moss-local voice-acting — best-of-64 by emotion (pilot)
What this is (for dummies). the same pilot we ran on MOSS-8B (delay architecture), repeated on
TTS-AGI/moss-local-transformer-voice-acting:
MOSS-TTS-Local (4.55B, 12-codebook RVQ @ 48 kHz, v2 tokenizer), full dramabox fine-tune + rank-256
voice-acting LoRA (checkpoint-1m, merged into the weights). identical 10 groups: 5 emotions × 2 prompts from
laion/voice-acting-prompts,
procedural instructions, 64 takes each, batched inference.
head-to-head: moss-local (48 kHz) vs MOSS-8B delay (24 kHz), same 10 groups
emotion
best target (local)
best target (8B)
genu (local)
genu (8B)
invWER (local)
invWER (8B)
arousal (local)
arousal (8B)
anger
3.33
3.40
0.45
0.60
1.00
0.81
2.37
2.14
fear
2.72
2.52
1.13
1.02
0.94
0.87
1.50
1.55
sadness
4.18
3.70
1.53
1.43
0.96
0.78
1.41
1.42
elation
3.28
3.49
0.86
0.71
0.99
0.81
2.43
2.00
disgust
2.94
3.12
0.58
0.61
0.97
0.88
1.77
1.84
top-3 pools — moss-local: target 2.30, genu 1.25, invWER 0.99 · MOSS-8B: target 2.33, genu 1.21, invWER 0.98. (rewards are normalised within each model run, so compare the raw scores, not the reward values.)
anger
vap_Anger_0burst: Teeth chattering (from cold or extreme fear)64/64 takes · best reward 1.81
SAY: Damn them all to the deepest circles of oblivion; their treachery will be repaid tenfold! This shivering is not fear, but the prelude to a storm of vengeance that will consume them all.
procedural instruction
extremely anger rage fury hate irascibility enragement vexation wrath peevishness annoyance. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: teeth chattering (from cold or extreme fear). loud, intense, high quality recording.
vap_Anger_1burst: Exasperated Sigh (audible breath of frustration)64/64 takes · best reward 1.77
SAY: To hell with their rules, their expectations, their suffocating little world! This burning rage will not be contained; it will be unleashed like a tempest upon the unworthy! They will learn the true meaning of fear when they face my wrath!
procedural instruction
extremely anger rage fury hate irascibility enragement vexation wrath peevishness annoyance. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: exasperated sigh (audible breath of frustration). loud, intense, high quality recording.
vap_Fear_0burst: Sharp, sudden inhale (a quick gasp of shock or horror)64/64 takes · best reward 2.01
SAY: Lest we forget this day, when shadows danced with demons, and the abyss opened its maw to swallow us whole, revealing a terror we can never truly escape!
procedural instruction
extremely fear terror dread apprehension alarm horror panic nervousness. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: sharp, sudden inhale (a quick gasp of shock or horror). loud, intense, high quality recording.
vap_Fear_1burst: Cracking knuckles (the sharp pop of joints, indicating readi64/64 takes · best reward 1.91
SAY: Alright, alright, easy now. Maybe I shouldn't have come here; this looks bad. I really hope I can get out of this mess.
procedural instruction
extremely fear terror dread apprehension alarm horror panic nervousness. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: cracking knuckles (the sharp pop of joints, indicating readiness or intimidation). loud, intense, high quality recording.
vap_Sadness_0burst: Phew! (breathy sound of relief)64/64 takes · best reward 2.04
SAY: Now that's over, I guess, huh? Feels kinda empty, not gonna lie. Whatever, gotta keep moving, right?
procedural instruction
extremely sadness sorrow grief melancholy dejection despair self pity sullenness heartache mournfulne. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: phew! (breathy sound of relief). loud, intense, high quality recording.
vap_Sadness_1burst: Oi! (A very informal, sharp call to get someone's attention,64/64 takes · best reward 2.60
SAY: Ah, destiny, you cruel mistress, why do you mock me so? Oi! Is this all I'm worth, a discarded pawn in your twisted game? My heart weeps, drowning in a sea of sorrow, forever lost to the echoes of what could have been.
procedural instruction
extremely sadness sorrow grief melancholy dejection despair self pity sullenness heartache mournfulne. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: oi! (a very informal, sharp call to get someone's attention, common in british english). loud, intense, high quality recording.
vap_Elation_0burst: Humming a cheerful, upbeat tune (to express happiness or lig64/64 takes · best reward 2.15
SAY: This is it; the moment I've dreamt of since I was a child, standing here on the precipice of destiny. The world unfolds before me, a tapestry woven with threads of infinite possibility, and I, I am the artist, ready to paint my masterpiece! I knew I
procedural instruction
extremely happiness excitement joy exhilaration delight jubilation bliss cheerfulness. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: humming a cheerful, upbeat tune (to express happiness or lightheartedness). loud, intense, high quality recording.
vap_Elation_1burst: Cracking knuckles (the sharp pop of joints, indicating readi64/64 takes · best reward 1.94
SAY: Having foreseen this glorious victory, my spirit now soars as high as the heavens, for destiny itself is about to be rewritten! Now, let the games begin!
procedural instruction
extremely happiness excitement joy exhilaration delight jubilation bliss cheerfulness. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: cracking knuckles (the sharp pop of joints, indicating readiness or intimidation). loud, intense, high quality recording.
vap_Disgust_0burst: Drinking Noises (audible liquid intake with gulps or slurps)64/64 takes · best reward 1.84
SAY: That swill... it assaults my very being! Never again shall my palate be subjected to such an unholy concoction, a transgression against taste itself!
procedural instruction
extremely disgust revulsion repulsion abhorrence loathing. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: drinking noises (audible liquid intake with gulps or slurps). loud, intense, high quality recording.
vap_Disgust_1burst: A sharp, attention-getting whistle64/64 takes · best reward 1.36
SAY: Right, the execrable miasma of malodorous prevarication, redolent of suppurating mendacity, has so permeated the very fabric of this odious charade that it engenders an abject abhorrence hitherto unknown in the annals of my profoundly disquieted sens
procedural instruction
extremely disgust revulsion repulsion abhorrence loathing. starts already emotional and escalates line by line until the feeling completely takes over. includes this vocal burst, performed fully: a sharp, attention-getting whistle. loud, intense, high quality recording.