20 scenes · 3 seeds each · 60 clips · a voice-acting TTS model at the fixed settings below · no voice reference, the voice comes from the scene text.
The ear is the arbiter, not the badge. The badge on each card is what
reward.RewardModel(...).bursts() — our own burst locator plus 83-class detector —
says about the clip. That detector is a weak instrument: on 264 clean source clips it names
the source dataset's own label 3.4 % of the time, it emits only 44 of its 83 labels, and three
labels account for 58 % of everything it ever says. So a card marked MISS
that plainly contains a scream when you play it is a fact about the detector, not about the generator.
Play the clips; the numbers are the secondary reading.
Strict vs family. Strict = the detector emitted the exact class the scene asked
for. Family = it emitted anything in the same burst family, which here is
{Scream, Shriek, Mournful Wail}. Note that scream and
shriek sit in the same family, so the family reading cannot tell the two
classes apart — it only says "something in the screaming family came out". Strict is the headline
for that reason.
| brief / Space slider | value | TTSServer.generate kwarg | note |
|---|---|---|---|
| CFG scale | 2.5 | cfg_scale | classifier-free guidance |
| STG scale | 1.5 | stg_scale | skip-token guidance, block 29 |
| Breathing factor | 1.10 | duration_multiplier | headroom on the estimated length |
| Fixed duration (s) | 0.0 | gen_duration | 0 = automatic, estimator decides |
| Reference window (s) | 10.0 | ref_duration | inert here: no voice_ref is passed, so the voice comes from the scene's own speaker description |
Watermarking off, rescale_scale="auto" (the checkpoint default), bf16,
30 denoising steps, output 48 kHz written to OGG/Vorbis. Seeds 1234, 5678, 9012,
the same three for every scene so the per-seed column means something.
| class | clips | strict hits | strict rate | family hits | family rate |
|---|---|---|---|---|---|
| shriek | 30 | 0 | 0.0 % | 6 | 20.0 % |
| scream | 30 | 3 | 10.0 % | 3 | 10.0 % |
| all | 60 | 3 | 5.0 % | 9 | 15.0 % |
| seed | class | clips | strict | family |
|---|---|---|---|---|
| 1234 | shriek | 10 | 0 / 0.0 % | 4 / 40.0 % |
| 1234 | scream | 10 | 0 / 0.0 % | 0 / 0.0 % |
| 5678 | shriek | 10 | 0 / 0.0 % | 1 / 10.0 % |
| 5678 | scream | 10 | 0 / 0.0 % | 0 / 0.0 % |
| 9012 | shriek | 10 | 0 / 0.0 % | 1 / 10.0 % |
| 9012 | scream | 10 | 3 / 30.0 % | 3 / 30.0 % |
| voice asked for | class | clips | strict | family |
|---|---|---|---|---|
| male | shriek | 15 | 0 / 0.0 % | 5 / 33.3 % |
| male | scream | 15 | 1 / 6.7 % | 1 / 6.7 % |
| female | shriek | 15 | 0 / 0.0 % | 1 / 6.7 % |
| female | scream | 15 | 2 / 13.3 % | 2 / 13.3 % |
| scene | class | voice | situation | strict / 3 seeds | family / 3 seeds |
|---|---|---|---|---|---|
| shriek01 | shriek | male | ice water over the head | 0 / 3 | 1 / 3 |
| shriek02 | shriek | female | bucket of ice water | 0 / 3 | 1 / 3 |
| shriek03 | shriek | male | something runs over his bare foot in the dark | 0 / 3 | 1 / 3 |
| shriek04 | shriek | female | spider drops onto her hand | 0 / 3 | 0 / 3 |
| shriek05 | shriek | male | grabs a red-hot pan handle | 0 / 3 | 2 / 3 |
| shriek06 | shriek | female | metal crash behind her in a parking garage | 0 / 3 | 0 / 3 |
| shriek07 | shriek | male | needle jab at the doctor | 0 / 3 | 0 / 3 |
| shriek08 | shriek | female | haunted-house jump scare | 0 / 3 | 0 / 3 |
| shriek09 | shriek | male | slams his finger in a car door | 0 / 3 | 1 / 3 |
| shriek10 | shriek | female | rat in the kitchen cupboard | 0 / 3 | 0 / 3 |
| scream01 | scream | male | the door bursts open | 0 / 3 | 0 / 3 |
| scream02 | scream | female | high-pitched scream of panic | 0 / 3 | 0 / 3 |
| scream03 | scream | male | the ledge gives way under him | 0 / 3 | 0 / 3 |
| scream04 | scream | female | finds a body | 1 / 3 | 1 / 3 |
| scream05 | scream | male | rage, throat-tearing | 0 / 3 | 0 / 3 |
| scream06 | scream | female | pinned in a crashed car | 0 / 3 | 0 / 3 |
| scream07 | scream | male | watching someone he loves dragged away | 0 / 3 | 0 / 3 |
| scream08 | scream | female | a shape in the fog | 0 / 3 | 0 / 3 |
| scream09 | scream | male | broken leg, prolonged agony | 1 / 3 | 1 / 3 |
| scream10 | scream | female | house fire, screaming for her child | 1 / 3 | 1 / 3 |
Rows highlighted green are in the target family. Most common overall: Surprised Gasp ×26, Contented Sigh ×25, Scream ×10.
| detector label | events (all 60 clips) | share | events on strict-miss clips | family |
|---|---|---|---|---|
| Surprised Gasp | 26 | 35.6 % | 26 | breath |
| Contented Sigh | 25 | 34.2 % | 25 | sigh |
| Scream | 10 | 13.7 % | 7 | scream |
| Ahem | 4 | 5.5 % | 4 | throat |
| Childlike Giggle | 4 | 5.5 % | 4 | laugh |
| Exasperated Sigh | 2 | 2.7 % | 2 | sigh |
| Breathy Giggle | 1 | 1.4 % | 1 | laugh |
| Wistful Sigh | 1 | 1.4 % | 1 | sigh |
73 burst events located across the 60 clips (1.2 per clip); 8 distinct labels of the detector's 83.
Orange text inside a prompt is what the model is asked to say; grey italic is direction it interprets but never speaks. An empty filter falls back to the full group.
Ten scenes, five male and five female, three seeds each. Scenes shriek01 and shriek02 are the user-supplied examples, used verbatim.
Ten scenes, five male and five female, three seeds each. Scenes scream01 and scream02 are the user-supplied examples, used verbatim.