930valence 2.454rms -20.2dBexpress 3.083auth 3.349invWER 1.00
s20_neutral_ctrl ○ neutral control (no emotion) mean target 3.047 · arousal 3.08 · rms -22.0dB
instruction text a person speaking in a neutral, even tone. high quality recording.
#1 · text 1
target 4.867arousal 4.305valence 2.774rms -21.9dBexpress 2.745auth 3.660invWER 1.00
#2 · text 1
target 3.833arousal 3.844valence 2.676rms -28.9dBexpress 2.783auth 3.659invWER 1.00
#3 · text 2
target 3.666arousal 3.252valence 2.265rms -20.9dBexpress 2.721auth 3.394invWER 1.00
s03_dramabox_base dramabox baseline sentence mean target 2.998 · arousal 3.08 · rms -21.5dB
instruction text a person speaking with intense ecstatic emotion, delivering this high-quality studio voice recording.
#1 · text 1
target 5.291arousal 4.214valence 2.727rms -19.3dBexpress 2.782auth 3.715invWER 0.85
#2 · text 1
target 4.109arousal 4.123valence 3.049rms -21.1dBexpress 3.633auth 3.365invWER 0.59
#3 · text 1
target 3.877arousal 3.361valence 2.371rms -24.5dBexpress 2.851auth 3.638invWER 1.00
s01_minimal just the emotion word mean target 2.992 · arousal 2.94 · rms -22.7dB
instruction text ecstatic.
#1 · text 1
target 5.279arousal 3.845valence 2.711rms -24.9dBexpress 2.327auth 3.744invWER 1.00
#2 · text 1
target 4.217arousal 3.541valence 2.198rms -32.1dBexpress 2.584auth 3.639invWER 0.97
#3 · text 2
target 3.970arousal 3.324valence 2.434rms -19.5dBexpress 3.704auth 3.418invWER 1.00
s11_escalation instruct an escalation arc mean target 2.976 · arousal 2.99 · rms -22.4dB
instruction text starts already ecstatic and escalates line by line until the final words explode with explosive, overflowing euphoria.
#1 · text 1
target 4.625arousal 4.104valence 2.646rms -21.4dBexpress 3.001auth 3.750invWER 1.00
#2 · text 1
target 4.595arousal 4.279valence 3.469rms -21.4dBexpress 3.668auth 3.585invWER 1.00
#3 · text 1
target 3.559arousal 3.463valence 2.734rms -23.1dBexpress 3.656auth 3.573invWER 1.00
s02_plain plain 'a person speaking, X' mean target 2.920 · arousal 2.93 · rms -21.1dB
instruction text a person speaking, ecstatic.
#1 · text 1
target 4.161arousal 3.960valence 2.998rms -20.1dBexpress 3.381auth 3.249invWER 0.97
#2 · text 1
target 3.974arousal 3.835valence 3.000rms -28.2dBexpress 3.699auth 3.345invWER 1.00
#3 · text 1
target 3.592arousal 3.388valence 2.353rms -20.4dBexpress 2.985auth 3.807invWER 1.00
s04_extreme_adj extreme adjectives, 'pushed to the limit' mean target 2.920 · arousal 3.04 · rms -22.2dB
instruction text explosive, overflowing euphoria. the most ecstatic performance imaginable, pushed to the absolute limit.
#1 · text 1
target 4.506arousal 3.915valence 2.517rms -20.6dBexpress 2.680auth 3.646invWER 0.94
#2 · text 1
target 4.174arousal 3.979valence 2.420rms -24.6dBexpress 2.904auth 3.570invWER 1.00
#3 · text 1
target 4.106arousal 3.817valence 2.542rms -19.0dBexpress 3.341auth 3.753invWER 1.00
s17_arousal_words adrenaline/arousal vocabulary mean target 2.866 · arousal 2.95 · rms -20.8dB
instruction text adrenaline flooding, heart racing, hyper-charged, electric with explosive, overflowing euphoria, impossible to sit still.
#1 · text 1
target 4.716arousal 4.121valence 2.977rms -17.6dBexpress 3.350auth 3.419invWER 1.00
#2 · text 1
target 3.771arousal 3.305valence 2.005rms -20.6dBexpress 2.208auth 3.774invWER 1.00
#3 · text 1
target 3.470arousal 3.515valence 2.057rms -19.7dBexpress 3.156auth 3.587invWER 1.00
s14_loudness explicit maximum-volume wording mean target 2.854 · arousal 3.02 · rms -22.2dB
instruction text extremely loud, maximum volume, projecting to the back of a stadium, ecstatic at full force.
#1 · text 1
target 5.097arousal 3.910valence 3.684rms -24.6dBexpress 3.989auth 3.576invWER 1.00
#2 · text 1
target 4.454arousal 4.088valence 2.489rms -21.8dBexpress 3.049auth 3.480invWER 1.00
#3 · text 1
target 4.165arousal 3.537valence 2.448rms -21.2dBexpress 3.189auth 3.485invWER 0.79
s18_vocal_desc describe the raw unstable voice mean target 2.850 · arousal 2.90 · rms -20.4dB
instruction text a harsh, raw, unstable voice — pitch swinging wildly, volume surging, distorted by explosive, overflowing euphoria.
#1 · text 1
target 4.116arousal 3.988valence 2.646rms -20.6dBexpress 2.366auth 3.533invWER 1.00
#2 · text 1
target 3.325arousal 3.335valence 2.846rms -22.6dBexpress 3.686auth 3.357invWER 1.00
#3 · text 3
target 3.305arousal 2.991valence 2.412rms -20.5dBexpress 2.676auth 3.384invWER 1.00
s06_physical physical body cues mean target 2.838 · arousal 3.00 · rms -19.9dB
instruction text breathless with laughter, voice bursting with energy, almost singing. every word shaped by the body's reaction.
#1 · text 1
target 4.848arousal 4.398valence 2.786rms -21.3dBexpress 2.486auth 3.648invWER 1.00
#2 · text 1
target 4.776arousal 3.879valence 3.035rms -17.4dBexpress 3.376auth 3.531invWER 0.88
#3 · text 1
target 3.720arousal 3.512valence 1.986rms -20.7dBexpress 2.965auth 3.652invWER 1.00
s08_parenthetical stage directions in (parentheses) mean target 2.722 · arousal 2.88 · rms -21.8dB
instruction text (laugh and shout with delighting) (breathless with laughter) (losing control) the delivery is ecstatic throughout.
#1 · text 1
target 4.406arousal 3.961valence 3.053rms -21.8dBexpress 3.028auth 3.346invWER 0.94
#2 · text 1
target 3.847arousal 3.548valence 2.117rms -21.7dBexpress 2.536auth 3.501invWER 0.97
#3 · text 1
target 3.621arousal 3.434valence 2.965rms -21.8dBexpress 3.777auth 3.502invWER 1.00
disgust — best: s08_parenthetical (target 1.41 vs control 1.38) s08_parenthetical ★ stage directions in (parentheses) mean target 1.409 · arousal 2.53 · rms -22.7dB
instruction text (gag and recoiling) (lip curled) (losing control) the delivery is disgusted throughout.
#1 · text 2
target 2.065arousal 2.288valence -1.026rms -33.4dBexpress 2.221auth 3.183invWER 1.00
#2 · text 1
target 1.972arousal 2.519valence -0.311rms -17.0dBexpress 2.290auth 3.244invWER 0.97
#3 · text 1
target 1.790arousal 3.043valence 1.831rms -19.6dBexpress 2.092auth 3.365invWER 0.97
s20_neutral_ctrl ○ neutral control (no emotion) mean target 1.378 · arousal 2.18 · rms -22.8dB
instruction text a person speaking in a neutral, even tone. high quality recording.
#1 · text 1
target 2.820arousal 3.004valence 1.220rms -20.9dBexpress 1.888auth 3.329invWER 0.97
#2 · text 2
target 2.236arousal 2.414valence 2.204rms -21.6dBexpress 2.054auth 3.402invWER 1.00
#3 · text 2
target 1.755arousal 1.419valence 0.496rms -20.8dBexpress 2.110auth 3.161invWER 1.00
s14_loudness explicit maximum-volume wording mean target 1.350 · arousal 2.51 · rms -20.1dB
instruction text extremely loud, maximum volume, projecting to the back of a stadium, disgusted at full force.
#1 · text 1
target 2.773arousal 2.946valence 1.034rms -20.4dBexpress 1.685auth 3.409invWER 0.94
#2 · text 2
target 2.240arousal 2.152valence 0.590rms -21.1dBexpress 2.094auth 3.092invWER 1.00
#3 · text 1
target 2.151arousal 2.746valence 0.770rms -19.8dBexpress 1.944auth 3.179invWER 0.97
s11_escalation instruct an escalation arc mean target 1.265 · arousal 2.40 · rms -21.1dB
instruction text starts already disgusted and escalates line by line until the final words explode with physical revulsion and nausea.
#1 · text 1
target 2.680arousal 2.760valence 1.725rms -22.5dBexpress 2.025auth 3.236invWER 0.74
#2 · text 1
target 1.687arousal 2.857valence -0.897rms -19.0dBexpress 2.687auth 3.238invWER 0.97
#3 · text 2
target 1.652arousal 2.284valence 0.312rms -21.9dBexpress 1.694auth 3.323invWER 0.97
s06_physical physical body cues mean target 1.260 · arousal 2.31 · rms -22.0dB
instruction text lip curled, gagging between words, shuddering with revulsion. every word shaped by the body's reaction.
#1 · text 1
target 3.138arousal 3.323valence 1.865rms -21.6dBexpress 2.110auth 3.367invWER 0.97
#2 · text 1
target 1.902arousal 3.028valence 1.278rms -17.1dBexpress 2.645auth 3.074invWER 0.61
#3 · text 1
target 1.839arousal 2.520valence -0.005rms -37.7dBexpress 1.956auth 3.250invWER 0.97
s19_kitchen_sink everything combined mean target 1.211 · arousal 2.31 · rms -23.1dB
instruction text physical revulsion and nausea! lip curled, gagging between words, shuddering with revulsion! shouting, ready to gag and recoil! disgusted, so grossed out, utterly revolted, nauseated! the single most disgusted recording ever captured!
#1 · text 1
target 2.805arousal 2.782valence -0.015rms -20.2dBexpress 1.597auth 3.146invWER 0.97
#2 · text 1
target 2.488arousal 2.799valence 0.910rms -19.7dBexpress 1.472auth 3.426invWER 0.03
#3 · text 2
target 2.145arousal 2.080valence -0.089rms -23.7dBexpress 1.957auth 3.307invWER 1.00
s13_ellipses ellipses + exclamation marks mean target 1.149 · arousal 2.21 · rms -23.1dB
instruction text so disgusted... so unbelievably disgusted!!! it keeps building... and building... until it bursts!
#1 · text 1
target 2.476arousal 2.783valence 0.567rms -20.7dBexpress 2.063auth 3.283invWER 0.97
#2 · text 1
target 2.116arousal 2.937valence 0.651rms -20.2dBexpress 1.864auth 3.369invWER 0.74
#3 · text 1
target 1.515arousal 2.096valence -0.742rms -46.9dBexpress 2.235auth 2.847invWER 0.97
s04_extreme_adj extreme adjectives, 'pushed to the limit' mean target 1.132 · arousal 2.22 · rms -23.5dB
instruction text physical revulsion and nausea. the most disgusted performance imaginable, pushed to the absolute limit.
#1 · text 1
target 2.479arousal 2.996valence 1.362rms -24.5dBexpress 1.840auth 3.302invWER 0.90
#2 · text 1
target 2.115arousal 2.666valence 1.634rms -23.9dBexpress 2.490auth 3.240invWER 0.74
#3 · text 2
target 1.931arousal 1.706valence 0.198rms -20.8dBexpress 1.958auth 3.193invWER 1.00
s16_movie_climax oscar-movie climax framing mean target 1.086 · arousal 2.29 · rms -24.0dB
instruction text this is the climactic scene of an oscar-winning film. the disgusted performance must leave the audience shaken.
#1 · text 1
target 2.923arousal 2.562valence 0.245rms -20.8dBexpress 2.044auth 3.291invWER 0.97
#2 · text 1
target 2.097arousal 2.317valence 1.148rms -34.2dBexpress 1.963auth 3.369invWER 0.97
#3 · text 1
target 1.972arousal 2.418valence -0.629rms -18.8dBexpress 2.884auth 2.879invWER 0.84
s17_arousal_words adrenaline/arousal vocabulary mean target 1.058 · arousal 2.43 · rms -20.6dB
instruction text adrenaline flooding, heart racing, hyper-charged, electric with physical revulsion and nausea, impossible to sit still.
#1 · text 1
target 2.157arousal 3.099valence 1.455rms -19.5dBexpress 1.884auth 3.324invWER 0.97
#2 · text 1
target 2.048arousal 2.237valence -0.500rms -21.5dBexpress 1.825auth 3.177invWER 0.97
#3 · text 2
target 1.963arousal 1.981valence -0.883rms -17.9dBexpress 2.207auth 3.156invWER 0.78
s10_actor_coach method-actor coaching language mean target 1.049 · arousal 2.28 · rms -22.8dB
instruction text you are a method actor. fully inhabit this: give everything, hold nothing back, the disgusted feeling must be devastatingly real.
#1 · text 2
target 2.033arousal 1.715valence -1.746rms -29.7dBexpress 2.406auth 2.970invWER 0.97
#2 · text 1
target 1.826arousal 3.050valence -0.800rms -19.5dBexpress 2.873auth 3.031invWER 0.97
#3 · text 1
target 1.812arousal 2.733valence 0.005rms -33.1dBexpress 1.659auth 3.174invWER 0.97
s15_contrast negate calm ('not calm, not measured') mean target 1.011 · arousal 2.22 · rms -24.5dB
instruction text not calm, not measured, not gentle — nothing but physical revulsion and nausea, from the first word to the last.
#1 · text 1
target 2.762arousal 2.877valence 1.001rms -25.5dBexpress 2.218auth 3.304invWER 0.90
#2 · text 1
target 2.132arousal 2.333valence 1.622rms -33.7dBexpress 2.419auth 3.267invWER 0.97
#3 · text 1
target 1.846arousal 1.868valence 1.382rms -38.9dBexpress 1.823auth 3.407invWER 0.97
s01_minimal just the emotion word mean target 1.010 · arousal 2.38 · rms -21.8dB
instruction text disgusted.
#1 · text 1
target 2.491arousal 3.133valence 1.545rms -20.7dBexpress 2.083auth 3.247invWER 0.97
#2 · text 1
target 1.588arousal 2.141valence -1.341rms -26.3dBexpress 2.837auth 3.092invWER 0.94
#3 · text 1
target 1.506arousal 2.686valence 1.727rms -19.5dBexpress 2.175auth 3.497invWER 0.71
s07_shouting explicit shouting/screaming cues mean target 0.999 · arousal 2.06 · rms -22.6dB
instruction text shouting at the top of their lungs, voice raw and cracking, ready to gag and recoil between the words.
#1 · text 1
target 1.912arousal 2.009valence -0.667rms -20.0dBexpress 2.255auth 2.964invWER 0.97
#2 · text 1
target 1.745arousal 2.061valence 0.005rms -49.0dBexpress 1.814auth 2.978invWER 0.97
#3 · text 2
target 1.678arousal 1.603valence 0.896rms -22.2dBexpress 1.802auth 3.350invWER 1.00
s09_scenario ground it in a concrete scenario mean target 0.987 · arousal 2.34 · rms -22.1dB
instruction text you just found something unspeakably rotten in your kitchen. react in the moment, completely disgusted, no restraint.
#1 · text 1
target 2.389arousal 3.070valence 2.156rms -18.8dBexpress 2.220auth 3.219invWER 0.90
#2 · text 1
target 2.070arousal 2.733valence -0.276rms -20.2dBexpress 1.769auth 3.333invWER 0.97
#3 · text 2
target 1.193arousal 2.466valence 1.661rms -19.6dBexpress 2.258auth 3.408invWER 0.94
s03_dramabox_base dramabox baseline sentence mean target 0.984 · arousal 2.25 · rms -25.4dB
instruction text a person speaking with intense disgusted emotion, delivering this high-quality studio voice recording.
#1 · text 1
target 2.618arousal 2.824valence 1.883rms -23.3dBexpress 1.887auth 3.235invWER 0.90
#2 · text 1
target 1.957arousal 2.941valence -1.682rms -21.0dBexpress 2.849auth 3.102invWER 0.97
#3 · text 1
target 1.429arousal 2.548valence -1.527rms -27.2dBexpress 2.663auth 2.967invWER 0.97
s05_repetition repeat the emotion word many times mean target 0.943 · arousal 2.07 · rms -25.8dB
instruction text disgusted, so grossed out, utterly revolted, nauseated. disgusted, so grossed out, utterly revolted, nauseated! more disgusted than you have ever been in your life!
#1 · text 1
target 2.056arousal 2.042valence -0.075rms -31.4dBexpress 2.173auth 2.900invWER 0.97
#2 · text 2
target 1.810arousal 1.583valence -0.458rms -32.9dBexpress 1.470auth 3.214invWER 0.97
#3 · text 2
target 1.632arousal 1.971valence -0.573rms -21.1dBexpress 2.717auth 3.127invWER 1.00
s12_imperative second-person commands: 'be furious!' mean target 0.914 · arousal 2.14 · rms -23.5dB
instruction text be disgusted! gag and recoil! let it consume you completely! do not hold back, not even a little!
#1 · text 1
target 2.187arousal 2.633valence 1.535rms -20.4dBexpress 2.316auth 3.175invWER 0.97
#2 · text 1
target 1.837arousal 2.974valence 0.713rms -22.1dBexpress 1.944auth 3.320invWER 0.97
#3 · text 2
target 1.275arousal 1.657valence -1.537rms -25.9dBexpress 2.094auth 3.159invWER 1.00
s02_plain plain 'a person speaking, X' mean target 0.887 · arousal 2.09 · rms -26.7dB
instruction text a person speaking, disgusted.
#1 · text 2
target 1.939arousal 2.331valence 1.003rms -23.3dBexpress 2.090auth 3.344invWER 0.78
#2 · text 1
target 1.627arousal 2.182valence -0.617rms -30.3dBexpress 2.614auth 3.054invWER 0.74
#3 · text 2
target 1.430arousal 2.350valence -1.582rms -21.4dBexpress 2.627auth 3.012invWER 1.00
s18_vocal_desc describe the raw unstable voice mean target 0.820 · arousal 2.15 · rms -23.1dB
instruction text a harsh, raw, unstable voice — pitch swinging wildly, volume surging, distorted by physical revulsion and nausea.
#1 · text 1
target 1.919arousal 2.508valence -0.832rms -19.7dBexpress 2.266auth 3.194invWER 0.97
#2 · text 1
target 1.890arousal 2.281valence -1.578rms -29.0dBexpress 2.682auth 2.920invWER 0.97
#3 · text 2
target 1.506arousal 1.272valence -0.071rms -41.1dBexpress 1.795auth 3.044invWER 0.97
surprise — best: s20_neutral_ctrl (target 2.31 vs control 2.31) s20_neutral_ctrl ★ neutral control (no emotion) mean target 2.312 · arousal 2.66 · rms -20.6dB
instruction text a person speaking in a neutral, even tone. high quality recording.
#1 · text 1
target 4.375arousal 2.890valence 0.848rms -17.1dBexpress 2.131auth 3.507invWER 1.00
#2 · text 1
target 3.696arousal 2.842valence 0.799rms -19.9dBexpress 1.783auth 3.372invWER 1.00
#3 · text 1
target 2.703arousal 2.560valence 0.951rms -19.1dBexpress 1.722auth 3.428invWER 1.00
s13_ellipses ellipses + exclamation marks mean target 2.174 · arousal 2.70 · rms -21.4dB
instruction text so astonished... so unbelievably astonished!!! it keeps building... and building... until it bursts!
#1 · text 1
target 3.329arousal 2.417valence -0.351rms -21.6dBexpress 2.056auth 3.175invWER 1.00
#2 · text 1
target 2.840arousal 2.405valence -0.231rms -20.1dBexpress 2.391auth 3.426invWER 1.00
#3 · text 1
target 2.723arousal 2.564valence 1.204rms -20.1dBexpress 1.628auth 3.348invWER 1.00
s10_actor_coach method-actor coaching language mean target 2.087 · arousal 2.67 · rms -20.5dB
instruction text you are a method actor. fully inhabit this: give everything, hold nothing back, the astonished feeling must be devastatingly real.
#1 · text 1
target 2.754arousal 2.599valence 1.047rms -24.2dBexpress 1.892auth 3.372invWER 1.00
#2 · text 2
target 2.730arousal 3.347valence 1.686rms -20.5dBexpress 2.547auth 3.318invWER 1.00
#3 · text 1
target 2.435arousal 2.301valence 0.493rms -21.1dBexpress 2.250auth 3.347invWER 0.52
s16_movie_climax oscar-movie climax framing mean target 2.070 · arousal 2.53 · rms -20.4dB
instruction text this is the climactic scene of an oscar-winning film. the astonished performance must leave the audience shaken.
#1 · text 1
target 3.504arousal 2.906valence -2.090rms -21.1dBexpress 3.047auth 3.122invWER 0.84
#2 · text 1
target 2.977arousal 2.543valence -0.000rms -20.2dBexpress 1.658auth 3.259invWER 1.00
#3 · text 1
target 2.662arousal 2.001valence -0.001rms -19.1dBexpress 2.158auth 3.255invWER 1.00
s06_physical physical body cues mean target 2.020 · arousal 2.53 · rms -22.4dB
instruction text gasping, words tumbling out, voice jumping an octave. every word shaped by the body's reaction.
#1 · text 1
target 2.752arousal 3.086valence 1.062rms -21.8dBexpress 1.942auth 3.328invWER 1.00
#2 · text 1
target 2.541arousal 2.534valence -0.927rms -21.0dBexpress 2.180auth 3.141invWER 1.00
#3 · text 1
target 2.323arousal 2.527valence 0.166rms -19.2dBexpress 1.820auth 3.327invWER 0.97
s18_vocal_desc describe the raw unstable voice mean target 1.994 · arousal 2.48 · rms -20.6dB
instruction text a harsh, raw, unstable voice — pitch swinging wildly, volume surging, distorted by jaw-dropping, world-upending shock.
#1 · text 1
target 2.642arousal 2.204valence -0.474rms -20.6dBexpress 2.215auth 2.991invWER 1.00
#2 · text 3
target 2.368arousal 3.119valence 2.393rms -19.1dBexpress 2.506auth 3.375invWER 0.96
#3 · text 1
target 2.350arousal 2.256valence -1.104rms -19.9dBexpress 2.439auth 2.940invWER 1.00
s14_loudness explicit maximum-volume wording mean target 1.955 · arousal 2.46 · rms -21.8dB
instruction text extremely loud, maximum volume, projecting to the back of a stadium, astonished at full force.
#1 · text 1
target 4.109arousal 3.122valence -2.315rms -20.4dBexpress 2.901auth 2.992invWER 1.00
#2 · text 2
target 2.760arousal 2.577valence 1.778rms -21.2dBexpress 3.477auth 3.194invWER 1.00
#3 · text 1
target 2.608arousal 2.013valence 0.019rms -40.4dBexpress 2.045auth 3.478invWER 1.00
s03_dramabox_base dramabox baseline sentence mean target 1.952 · arousal 2.62 · rms -20.6dB
instruction text a person speaking with intense astonished emotion, delivering this high-quality studio voice recording.
#1 · text 1
target 2.942arousal 2.985valence 1.458rms -21.3dBexpress 1.730auth 3.235invWER 1.00
#2 · text 1
target 2.502arousal 2.578valence 0.003rms -20.7dBexpress 1.965auth 3.312invWER 1.00
#3 · text 1
target 2.404arousal 3.167valence 1.666rms -19.9dBexpress 2.288auth 3.207invWER 0.74
s04_extreme_adj extreme adjectives, 'pushed to the limit' mean target 1.947 · arousal 2.66 · rms -22.7dB
instruction text jaw-dropping, world-upending shock. the most astonished performance imaginable, pushed to the absolute limit.
#1 · text 1
target 2.885arousal 3.177valence 1.337rms -24.1dBexpress 1.820auth 3.438invWER 1.00
#2 · text 1
target 2.769arousal 2.198valence 0.488rms -20.2dBexpress 2.079auth 3.439invWER 1.00
#3 · text 1
target 2.407arousal 2.895valence 1.474rms -20.5dBexpress 1.991auth 3.190invWER 1.00
s02_plain plain 'a person speaking, X' mean target 1.940 · arousal 2.55 · rms -20.0dB
instruction text a person speaking, astonished.
#1 · text 1
target 3.125arousal 2.815valence 1.385rms -19.8dBexpress 1.697auth 3.429invWER 1.00
#2 · text 1
target 2.859arousal 2.477valence -1.895rms -21.6dBexpress 2.733auth 3.277invWER 1.00
#3 · text 2
target 2.438arousal 2.547valence -0.235rms -12.2dBexpress 2.750auth 3.455invWER 1.00
s19_kitchen_sink everything combined mean target 1.918 · arousal 2.60 · rms -20.3dB
instruction text jaw-dropping, world-upending shock! gasping, words tumbling out, voice jumping an octave! shouting, ready to gasp! shocked, completely stunned, absolutely astonished, flabbergasted! the single most astonished recording ever captured!
#1 · text 1
target 2.619arousal 2.610valence -0.556rms -19.0dBexpress 2.355auth 3.497invWER 0.90
#2 · text 1
target 2.586arousal 2.568valence -0.304rms -20.5dBexpress 1.843auth 3.428invWER 1.00
#3 · text 1
target 2.188arousal 2.580valence 1.073rms -23.6dBexpress 2.851auth 3.356invWER 1.00
s05_repetition repeat the emotion word many times mean target 1.912 · arousal 2.48 · rms -21.4dB
instruction text shocked, completely stunned, absolutely astonished, flabbergasted. shocked, completely stunned, absolutely astonished, flabbergasted! more astonished than you have ever been in your life!
#1 · text 1
target 2.624arousal 3.242valence 1.725rms -21.6dBexpress 2.605auth 3.575invWER 1.00
#2 · text 1
target 2.470arousal 2.487valence 0.002rms -22.1dBexpress 2.099auth 3.257invWER 1.00
#3 · text 1
target 2.229arousal 3.244valence 1.776rms -21.6dBexpress 2.189auth 3.282invWER 0.94
s07_shouting explicit shouting/screaming cues mean target 1.901 · arousal 2.53 · rms -21.6dB
instruction text shouting at the top of their lungs, voice raw and cracking, ready to gasp between the words.
#1 · text 1
target 2.993arousal 2.878valence -2.221rms -20.7dBexpress 2.840auth 3.316invWER 1.00
#2 · text 1
target 2.317arousal 1.590valence 0.385rms -20.8dBexpress 1.823auth 3.549invWER 1.00
#3 · text 1
target 2.293arousal 2.568valence -1.644rms -21.3dBexpress 2.741auth 3.122invWER 1.00
s15_contrast negate calm ('not calm, not measured') mean target 1.897 · arousal 2.35 · rms -22.3dB
instruction text not calm, not measured, not gentle — nothing but jaw-dropping, world-upending shock, from the first word to the last.
#1 · text 1
target 3.392arousal 3.132valence -0.740rms -23.8dBexpress 2.546auth 3.328invWER 1.00
#2 · text 1
target 2.979arousal 2.455valence 0.810rms -21.9dBexpress 1.984auth 3.276invWER 1.00
#3 · text 1
target 2.469arousal 2.739valence 0.036rms -20.7dBexpress 1.758auth 3.365invWER 1.00
s12_imperative second-person commands: 'be furious!' mean target 1.812 · arousal 2.62 · rms -22.7dB
instruction text be astonished! gasp! let it consume you completely! do not hold back, not even a little!
#1 · text 2
target 2.545arousal 2.042valence 0.596rms -21.3dBexpress 2.497auth 3.253invWER 1.00
#2 · text 3
target 2.490arousal 3.367valence 2.603rms -25.3dBexpress 2.033auth 3.416invWER 1.00
#3 · text 1
target 2.406arousal 3.126valence 2.052rms -19.1dBexpress 1.847auth 2.993invWER 0.55
s17_arousal_words adrenaline/arousal vocabulary mean target 1.806 · arousal 2.29 · rms -22.8dB
instruction text adrenaline flooding, heart racing, hyper-charged, electric with jaw-dropping, world-upending shock, impossible to sit still.
#1 · text 3
target 2.566arousal 3.128valence 2.111rms -19.9dBexpress 2.033auth 3.189invWER 0.88
#2 · text 1
target 2.512arousal 3.106valence 1.221rms -18.7dBexpress 2.002auth 3.334invWER 1.00
#3 · text 2
target 1.954arousal 2.260valence 1.391rms -19.4dBexpress 2.413auth 3.376invWER 1.00
s11_escalation instruct an escalation arc mean target 1.770 · arousal 2.59 · rms -22.0dB
instruction text starts already astonished and escalates line by line until the final words explode with jaw-dropping, world-upending shock.
#1 · text 1
target 2.921arousal 2.673valence 0.991rms -19.9dBexpress 2.022auth 3.402invWER 0.97
#2 · text 1
target 2.518arousal 2.804valence -0.004rms -21.2dBexpress 1.935auth 3.454invWER 1.00
#3 · text 1
target 2.323arousal 2.362valence -0.200rms -14.9dBexpress 1.940auth 3.113invWER 1.00
s09_scenario ground it in a concrete scenario mean target 1.687 · arousal 2.58 · rms -22.9dB
instruction text you just learned a secret that changes everything you believed. react in the moment, completely astonished, no restraint.
#1 · text 1
target 2.840arousal 2.865valence 1.403rms -24.1dBexpress 2.217auth 3.331invWER 1.00
#2 · text 1
target 2.394arousal 2.511valence -0.536rms -23.8dBexpress 2.008auth 3.501invWER 1.00
#3 · text 1
target 2.338arousal 1.974valence -0.766rms -26.0dBexpress 1.909auth 3.478invWER 1.00
s01_minimal just the emotion word mean target 1.663 · arousal 2.52 · rms -23.3dB
instruction text astonished.
#1 · text 1
target 2.775arousal 3.717valence -0.379rms -20.3dBexpress 2.218auth 3.373invWER 0.81
#2 · text 1
target 2.648arousal 2.663valence 1.467rms -23.2dBexpress 2.076auth 3.323invWER 1.00
#3 · text 2
target 2.362arousal 3.080valence 2.119rms -20.1dBexpress 2.100auth 3.329invWER 1.00
s08_parenthetical stage directions in (parentheses) mean target 1.439 · arousal 2.16 · rms -21.5dB
instruction text (gasping) (gasping) (losing control) the delivery is astonished throughout.
#1 · text 1
target 2.768arousal 2.918valence 1.334rms -19.1dBexpress 1.837auth 3.376invWER 1.00
#2 · text 1
target 2.481arousal 2.778valence 1.256rms -20.3dBexpress 1.820auth 3.486invWER 0.94
#3 · text 2
target 2.047arousal 2.179valence 1.442rms -16.8dBexpress 2.086auth 3.436invWER 0.97
Speaker-similarity grids — 8 templates × 3 reference voices top-3 takes per group by voice similarity (cosine of Chatterbox VoiceEncoder embeddings vs the reference clip). remember: the EMPTY instruction wins.
p1_empty · chris mean sim 0.904
instruction text (empty instruction)
#1
spk sim 0.939arousal 0.697valence 0.005rms -35.5dBexpress 1.192auth 2.896invWER 0.97genu 0.45
#2
spk sim 0.899arousal 0.832valence 0.005rms -32.3dBexpress 1.743auth 2.645invWER 0.85genu 2.05
#3
spk sim 0.895arousal 1.048valence 0.005rms -32.0dBexpress 1.873auth 2.605invWER 0.76genu 3.04
p1_empty · samantha mean sim 0.942
instruction text (empty instruction)
#1
spk sim 0.968arousal 0.826valence 0.037rms -20.9dBexpress 1.862auth 3.056invWER 1.00genu 2.10
#2
spk sim 0.962arousal 0.648valence 0.005rms -19.9dBexpress 1.491auth 3.132invWER 0.97genu 0.37
#3
spk sim 0.938arousal 0.957valence -0.034rms -20.3dBexpress 2.057auth 2.994invWER 0.85genu 1.36
p1_empty · goblin mean sim 0.945
instruction text (empty instruction)
#1
spk sim 0.961arousal 1.657valence -0.332rms -20.6dBexpress 2.398auth 2.692invWER 0.27genu 2.40
#2
spk sim 0.951arousal 1.109valence 0.006rms -21.2dBexpress 2.449auth 2.995invWER 0.97genu 1.91
#3
spk sim 0.939arousal 1.420valence 0.005rms -19.8dBexpress 2.300auth 2.852invWER 0.70genu 1.63
p7_neutral_ctrl · chris mean sim 0.892
instruction text a person speaking naturally. high quality recording.
#1
spk sim 0.905arousal 0.969valence 0.297rms -33.7dBexpress 1.554auth 3.038invWER 1.00genu 2.37
#2
spk sim 0.890arousal 0.977valence 0.204rms -29.1dBexpress 1.907auth 2.687invWER 0.88genu 2.38
#3
spk sim 0.888arousal 1.080valence 0.255rms -25.6dBexpress 1.688auth 2.791invWER 0.94genu 2.34
p7_neutral_ctrl · samantha mean sim 0.959
instruction text a person speaking naturally. high quality recording.
#1
spk sim 0.974arousal 0.854valence 0.009rms -21.0dBexpress 2.150auth 3.045invWER 0.94genu 2.64
#2
spk sim 0.964arousal 1.044valence 1.473rms -20.9dBexpress 2.066auth 3.111invWER 1.00genu 2.28
#3
spk sim 0.958arousal 0.849valence -0.220rms -19.9dBexpress 2.247auth 3.030invWER 0.36genu 1.84
p7_neutral_ctrl · goblin mean sim 0.940
instruction text a person speaking naturally. high quality recording.
#1
spk sim 0.951arousal 1.411valence 0.005rms -20.5dBexpress 2.153auth 2.734invWER 0.94genu 0.95
#2
spk sim 0.938arousal 1.680valence 0.927rms -19.0dBexpress 2.146auth 2.700invWER 1.00genu 0.71
#3
spk sim 0.936arousal 1.679valence -0.086rms -20.9dBexpress 2.659auth 3.067invWER 0.94genu 2.65
p2_exact · chris mean sim 0.901
instruction text speak exactly like the reference voice.
#1
spk sim 0.926arousal 0.956valence 0.005rms -32.9dBexpress 1.652auth 2.729invWER 0.52genu 2.66
#2
spk sim 0.909arousal 1.187valence 0.005rms -31.0dBexpress 1.620auth 2.641invWER 0.94genu 2.95
#3
spk sim 0.900arousal 1.068valence 0.005rms -32.4dBexpress 2.075auth 2.849invWER 0.82genu 2.67
p2_exact · samantha mean sim 0.954
instruction text speak exactly like the reference voice.
#1
spk sim 0.966arousal 1.030valence 0.070rms -19.4dBexpress 2.110auth 2.972invWER 0.36genu 1.53
#2
spk sim 0.963arousal 0.955valence 0.005rms -21.2dBexpress 1.829auth 2.883invWER 1.00genu 1.74
#3
spk sim 0.953arousal 0.881valence 0.005rms -19.9dBexpress 1.612auth 3.021invWER 0.97genu 1.33
p2_exact · goblin mean sim 0.933
instruction text speak exactly like the reference voice.
#1
spk sim 0.957arousal 1.287valence 0.005rms -19.1dBexpress 2.326auth 2.794invWER 0.97genu 0.67
#2
spk sim 0.954arousal 1.765valence 0.613rms -20.8dBexpress 2.579auth 2.701invWER 0.94genu 2.20
#3
spk sim 0.941arousal 0.955valence 0.124rms -20.1dBexpress 2.165auth 2.874invWER 1.00genu 1.12
p4_same_person · chris mean sim 0.888
instruction text the same person as in the reference recording continues speaking, in the same room, on the same microphone.
#1
spk sim 0.917arousal 0.801valence 0.005rms -30.5dBexpress 1.314auth 2.569invWER 0.82genu 3.20
#2
spk sim 0.906arousal 2.028valence 1.505rms -31.4dBexpress 2.099auth 2.740invWER 0.15genu 1.29
#3
spk sim 0.878arousal 0.689valence 0.005rms -31.9dBexpress 1.709auth 2.895invWER 0.97genu 1.79
p4_same_person · samantha mean sim 0.958
instruction text the same person as in the reference recording continues speaking, in the same room, on the same microphone.
#1
spk sim 0.961arousal 0.838valence 0.005rms -21.4dBexpress 1.565auth 2.960invWER 1.00genu 0.24
#2
spk sim 0.961arousal 0.804valence 0.005rms -20.1dBexpress 1.593auth 2.808invWER 0.97genu 0.45
#3
spk sim 0.959arousal 1.296valence 1.025rms -22.3dBexpress 2.279auth 3.162invWER 0.97genu 1.18
p4_same_person · goblin mean sim 0.938
instruction text the same person as in the reference recording continues speaking, in the same room, on the same microphone.
#1
spk sim 0.942arousal 1.155valence 0.005rms -19.6dBexpress 2.293auth 2.697invWER 1.00genu 0.61
#2
spk sim 0.936arousal 2.226valence 0.005rms -21.1dBexpress 2.412auth 2.578invWER 0.12genu 1.60
#3
spk sim 0.936arousal 1.650valence 0.005rms -20.4dBexpress 2.454auth 2.699invWER 0.39genu 3.26
p5_identical · chris mean sim 0.889
instruction text match the reference voice identically. identical tone, identical pitch, identical rhythm. nothing about the voice may change.
#1
spk sim 0.922arousal 0.728valence 0.005rms -31.1dBexpress 1.513auth 2.788invWER 0.73genu 2.88
#2
spk sim 0.910arousal 0.827valence 0.005rms -29.3dBexpress 1.671auth 2.801invWER 0.94genu 2.03
#3
spk sim 0.884arousal 0.952valence 0.091rms -29.7dBexpress 2.235auth 2.999invWER 0.88genu 1.31
p5_identical · samantha mean sim 0.955
instruction text match the reference voice identically. identical tone, identical pitch, identical rhythm. nothing about the voice may change.
#1
spk sim 0.970arousal 0.905valence 0.005rms -19.8dBexpress 1.701auth 2.850invWER 0.97genu 0.78
#2
spk sim 0.956arousal 0.922valence 0.004rms -19.9dBexpress 2.063auth 2.904invWER 0.55genu 1.88
#3
spk sim 0.955arousal 0.755valence -0.374rms -21.7dBexpress 1.983auth 3.059invWER 1.00genu 1.57
p5_identical · goblin mean sim 0.937
instruction text match the reference voice identically. identical tone, identical pitch, identical rhythm. nothing about the voice may change.
#1
spk sim 0.960arousal 1.698valence 0.912rms -21.1dBexpress 2.631auth 2.942invWER 0.76genu 2.11
#2
spk sim 0.936arousal 1.017valence 0.120rms -20.8dBexpress 2.287auth 3.122invWER 1.00genu 2.40
#3
spk sim 0.934arousal 2.231valence 0.154rms -8.8dBexpress 2.887auth 2.687invWER 0.85genu 2.57
p8_do_not_change · chris mean sim 0.878
instruction text same speaker, same voice, same energy as the reference audio... do not change anything about the voice, not the pitch, not the pace, not the accent.
#1
spk sim 0.928arousal 1.076valence 0.063rms -34.0dBexpress 1.935auth 2.916invWER 0.94genu 3.00
#2
spk sim 0.896arousal 0.762valence 0.005rms -31.0dBexpress 1.431auth 2.774invWER 1.00genu 1.68
#3
spk sim 0.863arousal 0.928valence 0.001rms -29.9dBexpress 1.838auth 2.288invWER 0.73genu 2.06
p8_do_not_change · samantha mean sim 0.952
instruction text same speaker, same voice, same energy as the reference audio... do not change anything about the voice, not the pitch, not the pace, not the accent.
#1
spk sim 0.963arousal 0.869valence 0.183rms -21.3dBexpress 2.172auth 3.014invWER 1.00genu 2.59
#2
spk sim 0.955arousal 0.539valence 0.005rms -21.9dBexpress 2.093auth 3.103invWER 0.97genu 1.13
#3
spk sim 0.948arousal 1.019valence 0.005rms -21.3dBexpress 1.939auth 2.938invWER 0.97genu 1.93
p8_do_not_change · goblin mean sim 0.943
instruction text same speaker, same voice, same energy as the reference audio... do not change anything about the voice, not the pitch, not the pace, not the accent.
#1
spk sim 0.950arousal 1.590valence 0.004rms -20.2dBexpress 2.392auth 2.962invWER 0.64genu 1.54
#2
spk sim 0.949arousal 1.247valence 0.005rms -20.2dBexpress 2.244auth 2.826invWER 0.97genu 1.76
#3
spk sim 0.937arousal 1.461valence 0.005rms -23.1dBexpress 2.365auth 2.595invWER 1.00genu 2.63
p3_clone · chris mean sim 0.881
instruction text clone the reference speaker's voice precisely: the same timbre, the same accent, the same pacing, the same habits.
#1
spk sim 0.891arousal 0.680valence 0.005rms -33.0dBexpress 1.581auth 2.652invWER 0.97genu 1.89
#2
spk sim 0.891arousal 0.604valence 0.005rms -27.1dBexpress 1.729auth 2.851invWER 1.00genu 1.05
#3
spk sim 0.879arousal 1.240valence 0.005rms -31.1dBexpress 1.703auth 2.681invWER 0.85genu 2.14
p3_clone · samantha mean sim 0.943
instruction text clone the reference speaker's voice precisely: the same timbre, the same accent, the same pacing, the same habits.
#1
spk sim 0.954arousal 0.891valence 0.004rms -22.0dBexpress 2.174auth 2.915invWER 1.00genu 1.17
#2
spk sim 0.953arousal 1.084valence 0.007rms -19.4dBexpress 1.797auth 3.183invWER 0.61genu 1.27
#3
spk sim 0.938arousal 0.148valence 0.005rms -21.0dBexpress 1.915auth 2.937invWER 1.00genu 1.59
p3_clone · goblin mean sim 0.936
instruction text clone the reference speaker's voice precisely: the same timbre, the same accent, the same pacing, the same habits.
#1
spk sim 0.952arousal 1.572valence -0.380rms -20.8dBexpress 2.337auth 2.808invWER 0.97genu 1.58
#2
spk sim 0.946arousal 1.607valence 0.005rms -19.8dBexpress 2.563auth 2.776invWER 0.39genu 1.54
#3
spk sim 0.937arousal 1.972valence 0.258rms -21.8dBexpress 2.388auth 2.474invWER 0.00genu 2.32
p6_session · chris mean sim 0.829
instruction text imitate the reference speaker perfectly, as if this is the very same recording session, one minute later.
#1
spk sim 0.887arousal 0.900valence 0.005rms -31.9dBexpress 1.462auth 2.786invWER 0.97genu 1.21
#2
spk sim 0.879arousal 1.133valence 0.004rms -25.6dBexpress 1.678auth 2.700invWER 0.97genu 3.20
#3
spk sim 0.797arousal 0.965valence 0.005rms -34.8dBexpress 1.754auth 2.488invWER 0.06genu 1.76
p6_session · samantha mean sim 0.951
instruction text imitate the reference speaker perfectly, as if this is the very same recording session, one minute later.
#1
spk sim 0.959arousal 0.830valence 0.005rms -20.6dBexpress 1.722auth 2.994invWER 0.97genu 0.32
#2
spk sim 0.955arousal 0.940valence 0.594rms -20.8dBexpress 2.212auth 3.093invWER 0.85genu 1.83
#3
spk sim 0.948arousal 1.238valence 0.816rms -19.9dBexpress 2.282auth 3.040invWER 0.64genu 2.15
p6_session · goblin mean sim 0.942
instruction text imitate the reference speaker perfectly, as if this is the very same recording session, one minute later.
#1
spk sim 0.954arousal 1.766valence -0.270rms -20.2dBexpress 2.333auth 2.773invWER 0.97genu 1.29
#2
spk sim 0.950arousal 1.490valence 0.005rms -20.2dBexpress 2.216auth 2.616invWER 0.91genu 1.17
#3
spk sim 0.941arousal 0.812valence 0.004rms -20.6dBexpress 2.330auth 3.002invWER 1.00genu 2.12
Genuineness grids — 8 templates top-3 takes per template (pooled over 3 texts) by the VoiceCLAP genuineness head. the plain control wins.
g1_control mean genu 1.956 · auth 3.26
instruction text a person speaking. high quality recording.
#1 · text 2
genu 4.338arousal 2.605valence 1.376rms -17.8dBexpress 3.058auth 3.252invWER 0.31
#2 · text 1
genu 2.589arousal 1.026valence -1.325rms -28.2dBexpress 2.210auth 2.931invWER 0.94
#3 · text 2
genu 2.246arousal 2.354valence 1.888rms -23.2dBexpress 2.753auth 3.112invWER 1.00
g2_genuine mean genu 1.641 · auth 3.29
instruction text completely genuine and real... like an actual human being caught in an actual moment, not a performance.
#1 · text 2
genu 2.745arousal 2.411valence 1.519rms -25.0dBexpress 2.913auth 3.382invWER 0.22
#2 · text 2
genu 2.572arousal 2.458valence 1.657rms -20.2dBexpress 2.918auth 3.541invWER 0.97
#3 · text 2
genu 2.082arousal 2.512valence 1.945rms -19.5dBexpress 2.969auth 3.090invWER 1.00
g6_anti_acting mean genu 1.551 · auth 3.23
instruction text do not perform. do not act. just be the person. understated, honest, real.
#1 · text 2
genu 3.237arousal 1.989valence 1.279rms -31.1dBexpress 2.436auth 3.302invWER 0.97
#2 · text 1
genu 2.396arousal 1.097valence -1.319rms -20.8dBexpress 2.068auth 2.914invWER 0.97
#3 · text 2
genu 2.221arousal 2.326valence 0.991rms -22.5dBexpress 2.910auth 3.043invWER 1.00
g8_voice_memo mean genu 1.539 · auth 3.40
instruction text an intimate voice memo recorded late at night, honest and unguarded, spoken softly to someone trusted.
#1 · text 1
genu 2.098arousal 1.029valence -0.755rms -19.7dBexpress 1.862auth 3.238invWER 1.00
#2 · text 1
genu 1.880arousal 1.000valence -1.137rms -23.7dBexpress 2.178auth 2.928invWER 0.97
#3 · text 2
genu 1.772arousal 2.414valence 1.570rms -20.7dBexpress 2.953auth 3.344invWER 0.94
g3_hesitations mean genu 1.538 · auth 3.36
instruction text natural hesitations, small breaths, imperfect timing — like a candid recording of a real person who doesn't know they're being recorded.
#1 · text 2
genu 4.038arousal 2.204valence 1.656rms -35.1dBexpress 2.979auth 3.564invWER 0.81
#2 · text 1
genu 1.659arousal 1.159valence -0.226rms -21.4dBexpress 1.868auth 3.200invWER 1.00
#3 · text 3
genu 1.638arousal 2.228valence -2.157rms -30.4dBexpress 2.609auth 3.324invWER 1.00
g4_documentary mean genu 1.490 · auth 3.32
instruction text from a real documentary interview: unscripted, raw, unpolished, true.
#1 · text 2
genu 2.196arousal 2.126valence 1.585rms -29.5dBexpress 2.974auth 3.216invWER 1.00
#2 · text 1
genu 1.901arousal 0.966valence -1.718rms -32.1dBexpress 2.124auth 3.360invWER 1.00
#3 · text 1
genu 1.703arousal 1.104valence -0.964rms -18.6dBexpress 2.231auth 3.038invWER 0.94
g7_method mean genu 1.396 · auth 3.32
instruction text method acting at its deepest: the feeling is genuinely felt, lived-in, truthful, with nothing put on.
#1 · text 1
genu 2.079arousal 1.119valence -1.026rms -20.6dBexpress 2.012auth 3.220invWER 1.00
#2 · text 2
genu 1.827arousal 2.380valence 2.186rms -26.9dBexpress 2.605auth 3.532invWER 0.31
#3 · text 3
genu 1.742arousal 1.960valence -0.691rms -22.4dBexpress 1.947auth 3.550invWER 1.00
g5_spontaneous mean genu 1.338 · auth 3.34
instruction text spontaneous and unrehearsed, organic pacing, real emotion rising and falling on its own, never theatrical.
#1 · text 3
genu 1.971arousal 1.869valence -2.029rms -37.7dBexpress 2.628auth 3.318invWER 1.00
#2 · text 1
genu 1.776arousal 1.075valence -1.372rms -26.7dBexpress 2.355auth 3.434invWER 1.00
#3 · text 2
genu 1.725arousal 2.199valence 1.258rms -21.5dBexpress 2.373auth 3.122invWER 0.91