Procedural voice captions — current defaults

14 clips re-annotated with vocalburst-locator model_v2.pt + vocal-burst-detector-v2 · 16 bursts written into the scripts · ← caption grid · ← previous burst page (v1)

How to read a caption. Every clip gets one GENERAL line describing how the voice sounds overall, then a SCRIPT with one line per sentence.

(round brackets)a delivery cue at the head of a sentence, or a (vocal burst) written inline at the moment the locator found it. [square brackets]a silence, [pause X.Xs], taken from the Parakeet-v3 word timestamps (any gap ≥ 0.30 s).

The two burst models are different stages. The locator (laion/vocalburst-locator, model_v2.pt) says where a burst is: a 50 fps burst probability per frame over a fixed 30 s window. Longer audio is scanned in overlapping 30 s windows whose frame probabilities are stitched onto one timeline before events are extracted, so a burst on a seam is neither lost nor counted twice. The classifier (laion/vocal-burst-detector-v2) says what it is: it names the cut-out span with one of 82 taxonomy classes, or vetoes it with no_burst.

Post-processing here is threshold 0.5 · merge_gap 0.1 s · min_duration 0.1 s — the operating point measured for model_v2.pt on 992 held-out real clips (event F1 0.607), not the v1 defaults of 0.65 / 0.30 / 0.50. Spans whose P(no_burst) ≥ 0.5 are dropped. Where the same clip was annotated by the old pipeline, expand previous output under a card to compare.

character pain_scream5 9.0s · 1 locator window · 2 burst(s) kept of 2 located · genu 1.94 · blend 3.28
GENERAL
very vulnerable, very jerky, very scattered, very tense, very harsh, young, low-register, trepidation, submission, physical pain, semi-genuine
SCRIPT
(very vulnerable, very scattered, very raspy, very concern, very submission, very physical pain, genuine, with-bursts) (Contented Sigh) No, [pause 0.9s] no, [pause 0.3s] oh, it hurts.
(very breathy, very vulnerable, very scattered, very weariness, no curiosity, very helplessness, genuine, with-bursts) [pause 0.3s] Ah, (Contented Sigh) Eng, [pause 0.6s] get it off me.
Located spans
1.00–1.38s → Contented Sigh (p=0.62) 5.96–6.30s → Contented Sigh (p=0.43)
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (5 burst(s) kept of 10 spans)
(Surprised Gasp) No, no, (Surprised Gasp) oh, it hurts. (Surprised Gasp) Ah, (Slap Face) Hang, (Slap Face) get it off me.
character zombie5 12.4s · 1 locator window · 1 burst(s) kept of 1 located · genu 2.21 · blend 3.67
GENERAL
very vulnerable, very meek, very scattered, very soft-onset, very flat, young, feminine, low-register, very slow, lethargy, feeling horny, warmth, semi-genuine
SCRIPT
(very soft-onset, very scattered, very flat, lethargy, submission, insensitivity, semi-genuine, with-bursts) So hungry.
(very flat, very scattered, very smooth, burnout, no fascination, sullenness, semi-genuine, with-bursts) [pause 0.5s] Hmm. (Wistful Sigh)
(very scattered, very flat, very mumbled, desperation, pessimism, carnal desire, semi-genuine, with-bursts) [pause 1.3s] Help me.
(very scattered, very soft-onset, very flat, very burnout, very desperation, very unease, genuine, with-bursts) [pause 0.3s] I can't stop.
Located spans
3.32–3.60s → Wistful Sigh (p=0.24)
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (6 burst(s) kept of 11 spans)
So hungry. Hmm. (Relief Sigh) Help (Lip Smack) me. (Relief Sigh) I can't (Slap Face) (Lip Smack) stop. (Chuckle)
character goblin5 8.3s · 1 locator window · 1 burst(s) kept of 1 located · genu 1.74 · blend 6.53
GENERAL
very warm, very soft-onset, very flat, very mumbled, very organic, middle-aged, low-register, very slow, pining, helplessness, unease, semi-genuine, with-bursts
SCRIPT
(very flat, very smooth, very non-narrative, no interest, very fatigue, no deep focus, genuine) Huh. (Chuckle)
(very flat, very warm, very smooth, helplessness, concern, no interest, semi-genuine, with-bursts) [pause 0.4s] What have we here?
(very whispered, very flat, very warm, feeling turned on, nostalgia, fondness, semi-genuine, with-bursts) [pause 0.5s] A lost little traveller with shiny pockets?
Located spans
0.22–0.70s → Chuckle (p=0.58)
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (5 burst(s) kept of 7 spans)
(Clears Throat) Huh. (Deep Breath) What have (Purr) we here? (Slap Face) A lost little traveller with (Lip Smack) shiny pockets?
character mouse5 11.8s · 1 locator window · 0 burst(s) kept of 0 located · genu 0.97 · blend 2.9
GENERAL
very raspy, very scattered, very jerky, very monologic, very dramatic, middle-aged, low-register, very slow, performed
SCRIPT
(very raspy, very scattered, very dramatic, wonder, surprise, excitement, semi-genuine) Eee wait, wait, ooh, did you see it?
(very smooth, very raspy, very swelling, awe, amazement, no deep focus, semi-genuine, with-bursts) Did you see how teeny tiny it was?
Located spans
no span above threshold
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (6 burst(s) kept of 10 spans)
Eee wait wait ooh did you see it? Did you see how teeny tiny it was? (Slap Face) (Lip Smack) (Lip Smack) (Lip Smack) (Lip Smack) (Slap Face)
character sad_woman5 11.8s · 1 locator window · 1 burst(s) kept of 1 located · genu 1.35 · blend 1.56
GENERAL
very negative, very scattered, very vulnerable, very darkening, very meek, young, feminine, mid-register, very slow, helplessness, unease, contrition, semi-genuine, burst-free
SCRIPT
(very negative, very smooth, very meek, anxiety, helplessness, contrition, semi-genuine) No [pause 0.7s] no please [pause 0.3s] Don't leave me to (Contented Sigh) I can't [pause 0.5s] I can't lose anyone else
Located spans
5.10–5.36s → Contented Sigh (p=0.41)
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (2 burst(s) kept of 3 spans)
No no please Don't leave me to (Deep Breath) I can't I can't lose anyone else (Surprised Gasp)
character asmr_woman5 11.8s · 1 locator window · 1 burst(s) kept of 1 located · genu 1.65 · blend 3.1
GENERAL
very whispered, very monologic, very soft-onset, very noisy, very monotone, middle-aged, masculine, low-register, very slow, compassion, foreboding, lethargy, semi-genuine
SCRIPT
(very whispered, very warm, very smooth, warmth, butterflies in the stomach, lust, semi-genuine, with-bursts) (Deep Breath) There now, my love, close your eyes.
(very whispered, very flat, very smooth, devotion, carnal desire, fondness, semi-genuine, with-bursts) [pause 0.3s] I have you, safe and warm.
Located spans
0.10–0.98s → Deep Breath (p=0.17)
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (2 burst(s) kept of 3 spans)
(Hiss) There now, my love, close your eyes. I have you, safe and warm. (Breathy Giggle)
character ranting5 11.9s · 1 locator window · 0 burst(s) kept of 0 located · genu 1.52 · blend 9.62
GENERAL
very denasal, very negative, very thin-head, very jerky, very scattered, young, masculine, low-register, slightly slow, trepidation, powerlessness, agony, semi-genuine, with-bursts
SCRIPT
(very scattered, very breathy, very vulnerable, very anguish, very helplessness, very having a crush, genuine, with-bursts) Enough I am sick to death of this.
(very whispered, very denasal, very thin-head, spite, anxiety, no interest, semi-genuine, with-bursts) You pathetic, spineless fool.
(very raspy, very whispered, very negative, foreboding, adoration, feeling horny, semi-genuine, with-bursts) How dare you how dare you waste my time again and again?
Located spans
no span above threshold
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (2 burst(s) kept of 3 spans)
Enough. (Breathy Giggle) I am sick to death of this. You pathetic, (Slap Face) spineless fool. How dare you how dare you waste my time again and again?
character dragon5 7.1s · 1 locator window · 0 burst(s) kept of 0 located · genu 0.97 · blend 4.16
GENERAL
very raspy, very throaty, very full, very formal, very chesty, middle-aged, masculine, low-register, very slow, performed
SCRIPT
(very raspy, very monologic, very whispered, carnal desire, sadism, infatuation, semi-genuine) You dare enter my lair, mortal.
(very monologic, very formal, very metallic, performed, burst-free) Your bones will join the others.
Located spans
no span above threshold
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (3 burst(s) kept of 6 spans)
You dare enter my lair, mortal. (Slap Face) Your bones will (Displeased Grunt) join the others. (Wolf Whistle)
character evil_ghost5 7.1s · 1 locator window · 1 burst(s) kept of 1 located · genu 1.3 · blend 6.75
GENERAL
very throaty, very raspy, very chesty, very full, very formal, elderly, low-register, very slow, foreboding, malice, submission, semi-genuine, with-bursts
SCRIPT
(very throaty, very chesty, very smooth, sadism, anguish, lust, semi-genuine, with-bursts) Ugh (Exhausted Groan) you should not have come here.
(very throaty, very chesty, very formal, performed, with-bursts) [pause 0.6s] This house feeds on the living.
Located spans
0.22–1.20s → Exhausted Groan (p=0.58)
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (1 burst(s) kept of 1 spans)
(Exasperated Sigh) You should not have come here. This house feeds on the living.
in-the-wild ZH_B00039_S04341_W000000 3.8s · 1 locator window · 0 burst(s) kept of 0 located · genu 2.22 · blend 9.1
GENERAL
very whispered, very soft-onset, very thin-head, very flat, very warm, elderly, masculine, low-register, very slow, weariness, carnal desire, wistfulness, semi-genuine, with-bursts
SCRIPT
(very flat, very smooth, very scattered, romantic desire, emotional blunting, feeling horny, semi-genuine, with-bursts) Tiva.
(very steady, very whispered, very flat, insensitivity, homesickness, exhaustion, semi-genuine, with-bursts) [pause 1.2s] Tia Baya Gokufu Tsiwa.
Located spans
no span above threshold
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (3 burst(s) kept of 5 spans)
Tiva. (Slap Face) (Deep Breath) Tia Baya Gokufu Tsiwa. (Clears Throat)
in-the-wild FR_B00001_S08689_W000031 18.2s · 1 locator window · 0 burst(s) kept of 0 located · genu 0.92 · blend 8.05
GENERAL
very whispered, very formal, very narrator, very steady, very smooth, middle-aged, masculine, low-register, slow, performed, with-bursts
SCRIPT
(very smooth, very crisp, very hi-fi, no fascination, no absorption, detachment, semi-genuine) Soudain, on frappe à sa porte.
(very smooth, very thin-head, very formal, detachment, spite, no curiosity, semi-genuine, with-bursts) [pause 0.3s] Chloé sursaute.
(very smooth, very chesty, very formal, boredom, desperation, no fascination, semi-genuine, with-bursts) [pause 0.6s] Elle se sent incapable de bouger.
(very smooth, very formal, very chesty, no fascination, no engrossment, no contemplation, semi-genuine, with-bursts) [pause 0.6s] On frappe à nouveau.
(very whispered, very smooth, very noisy, performed, with-bursts) [pause 0.3s] Chloé retient sa respiration.
(very smooth, very monologic, very formal, performed, with-bursts) [pause 0.6s] Elle a peur.
(very smooth, very formal, very commanding, performed) Elle a l'impression de vivre un mauvais rêve.
Located spans
no span above threshold
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (0 burst(s) kept of 0 spans)
Soudain, on frappe à sa porte. Chloé sursaute. Elle se sent incapable de bouger. On frappe à nouveau. Chloé retient sa respiration. Elle a peur. Elle a l'impression de vivre un mauvais rêve.
in-the-wild EN_B00000_S03298_W000123 13.4s · 1 locator window · 1 burst(s) kept of 1 located · genu 5.13 · blend 5.22
GENERAL
very oral-bright, very chesty, very full, very naturalistic, very throaty, adult, masculine, low-register, very existential void, very meditation, no ribbing, genuine, with-bursts
SCRIPT
(very oral-bright, very dialogic, very throaty, no interest, no brooding, disdain, genuine, burst-free) So that people go, What is that? (Breathy Giggle)
(very oral-bright, very throaty, very commanding, no fascination, no thoughtfulness, no relaxation, genuine) I need to find out about that.
(very oral-bright, very throaty, very chesty, very existential void, very pensiveness, no fascination, genuine) [pause 0.7s] So this is me making stuff up right in the moment, but I'm like, okay, I'm gonna use these oblique strategies as a way of
Located spans
1.50–1.84s → Breathy Giggle (p=0.35)
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (0 burst(s) kept of 0 spans)
So that people go, What is that? I need to find out about that. So this is me making stuff up right in the moment, but I'm like, okay, I'm gonna use these oblique strategies as a way of
in-the-wild EN_B00026_S00187_W000083 3.2s · 1 locator window · 0 burst(s) kept of 0 located · genu 0.65 · blend 7.38
GENERAL
very whispered, very smooth, very formal, very steady, very pleasing, adult, feminine, mid-register, slow, performed, with-bursts
SCRIPT
(very whispered, very smooth, very soft-onset, performed, with-bursts) The woman coos and sings a sweet song
Located spans
no span above threshold
previous output — locator v1 + multi-label classifier, defaults 0.75 / 0.30 / 0.30 (0 burst(s) kept of 0 spans)
The woman coos and sings a sweet song
long-form longform_concat_91s 91.4s · 4 locator windows · 8 burst(s) kept of 8 located · genu 1.43 · blend 1.9
Constructed by concatenating the nine character clips above — the only way to show the >30 s path with audio that ships in this repo.
GENERAL
very scattered, very jerky, very vulnerable, very meek, very flat, young, androgynous, low-register, very slow, submission, sexual lust, pessimism, semi-genuine, burst-free
SCRIPT
(very soft-onset, very flat, very scattered, lethargy, helplessness, detachment, semi-genuine, with-bursts) So hungry.
(very flat, very scattered, very smooth, weariness, no intrigue, mournfulness, semi-genuine, with-bursts) [pause 0.5s] Hmm. (Wistful Sigh)
(very scattered, very flat, very non-narrative, submission, unease, feeling horny, semi-genuine, with-bursts) [pause 1.4s] Help me.
(very scattered, very soft-onset, very flat, very burnout, very desperation, very pessimism, genuine, with-bursts) [pause 0.3s] I can't stop.
(very scattered, very breathy, very jerky, no fascination, schadenfreude, powerlessness, semi-genuine, with-bursts) [pause 3.9s] Eee, (Surprised Gasp) wait!
(very sharp-onset, very scattered, very dialogic, performed) [pause 0.3s] Wait!
(very scattered, very thin, very smooth, performed) [pause 0.3s] Ooh!
(very scattered, very breathy, very dialogic, awestruck, sadism, astonishment, semi-genuine, with-bursts) Did you see it?
(very smooth, very swelling, very dramatic, awestruck, amazement, no attention, semi-genuine, with-bursts) Did you see how teeny tiny it was.
(very flat, very smooth, very non-narrative, no interest, very weariness, no concentration, genuine, with-bursts) [pause 3.2s] Huh. (Chuckle)
(very flat, very warm, very smooth, powerlessness, anxiety, no curiosity, semi-genuine, with-bursts) [pause 0.5s] What have we here?
(very whispered, very flat, very warm, feeling turned on, yearning, infatuation, semi-genuine, with-bursts) [pause 0.5s] A lost little traveler with shiny pockets.
(very flat, very scattered, very non-narrative, no fascination, very having a crush, very sullenness, genuine, with-bursts) [pause 2.5s] Enough.
(very breathy, very whispered, very denasal, submission, worry, suffering, semi-genuine, with-bursts) [pause 1.1s] I am sick to death of this.
(very whispered, very denasal, very thin-head, desire to harm, worry, no intrigue, semi-genuine, with-bursts) [pause 0.3s] You pathetic, spineless fool.
(very smooth, very thin-head, very monologic, performed, with-bursts) How dare you?
(very raspy, very whispered, very smooth, butterflies in the stomach, yearning, lust, semi-genuine, with-bursts) How dare you waste my time again and again.
(very smooth, very flat, very scattered, apathy, no intrigue, despair, semi-genuine) [pause 0.3s] No.
(very smooth, very smooth, very flat, desperation, worry, no curiosity, semi-genuine, with-bursts) [pause 0.6s] No, please.
(very smooth, very smooth, very monologic, performed, with-bursts) [pause 0.3s] Don't leave me to.
(very scattered, very breathy, child, desperation, anguish, agony, semi-genuine, with-bursts) [pause 1.2s] (Contented Sigh) I can't.
(very smooth, very thin, very jerky, adoration, helplessness, trust, semi-genuine, with-bursts) [pause 0.5s] I can't lose anyone else.
(very flat, very steady, very dissociated, sympathy, fondness, desire, semi-genuine, with-bursts) [pause 5.0s] There now, my love.
(very whispered, very smooth, very steady, performed, with-bursts) Close your eyes.
(very whispered, very flat, very smooth, forgiveness, sexual lust, infatuation, semi-genuine, with-bursts) [pause 0.3s] I have you safe and warm.
(very raspy, very whispered, very denasal, romantic desire, carnal desire, sadism, semi-genuine, with-bursts) [pause 2.1s] You dare enter my lair, mortal.
(very monologic, very formal, very smooth, performed, burst-free) Your bones will join the others. (Contented Sigh)
(very scattered, child, very jerky, no curiosity, dejection, indifference, semi-genuine, with-bursts) [pause 2.9s] (Contented Sigh) No.
(very sharp-onset, very dialogic, very scattered, performed, with-bursts) [pause 0.9s] No.
(very scattered, very breathy, very dialogic, very physical pain, very helplessness, very anxiety, genuine, with-bursts) [pause 0.5s] Oh, it hurts. (Contented Sigh)
(very scattered, very breathy, very vulnerable, helplessness, having a crush, ache, semi-genuine, with-bursts) [pause 1.1s] Hang, get it off me.
(very smooth, very chesty, very formal, schadenfreude, no interest, fury, semi-genuine, with-bursts) [pause 1.3s] (Exhausted Groan) You should not have come here.
(very throaty, very chesty, very formal, performed, with-bursts) [pause 0.7s] This house feeds on the living.
Located spans
3.32–3.60s → Wistful Sigh (p=0.24) 13.08–13.52s → Surprised Gasp (p=0.26) 24.64–24.90s → Chuckle (p=0.29) 49.58–49.78s → Contented Sigh (p=0.32) 74.88–75.06s → Contented Sigh (p=0.13) 76.28–76.64s → Contented Sigh (p=0.57) 81.24–81.58s → Contented Sigh (p=0.43) 84.54–85.50s → Exhausted Groan (p=0.56)

Generated by build_burst_demo_v2.py · pipeline in burst_captions.py. Every dimension, emotion, timestamp and burst on this page is a model prediction; nothing is hand-written.