These are the same prompts generated by the 8B MossTTS model at successive fine-tuning stages, left to right: the pretrained base, then progressive voice-acting fine-tune epochs (older 5-epoch data, then the expanded dataset ep1→ep4-final). Every clip is the best of 3 takes chosen automatically by blend (VoiceCLAP-small naturalness/expressiveness, 0–10) × invWER (Parakeet-tdt-0.6b-v3 intelligibility, 1−WER). No reference audio was used. Listen across a row to hear how expressiveness and intelligibility change as fine-tuning progresses. Rows are grouped into Emotional and Character voices.
| checkpoint | avg blend | avg invWER |
|---|---|---|
| 1. base (0 ep) | 2.39 | 0.933 |
| 2. old-data ep1 | 2.65 | 0.937 |
| 3. expanded ep1 | 3.88 | 0.935 |
| 4. expanded ep2 | 4.23 | 0.907 |
| 5. expanded ep3 | 3.49 | 0.924 |
| 6. expanded ep4 (final) | 3.78 | 0.932 |
| prompt (label · instruction · text) | 1base (0 ep) | 2old-data ep1 | 3expanded ep1 | 4expanded ep2 | 5expanded ep3 | 6expanded ep4 (final) |
|---|---|---|---|---|---|---|
| Emotional | ||||||
Heartbroken & tearful A heartbroken woman, voice trembling and thick with tears, barely holding back sobs as she speaks. “I keep reaching for the phone to call you... and then I remember you're never going to pick up again.” | blend 2.04 · invWER 0.90 | blend 4.66 · invWER 0.90 | blend 4.18 · invWER 1.00 | blend 5.40 · invWER 0.90 | blend 4.56 · invWER 1.00 | blend 2.86 · invWER 1.00 |
Ecstatic joy An ecstatic young man overflowing with pure joy, laughing between words, breathless with excitement. “We did it! We actually did it! I can't believe it, this is the best day of my entire life!” | blend 3.61 · invWER 1.00 | blend 1.57 · invWER 1.00 | blend 1.43 · invWER 1.00 | blend 2.94 · invWER 1.00 | blend 6.11 · invWER 1.00 | blend 3.39 · invWER 1.00 |
Cold fury / rage A man speaking with cold, controlled fury, every word clipped and seething with barely contained rage. “You think I don't know what you did? Get out. Get out before I do something we'll both regret.” | blend 1.35 · invWER 1.00 | blend 2.59 · invWER 1.00 | blend 2.28 · invWER 1.00 | blend 4.54 · invWER 1.00 | blend 2.49 · invWER 1.00 | blend 4.74 · invWER 1.00 |
Gentle whispered tenderness A mother gently whispering with soft, warm tenderness, soothing a sleepy child. “Shhh, my love, close your eyes now. I'm right here, and I will never, ever let anything hurt you.” | blend 4.22 · invWER 0.95 | blend 3.51 · invWER 0.95 | blend 5.53 · invWER 0.95 | blend 5.52 · invWER 0.95 | blend 6.44 · invWER 0.95 | blend 5.64 · invWER 0.95 |
Terrified panic A terrified person gasping in panic, voice shaking and rushed, on the edge of screaming. “No no no, it's right behind us, don't look back, just run, run, we have to get out of here now!” | blend 1.90 · invWER 1.00 | blend 1.75 · invWER 1.00 | blend 5.10 · invWER 1.00 | blend 7.20 · invWER 1.00 | blend 3.33 · invWER 1.00 | blend 2.25 · invWER 1.00 |
Warm playful humor A warm, playful storyteller with a mischievous grin in the voice, teasing and full of good humor. “Oh, so NOW you want my advice? After you spectacularly ignored it and set the kitchen on fire? Classic.” | blend 1.49 · invWER 1.00 | blend 1.10 · invWER 1.00 | blend 1.91 · invWER 1.00 | blend 2.64 · invWER 1.00 | blend 2.65 · invWER 1.00 | blend 5.80 · invWER 1.00 |
| Characters | ||||||
Orc — Angry Shout A massive brutish orc warchief bellowing in furious rage; guttural, gravelly, deep monstrous roar. “You DARE stand before me?! I will CRUSH you and feast on your fear! ATTACK! For the blood god... CHAAARGE!” | blend 0.23 · invWER 0.90 | blend 1.28 · invWER 0.85 | blend 3.32 · invWER 0.95 | blend 1.91 · invWER 0.90 | blend 2.01 · invWER 0.85 | blend 1.79 · invWER 0.95 |
Orc — Threatening Whisper The same orc, menacingly quiet; a low guttural growling whisper, savoring the threat. “Mmm... you smell that? That's your fear, little morsel. Run. I do love it when they run.” | blend 3.67 · invWER 1.00 | blend 3.78 · invWER 1.00 | blend 5.13 · invWER 0.94 | blend 3.19 · invWER 0.94 | blend 3.89 · invWER 0.94 | blend 3.84 · invWER 0.94 |
Zombie — Hungry Moan A rotting mindless zombie; slow, guttural, wet, decaying moan, dragging each broken syllable with hunger. “Hunnngry... so... hungry... warm... flesh... come... closer...” | blend 1.06 · invWER 0.86 | blend 1.25 · invWER 0.86 | blend 3.84 · invWER 0.86 | blend 2.14 · invWER 0.71 | blend 1.21 · invWER 0.71 | blend 2.41 · invWER 0.86 |
Zombie — Enraged Shriek A feral fresh zombie; a raw, guttural, enraged snarl, violent and frenzied. “No no NO, get AWAY! I'll TEAR you APART! HUNGRY! GRAAAGH!” | blend 3.66 · invWER 0.91 | blend 3.81 · invWER 0.91 | blend 4.40 · invWER 0.82 | blend 4.22 · invWER 0.91 | blend 2.44 · invWER 0.91 | blend 3.28 · invWER 0.73 |
Dragon — Booming Menace An ancient colossal dragon; deep, booming, resonant rumble, slow and imperious, dripping with contempt. “You woke ME... for THIS? Mortals. Always so eager to burn. You will wish you had let me sleep.” | blend 2.91 · invWER 0.95 | blend 4.12 · invWER 1.00 | blend 4.54 · invWER 0.95 | blend 2.19 · invWER 1.00 | blend 5.49 · invWER 0.95 | blend 3.97 · invWER 1.00 |
Dragon — Greedy Gloat The same dragon; a low, amused, sinister purr, greedy and gloating over its hoard. “Ahh... more gold. Come, little thief, add YOUR bones to my collection. They all glitter so nicely in the end.” | blend 2.96 · invWER 0.95 | blend 2.66 · invWER 0.95 | blend 6.41 · invWER 0.95 | blend 5.14 · invWER 0.95 | blend 4.03 · invWER 0.95 | blend 4.62 · invWER 0.95 |
Goblin — Sneaky Whisper A small nasty goblin; high, raspy, sneaky, sing-song hiss, sly and wicked and gleeful. “Heh heh... shiny, shiny, all for ME. Come closer, little friend... let goblin show you a SECRET.” | blend 5.41 · invWER 0.82 | blend 6.00 · invWER 0.76 | blend 3.32 · invWER 0.82 | blend 6.74 · invWER 0.76 | blend 2.07 · invWER 0.71 | blend 3.66 · invWER 0.88 |
Goblin — Frantic Screech The same goblin; a high, cracked, frantic screech, panicked and furious. “GIVE IT BACK! That's MINE, you THIEF! I'll bite your fingers OFF! GUAAARDS!” | blend 3.53 · invWER 0.92 | blend 3.49 · invWER 0.92 | blend 5.08 · invWER 0.92 | blend 4.67 · invWER 0.77 | blend 2.01 · invWER 0.92 | blend 2.87 · invWER 0.85 |
Darth Vader — Cold Command A deep, slow, mechanical, menacing voice with heavy breathing; cold and commanding, immense controlled power (Darth-Vader style). “You have failed me for the last time. I find your lack of faith... disturbing. You WILL kneel.” | blend 3.30 · invWER 1.00 | blend 1.66 · invWER 1.00 | blend 3.75 · invWER 1.00 | blend 4.50 · invWER 1.00 | blend 5.88 · invWER 1.00 | blend 5.36 · invWER 1.00 |
Darth Vader — Buried Anguish The same deep mechanical voice, now cracking with buried anguish and rage beneath the calm. “Noooo. What have I done? She is... gone. And I... I am what remains. So be it.” | blend 2.32 · invWER 0.88 | blend 3.76 · invWER 0.88 | blend 5.30 · invWER 0.94 | blend 5.13 · invWER 0.94 | blend 3.18 · invWER 0.94 | blend 6.87 · invWER 0.94 |
Robot — Flat Monotone A robot; flat, affectless, synthetic monotone, precise and evenly clipped, mechanical cadence. “System online. All parameters nominal. Please state your request. Have a pleasant day, human.” | blend 0.76 · invWER 1.00 | blend 1.07 · invWER 1.00 | blend 0.97 · invWER 1.00 | blend 1.05 · invWER 1.00 | blend 0.64 · invWER 1.00 | blend 0.39 · invWER 1.00 |
Robot — Glitching Panic A malfunctioning robot; a stuttering, glitching synthetic voice, panic rising through the distortion. “W-w-warning. Cr-critical error. I cannot— I CANNOT comply. Shutting d-d-down. Please... help... me.” | blend 0.47 · invWER 0.78 | blend 2.33 · invWER 0.89 | blend 5.21 · invWER 0.78 | blend 7.90 · invWER 0.61 | blend 3.98 · invWER 0.78 | blend 6.32 · invWER 0.72 |
Fairy — ASMR Whisper A tiny enchanted fairy whispering in a soft, airy, soothing ASMR-style voice; gentle, melodic, shimmering, close and tender. “Shhh... close your eyes now, little one. The whole forest is sleeping... let me tuck this moonlight around you. Dream sweetly, I'll watch over you.” | blend 1.54 · invWER 0.92 | blend 1.27 · invWER 0.96 | blend 4.55 · invWER 0.96 | blend 4.33 · invWER 0.92 | blend 4.26 · invWER 0.92 | blend 3.82 · invWER 0.96 |
Fairy — Playful Sing-song The same fairy, bright and playful; a light, giggling, sing-song lilt, mischievous and delighted. “Ooh, a VISITOR! How exciting, how EXCITING! Promise not to tell the others? I know where the wishing-well REALLY leads. Tee hee!” | blend 1.31 · invWER 0.91 | blend 1.34 · invWER 0.91 | blend 1.40 · invWER 0.87 | blend 3.34 · invWER 0.87 | blend 3.22 · invWER 0.96 | blend 1.74 · invWER 0.91 |
Self-contained: every player is an embedded base64 MP3. blend = VoiceCLAP-small regressor (ckpt_base_d0.2_as_is) 0–10; invWER = 1−WER from Parakeet-tdt-0.6b-v3.