Each embedded DramaBox clip is reinterpreted by MOSS-TTS-4.55B voice-acting-v2 (the base checkpoint for our LoRAs): the original audio is fed back as the reference together with its own GENERAL/SCRIPT prompt, and the model regenerates the same line as a batched group of 8 candidates. Candidates are ranked by emotional similarity to the original × inverse WER; the top 3 are shown next to the original. Regenerating and reward-ranking cleans up disfluencies and mistakes while staying faithful to the original emotion.
Reward & WER by candidate-selection strategy
Reward = emotional similarity (cosine of 40 EmoNet emotions + arousal + valence, reference vs candidate) × (1−WER). WER from faster-whisper vs the intended script words. Averaged over all 124 clips. Higher reward & emo-sim, lower WER = better. Selecting the best few from a larger group raises reward and cuts WER — the more candidates, the better the top picks.
strategy
mean reward
mean WER
mean emo-sim
single candidate (avg of 8)
0.107
0.862
0.595
best-of-4 (top1 of 4)
0.250
0.648
0.620
best-of-8 (top1 of 8)
0.340
0.499
0.634
top2-of-4
0.180
0.753
0.608
top3-of-8
0.220
0.694
0.614
Top-2-of-4 vs Top-3-of-8: reward 0.180 → 0.220 (+22%), WER 0.753 → 0.694. Doubling the candidate pool (4→8) and taking the top 3 improves the average reward and lowers WER.
#0 · “so it's so big I could barely hear myself all those people just looking. But then the music started and I remembered the”
A feminine voice, voice at peak biological vigor. Timbre: perfect; standard human baseline; standard. Delivery: casual, monologue, conversational. Emotion: relief, interest, pride, contentment, elation. Genuineness 1.631
● ORIGINAL · 28.64s
the embedded DramaBox clip
✦ TOP 1
reward 0.125 · emo-sim 0.615 · WER 0.797 · 33.12s
“So it's so big, I could barely hear myself, all those people just looking. But then the m”
✦ TOP 2
reward 0.062 · emo-sim 0.443 · WER 0.859 · 36.72s
“So it's so big. I could barely hear myself, all those people just looking. But then the mu”
✦ TOP 3
reward 0.021 · emo-sim 0.664 · WER 0.969 · 29.2s
“So it's so big I could barely hear myself all those people just looking. But then the mus”
#1 · “so it's so big I could barely hear myself all those people just looking. But then the music started and I remembered the”
A feminine voice, voice at peak biological vigor. Timbre: perfect; standard human baseline; standard. Delivery: casual, conversational, monologue. Emotion: interest, relief, elation, pride, pleasure ecstasy. Genuineness
● ORIGINAL · 23.65s
the embedded DramaBox clip
✦ TOP 1
reward 0.344 · emo-sim 0.351 · WER 0.019 · 27.44s
“So it's so big. I could barely hear myself, all those people just looking. But then the m”
✦ TOP 2
reward 0.045 · emo-sim 0.394 · WER 0.885 · 24.96s
“So it's so big I could barely hear myself all those people just looking. But then the mus”
✦ TOP 3
reward 0.024 · emo-sim 0.610 · WER 0.962 · 24.56s
“So it's so big I could barely hear myself all those people just looking. But then the mus”
#2 · “so very quiet, felt so full of sound before all the singing”
A feminine voice, the acoustic signal is dominated by. Timbre: perfect; highly pleasant; standard. Delivery: monologue, dramatic, casual. Emotion: contentment, emotional numbness, longing, pleasure ecstasy, disappointmen
● ORIGINAL · 4.99s
the embedded DramaBox clip
✦ TOP 1
reward 0.887 · emo-sim 0.887 · WER 0.000 · 6.48s
“So very quiet, felt so full of sound before all the singing.”
✦ TOP 2
reward 0.697 · emo-sim 0.697 · WER 0.000 · 6.72s
“So very quiet, felt so full of sound before all the singing.”
✦ TOP 3
reward 0.691 · emo-sim 0.691 · WER 0.000 · 6.16s
“So very quiet, felt so full of sound before all the singing.”
#3 · “so it's so big I could barely hear myself”
A feminine voice, voice at peak biological vigor. Timbre: the high frequencies are distinctly; standard human baseline; standard. Delivery: casual, conversational, dramatic. Emotion: interest, elation, pleasure ecstasy,
● ORIGINAL · 5.1s
the embedded DramaBox clip
✦ TOP 1
reward 0.368 · emo-sim 0.368 · WER 0.000 · 5.84s
“So it's so big, I could barely hear myself.”
✦ TOP 2
reward 0.000 · emo-sim 0.605 · WER 1.000 · 5.2s
“So it's so big I could barely hear myself So it's so big I could barely hear myself”
✦ TOP 3
reward 0.000 · emo-sim 0.314 · WER 1.000 · 5.68s
“So it's so big, I could barely hear myself. So it's so big, I could barely hear myself.”
#4 · “all those people just looking. But then the music started”
A feminine voice, the acoustic signature reflects a fully. Timbre: the high frequencies are distinctly; standard human baseline; standard. Delivery: casual, dramatic, playful. Emotion: astonishment surprise, disappointme
● ORIGINAL · 5.51s
the embedded DramaBox clip
✦ TOP 1
reward 0.000 · emo-sim 0.702 · WER 1.000 · 5.2s
“All those people just looking. But then the music started. All those people just looking”
✦ TOP 2
reward 0.000 · emo-sim 0.750 · WER 1.000 · 5.36s
“All those people just looking, but then the music started. All those people just looking,”
✦ TOP 3
reward 0.000 · emo-sim 0.633 · WER 1.000 · 6.16s
“All those people just looking, but then the music started. All those people just looking,”
#5 · “and I remembered the high notes it felt”
A feminine voice, clear. Timbre: the high frequencies are distinctly; standard human baseline; standard. Delivery: casual, dramatic, conversational. Emotion: relief, interest, pleasure ecstasy, longing, embarrassment. Ge
● ORIGINAL · 3.63s
the embedded DramaBox clip
✦ TOP 1
reward 0.744 · emo-sim 0.744 · WER 0.000 · 2.56s
“and I remembered the high notes it felt.”
✦ TOP 2
reward 0.659 · emo-sim 0.659 · WER 0.000 · 3.36s
“And I remembered the high notes it felt.”
✦ TOP 3
reward 0.633 · emo-sim 0.633 · WER 0.000 · 3.12s
“and I remembered the high notes it felt.”
#6 · “right. It w- wasn't scary anymore, good, like like I could reach all the way up”
A the acoustic profile features, clear. Timbre: the high frequencies are distinctly; highly functional but distinctly neutral-cool; standard. Delivery: dramatic, casual, cartoonish. Emotion: relief, hope enthusiasm optim
● ORIGINAL · 6.46s
the embedded DramaBox clip
✦ TOP 1
reward 0.496 · emo-sim 0.562 · WER 0.118 · 9.92s
“Right. It wasn't scary anymore. Good. Like I could reach all the way up.”
✦ TOP 2
reward 0.094 · emo-sim 0.797 · WER 0.882 · 7.12s
“Right. It wasn't scary anymore. Good. Like I could reach all the way up. Right. It wa”
✦ TOP 3
reward 0.043 · emo-sim 0.734 · WER 0.941 · 8.56s
“Right! It wasn't scary anymore! Good! Like, like I could reach all the way up! Right! It”
#7 · “and stay there. I really did it,”
A the acoustic profile features, clear. Timbre: the high frequencies are distinctly; standard human baseline; standard. Delivery: dramatic, casual, cartoonish. Emotion: relief, interest, hope enthusiasm optimism, elation
● ORIGINAL · 2.95s
the embedded DramaBox clip
✦ TOP 1
reward 0.781 · emo-sim 0.781 · WER 0.000 · 3.68s
“And stay there! I really did it!”
✦ TOP 2
reward 0.672 · emo-sim 0.672 · WER 0.000 · 3.28s
“And stay there. I really did it.”
✦ TOP 3
reward 0.000 · emo-sim 0.551 · WER 1.000 · 4.0s
“And stay there. I really did it and stay there. I really did it”
#8 · “so very quiet, felt so full of sound before all the singing”
A feminine voice, the acoustic signal is dominated by. Timbre: perfect; highly pleasant; standard. Delivery: monologue, dramatic, casual. Emotion: contentment, emotional numbness, longing, pleasure ecstasy, disappointmen
● ORIGINAL · 4.99s
the embedded DramaBox clip
✦ TOP 1
reward 0.797 · emo-sim 0.797 · WER 0.000 · 7.76s
“So very quiet felt so full of sound before all the singing.”
✦ TOP 2
reward 0.773 · emo-sim 0.773 · WER 0.000 · 6.08s
“So very quiet, felt so full of sound before all the singing.”
✦ TOP 3
reward 0.767 · emo-sim 0.767 · WER 0.000 · 6.64s
“So very quiet, felt so full of sound before all the singing.”
#10 · “good. Not here. I thought s- I I thought it was just the sun playing tricks. But the shimmer, i- i- a pond? It's it's c-”
A the acoustic profile features, the acoustic signal is dominated by. Timbre: treble frequencies are heavily emphasized; distinctly tinny and highly top-heavy; standard. Delivery: casual, dramatic, cartoonish. Emotion: f
● ORIGINAL · 38.2s
the embedded DramaBox clip
✦ TOP 1
reward 0.223 · emo-sim 0.670 · WER 0.667 · 12.32s
“Good not here. I thought I thought it was just the Sun playing tricks, but the shimmer Ap”
✦ TOP 2
reward 0.183 · emo-sim 0.894 · WER 0.795 · 38.4s
“Good. Not here. I thought it was just the sun playing tricks. But the shimmer... A pond? ”
✦ TOP 3
reward 0.126 · emo-sim 0.754 · WER 0.833 · 34.56s
“Good, not here. I thought it was just the sun playing tricks. But the shimmer, a pond, i”
#11 · “A cruel dream, I should I should just walk towards it, even if even if it's not what I think, it's just so far, so so fa”
A the audio delivers a, voice at peak biological vigor. Timbre: perfect; standard human baseline; standard. Delivery: casual, conversational, playful. Emotion: contemplation, sadness, fatigue exhaustion, interest, helple
● ORIGINAL · 21.7s
the embedded DramaBox clip
✦ TOP 1
reward 0.141 · emo-sim 0.410 · WER 0.655 · 6.88s
“A cool dream. I should just walk towards it. A cool dream. I should just walk towards i”
✦ TOP 2
reward 0.113 · emo-sim 0.658 · WER 0.828 · 18.0s
“A cruel dream. I should just walk towards it even if it's not what I think. It's just so”
✦ TOP 3
reward 0.000 · emo-sim 0.483 · WER 1.000 · 13.6s
“A cruel dream I should I should just walk towards it even if even if it's not what I think”
#12 · “good. Not here. I thought s- I I thought it was just the sun playing tricks. But”
A the audio delivers a, the acoustic signature reflects a fully. Timbre: perfect; standard human baseline; standard. Delivery: casual, dramatic, conversational. Emotion: astonishment surprise, confusion, interest, contem
● ORIGINAL · 5.44s
the embedded DramaBox clip
✦ TOP 1
reward 0.287 · emo-sim 0.812 · WER 0.647 · 7.52s
“Good. Not here. I thought it was just the sun playing tricks. But. Good. Not here. I thoug”
✦ TOP 2
reward 0.284 · emo-sim 0.806 · WER 0.647 · 7.84s
“Good. Not here. I thought it was just the sun playing tricks. But good. Not here. I t”
✦ TOP 3
reward 0.284 · emo-sim 0.806 · WER 0.647 · 8.88s
“Good. Not here. I thought it was just the sun playing tricks. But good. Not here. I tho”
#13 · “But the shimmer, i- i- a pond? It's it's c- could it be there? Eee! Bone? I can't”
A the acoustic profile features, clear. Timbre: treble frequencies are heavily emphasized; distinctly tinny and highly top-heavy; lmost completely obliterated by violent. Delivery: dramatic, casual, cartoonish. Emotion:
● ORIGINAL · 6.81s
the embedded DramaBox clip
✦ TOP 1
reward 0.087 · emo-sim 0.914 · WER 0.905 · 12.96s
“But the shimmer... a pond? It's... it's... could it be there? E! Bone! But the shimmer.”
✦ TOP 2
reward 0.084 · emo-sim 0.877 · WER 0.905 · 8.4s
“But the shimmer... A pond? It's...it's...could it be there? Eh...bone? I can't... But”
✦ TOP 3
reward 0.082 · emo-sim 0.862 · WER 0.905 · 8.4s
“But the shimmer upon it's it's could it be there? It bone I can't but the shimmer upon it”
#14 · “after all this, some someone has to be, I can't believe it!”
A the acoustic profile features, the acoustic signal is dominated by. Timbre: treble frequencies are heavily emphasized; distinctly tinny and highly top-heavy; lmost completely obliterated by violent. Delivery: dramatic,
● ORIGINAL · 5.09s
the embedded DramaBox clip
✦ TOP 1
reward 0.757 · emo-sim 0.820 · WER 0.077 · 6.56s
“Huh, after all this, some someone has to be, I can't believe it.”
✦ TOP 2
reward 0.623 · emo-sim 0.900 · WER 0.308 · 7.2s
“If the world is some, someone has to be. I can't believe it.”
✦ TOP 3
reward 0.000 · emo-sim 0.522 · WER 1.000 · 5.84s
“After all this, some someone has to be, I can't believe it! After all this, some someone ”
#15 · “Please, please let someone be there. Just just a little water please Oh s-”
A the acoustic profile features, clear. Timbre: treble frequencies are heavily emphasized; distinctly tinny and highly top-heavy; lmost completely obliterated by violent. Delivery: dramatic, casual, cartoonish. Emotion:
● ORIGINAL · 7.71s
the embedded DramaBox clip
✦ TOP 1
reward 0.119 · emo-sim 0.836 · WER 0.857 · 10.64s
“Please, please let someone be there. Just a little water, please. Oh, s-please, please le”
✦ TOP 2
reward 0.057 · emo-sim 0.792 · WER 0.929 · 10.8s
“please, please let someone be there, just a little water, please. oh sssssssssss please,”
✦ TOP 3
reward 0.000 · emo-sim 0.771 · WER 1.000 · 9.76s
“Please, please let someone be there. Just, just a little water, please. Oase, please, pl”
#16 · “this is this is insane I'm I'm actually crying because”
A the acoustic profile features, clear. Timbre: treble frequencies are heavily emphasized; distinctly tinny and highly top-heavy; lmost completely obliterated by violent. Delivery: dramatic, casual, ranting/angry. Emotio
● ORIGINAL · 4.76s
the embedded DramaBox clip
✦ TOP 1
reward 0.328 · emo-sim 0.985 · WER 0.667 · 5.92s
“This is, this is insane. I'm actually crying because... This is, this is insane. I'm actu”
✦ TOP 2
reward 0.311 · emo-sim 0.933 · WER 0.667 · 5.12s
“This is... This is insane! I'm actually crying because... This is... This is insane! I'”
✦ TOP 3
reward 0.227 · emo-sim 0.909 · WER 0.750 · 4.08s
“This is- this is insane! I'm- I'm- This is- this is insane! I'm- I'm-”
#17 · “because of water.”
A the audio delivers a, the acoustic signature reflects a fully. Timbre: perfect; standard human baseline; lmost completely obliterated by violent. Delivery: conversational, dramatic, casual. Emotion: distress, helplessn
● ORIGINAL · 8.39s
the embedded DramaBox clip
✦ TOP 1
reward 0.000 · emo-sim 0.189 · WER 1.000 · 1.52s
“Because of water. Because of water.”
✦ TOP 2
reward 0.000 · emo-sim 0.055 · WER 1.000 · 1.44s
“Because of water. Because of water.”
✦ TOP 3
reward 0.000 · emo-sim 0.014 · WER 1.000 · 2.8s
“because of water. Because of water.”
#18 · “A cruel dream, I should I should just walk towards it,”
A the audio delivers a, voice at peak biological vigor. Timbre: perfect; standard human baseline; standard. Delivery: casual, conversational, playful. Emotion: distress, fatigue exhaustion, interest, sadness, helplessnes
● ORIGINAL · 12.36s
the embedded DramaBox clip
✦ TOP 1
reward 0.337 · emo-sim 0.412 · WER 0.182 · 5.6s
“A cruel dream, I should just walk towards it.”
✦ TOP 2
reward 0.188 · emo-sim 0.516 · WER 0.636 · 5.36s
“A cruel dream. I should just walk towards it. A cruel dream. I should just walk towards”
✦ TOP 3
reward 0.171 · emo-sim 0.469 · WER 0.636 · 8.8s
“A cruel dream. I should just walk towards it. A cruel dream. I should just walk towards”
#19 · “even if even if it's not what I think,”
A the audio delivers a, the acoustic signature reflects a fully. Timbre: the acoustic signal is completely; highly pleasant; the acoustic signal is heavily. Delivery: whisper-talk, monologue, dramatic. Emotion: contempla
● ORIGINAL · 4.65s
the embedded DramaBox clip
✦ TOP 1
reward 0.903 · emo-sim 0.903 · WER 0.000 · 4.88s
“Even if, even if it's not what I think.”
✦ TOP 2
reward 0.624 · emo-sim 0.780 · WER 0.200 · 3.28s
“even if it's not what I think.”
✦ TOP 3
reward 0.379 · emo-sim 0.949 · WER 0.600 · 3.84s
“even if it's not what I think. Even if it's not what I think.”
#20 · “it's just so far, so so far.”
A the audio delivers a, the acoustic profile is heavily defined. Timbre: the acoustic signal is completely; highly pleasant; phonation is exceptionally consistent and. Delivery: whisper-talk, monologue, asmr. Emotion: fa
● ORIGINAL · 4.69s
the embedded DramaBox clip
✦ TOP 1
reward 0.000 · emo-sim 0.907 · WER 1.000 · 4.64s
“It's just so far so so far It's just so far so so far”
✦ TOP 2
reward 0.000 · emo-sim 0.801 · WER 1.000 · 3.6s
“It's just so far so so far. It's just so far so so far”
✦ TOP 3
reward 0.000 · emo-sim 0.900 · WER 1.000 · 5.28s
“It's just so far so so far It's just so far so so far”
#24 · “it all happened so fast well, everything, you know, one minute everything was a bunch of b- b- you know, b- like normal ”
A feminine voice, voice at peak biological vigor. Timbre: perfect; standard human baseline; standard. Delivery: casual, conversational, monologue. Emotion: interest, fatigue exhaustion, embarrassment, amusement, impatien
● ORIGINAL · 9.39s
the embedded DramaBox clip
✦ TOP 1
reward 0.079 · emo-sim 0.681 · WER 0.885 · 10.24s
“It all happened so fast. Well, everything, you know, one minute everything was a bunch of”
✦ TOP 2
reward 0.055 · emo-sim 0.478 · WER 0.885 · 12.48s
“It all happened so fast. Well, everything, you know, one minute, everything was a bunch o”
✦ TOP 3
reward 0.047 · emo-sim 0.613 · WER 0.923 · 8.8s
“It all happens so fast. Well, everything, you know, one minute, everything was a bunch of,”
#25 · “this. I keep replaying it in my head, trying to understand where it all went wrong, searching for some tiny sign I misse”
A feminine voice, voice at peak biological vigor. Timbre: perfect; standard human baseline; standard. Delivery: conversational, casual, monologue. Emotion: contemplation, concentration, interest, confusion, sadness. Genu
● ORIGINAL · 6.99s
the embedded DramaBox clip
✦ TOP 1
reward 0.847 · emo-sim 0.847 · WER 0.000 · 9.68s
“This. I keep replaying it in my head, trying to understand where it all went wrong, search”
✦ TOP 2
reward 0.000 · emo-sim 0.848 · WER 1.000 · 7.76s
“This. I keep replaying it in my head, trying to understand where it all went wrong, search”
✦ TOP 3
reward 0.000 · emo-sim 0.778 · WER 1.000 · 9.68s
“This, I keep replaying it in my head, trying to understand where it all went wrong, search”
#26 · “It's absurd really, like a bad dream, but”
A feminine voice, the acoustic signature reflects a fully. Timbre: perfect; standard human baseline; standard. Delivery: casual, storytelling, monologue. Emotion: disgust, interest, pleasure ecstasy, embarrassment, intox
● ORIGINAL · 2.78s
the embedded DramaBox clip
✦ TOP 1
reward 0.000 · emo-sim 0.576 · WER 1.000 · 4.24s
“It's absurd really, like a bad dream, but it's absurd really, like a bad dream, but...”
✦ TOP 2
reward 0.000 · emo-sim 0.776 · WER 1.000 · 4.8s
“It's absurd really, like a bad dream, but... It's absurd really, like a bad dream, but...”
✦ TOP 3
reward 0.000 · emo-sim 0.603 · WER 1.000 · 5.2s
“It's absurd really like a bad dream, but it's absurd really like a bad dream, but”
#27 · “it's not. It's just here, this the silence. The silence is the worst part,”
A feminine voice, voice at peak biological vigor. Timbre: perfect; highly functional but distinctly neutral-cool; standard. Delivery: monologue, dramatic, casual. Emotion: distress, sadness, helplessness, disappointment,
● ORIGINAL · 7.82s
the embedded DramaBox clip
✦ TOP 1
reward 0.202 · emo-sim 0.539 · WER 0.625 · 14.16s
“It's not. It's just here, this the silence. It's not. It's just here, this the silence.”
✦ TOP 2
reward 0.000 · emo-sim 0.874 · WER 1.000 · 11.04s
“It's not it's just here This the silence the silence is the worst part It's not It's j”
✦ TOP 3
reward 0.000 · emo-sim 0.564 · WER 1.000 · 9.28s
“It's not. It's just here. This, the silence. The silence is the worst part. It's not. ”
#28 · “it's deafening. We're watch the steady I must say Everyone keeps saying it will be okay that time to heal wounds tends i”
A feminine voice, voice at peak biological vigor. Timbre: perfect; standard human baseline; standard. Delivery: casual, conversational, whisper-talk. Emotion: fatigue exhaustion, interest, sadness, helplessness, embarras
● ORIGINAL · 13.59s
the embedded DramaBox clip
✦ TOP 1
reward 0.077 · emo-sim 0.364 · WER 0.787 · 21.04s
“It's deafening. Where watch the steady I must say. Everyone keeps saying it will be okay”
✦ TOP 2
reward 0.009 · emo-sim 0.407 · WER 0.979 · 17.92s
“It's deafening. We watch the steady I must say everyone keeps saying it will be okay that”
✦ TOP 3
reward 0.000 · emo-sim 0.344 · WER 1.021 · 17.44s
“It's deafening. We're watch the steady, I must say. Everyone keeps saying it will be oka”