Judge — for every (sample × threshold) we send the audio plus the produced
(start, end, caption) list to Gemini 3.1 Pro and ask for three integer 0–5 scores
(5 = perfect, 0 = random): caption quality, timestamp accuracy, and completeness
(do the detections cover ALL real vocal bursts, penalizing both misses and false positives).
The goal: find the detection confidence threshold that maximises the overall Gemini judgment.
Pipeline & models live in the Universal Audio Annotation Pipeline repo.
Scoring of empty detections: when a clip has no detections, caption/timestamp are N/A
(excluded from those averages) and completeness is judged on whether real bursts were missed.
Overall = mean(completeness, caption, timestamp).
Results — average Gemini scores per threshold (ranked by overall)
threshold
overall
completeness
caption quality
timestamp accuracy
clips w/ detection
🥇 0.88
3.475
3.113
3.244
4.067
90/150
🥈 0.89
3.469
3.148
3.18
4.079
90/150
🥉 0.85
3.466
3.108
3.222
4.067
92/150
0.9
3.445
3.095
3.239
4.0
90/150
0.86
3.411
3.047
3.143
4.044
91/150
0.87
3.39
3.068
3.102
4.0
90/150
0.91
3.364
3.047
3.136
3.909
89/150
0.92
3.363
3.02
3.149
3.92
89/150
Averages over 150 samples (caption & timestamp averaged over clips that produced ≥1 detection).
🥇🥈🥉 mark the top three, shown with audio below.
6.56–8.98sconf 0.90 The audio features a distinct, deep sigh, characterized by a slow, audible exhalation. The sound is prolonged and conveys a sense of relief or resignation.
5.80–13.92sconf 1.00 The audio features a series of distinct, low-pitched vocalizations, resembling soft grunts or soft sighs. These sounds are repetitive and have a somewhat guttural quality, suggesting a relaxed or contemplative state.
52.32–56.58sconf 1.00 The audio features a series of distinct, forceful coughs, characteristic of a human clearing their throat. The coughs are sharp and somewhat wet, suggesting the presence of mucus.
42.22–48.16sconf 1.00 The audio features a series of distinct, low-pitched vocalizations, resembling grunts or throat-clearing sounds. These sounds are short, guttural, and occur intermittently, suggesting a deliberate effort to clear the throat.
52.86–60.00sconf 1.00 The audio features a series of distinct, low-pitched chuckles from a male voice, interspersed with brief pauses. The laughter is light and airy, suggesting amusement.
48.10–54.92sconf 0.99 The audio features distinct, wet smacking sounds, consistent with the act of drinking. The sounds are sharp and close, suggesting a direct interaction with a liquid.
47.74–54.14sconf 1.00 The audio features a series of distinct, high-pitched vocalizations, resembling laughter or giggling, produced by a person with a perceived female voice. The laughter is light and airy, with a clear, joyful quality.
4.18–13.10sconf 1.00 The audio features a series of distinct, low-pitched vocalizations that resemble humming or humming. These sounds are sustained and have a resonant quality, suggesting a human voice.
49.24–54.58sconf 1.00 The audio features a female voice, perceived as adult, emitting a series of high-pitched, lighthearted giggles. The giggles are brief and punctuated, suggesting a playful or amused emotional state.
44.26–50.86sconf 0.96 and laughter. The audio contains distinct human laughter, characterized by a high-pitched, almost squealing quality, suggesting a strong expression of amusement or mirth.
0.18–2.92sconf 0.99 The audio features a distinct, high-pitched vocalization, resembling a wail or a cry, produced by a person with a perceived female voice. The sound is sustained and appears to be a continuous expression of distress or pain.
0.34–3.54sconf 1.00 A high-pitched, sustained vocalization, resembling a wail or scream, is present. The sound is loud and piercing, with a clear, piercing quality.
0.48–1.38sconf 0.99 The audio features a high-pitched, sustained vocalization, resembling a wail or cry, produced by a person with a perceived female voice. The sound is continuous and conveys a sense of distress or intense emotion.
0.26–2.98sconf 0.99 A high-pitched, sustained vocalization, resembling a wail or scream, is present. The sound is loud and appears to be produced by a human, with a distinct, piercing quality.
0.48–2.00sconf 0.99 A high-pitched, sustained vocalization, resembling a wail or a cry, is present. The sound is continuous and wavering in pitch, conveying a sense of distress or urgency.
0.08–1.96sconf 0.98 The audio features a distinct, high-pitched vocalization, resembling a sustained wail or cry, produced by a person with a perceived female voice. The sound is continuous and appears to be produced with significant vocal effort.
0.32–2.66sconf 0.99 A low-pitched, sustained vocalization, perceived as male, is present. The sound is a continuous, deep hum with a resonant quality, resembling a prolonged moan or a deep sigh.
0.16–2.72sconf 0.95 The audio features a distinct, deep inhalation followed by a prolonged, audible exhalation. The sound is characterized by a slow, deliberate inhale and exhale, indicating a significant release of air.
0.04–2.54sconf 0.96 The audio features a distinct, sharp, and wet-sounding inhalation followed by a brief, forceful exhalation. The sound is clear and close, suggesting a direct and forceful expulsion of air.
0.04–2.78sconf 0.99 The audio features a series of sharp, quick, and distinct gasping sounds, characteristic of a person struggling to catch their breath. The gasps are quick and shallow, indicating a sudden intake of air.
0.02–2.50sconf 1.00 The audio features a series of distinct, sharp, and forceful exhalations, characteristic of panting. These sounds are consistent with rapid, shallow breaths, indicating a state of physical exertion or distress.
0.06–2.44sconf 1.00 The audio features a series of distinct, sharp, and forceful exhalations, characteristic of heavy breathing or panting. The sounds are consistent in their intensity and rhythm, suggesting a continuous physiological process.
21.56–28.28sconf 0.95 The audio features a series of distinct, wet, and guttural vocalizations, consistent with a burp or a deep, wet cough. The sounds are short, sharp, and occur in quick succession.
39.96–47.48sconf 1.00 The audio features a series of distinct, high-pitched vocalizations, resembling laughter or giggling, produced by a person with a perceived female voice. The sounds are short, sharp, and convey a sense of lightheartedness and amusement.
0.02–8.46sconf 0.99 The audio features a series of distinct, wet, and gurgling sounds, consistent with liquid being poured or slurped. These sounds are accompanied by a subtle, continuous, low-frequency hum in the background.
12.26–16.14sconf 0.99 The audio features a series of distinct, high-pitched, and somewhat strained vocalizations, resembling laughter. These sounds are short and sharp, with a clear, almost cartoonish quality.
23.26–28.14sconf 0.95 The audio features a series of distinct, sharp, and percussive sounds, consistent with the sound of a whip being cracked. Each sound is clear and forceful, with a brief, resonant decay.
32.10–37.82sconf 0.98 The audio features a series of distinct, low-pitched vocalizations, resembling grunts or short, guttural sounds. These sounds are repetitive and have a somewhat muffled quality, suggesting a non-verbal expression.
47.90–54.36sconf 0.99 The audio features a series of distinct, high-pitched vocalizations, resembling laughter or giggling, produced by a person perceived as female. The sounds are short, sharp, and repetitive, suggesting a lighthearted or playful mood.
16.62–22.70sconf 0.99 The audio features a series of high-pitched, somewhat strained vocalizations, resembling laughter or giggling, produced by a person perceived as female. The sounds are short, sharp, and repetitive, with a light, airy quality.
33.08–37.40sconf 0.98 The audio features a person laughing, with a distinct, high-pitched, and somewhat strained quality. The laughter is brief and appears to be a single, isolated vocalization.
46.54–47.04sconf 0.93 The audio features a single, distinct vocalization from a perceived male speaker. The sound is a guttural, low-pitched grunt, characterized by a deep, resonant quality.
47.40–49.88sconf 0.95 The audio features a series of distinct, wet, and guttural vocalizations, consistent with a burp or retch. The sounds are short, sharp, and occur in quick succession.
0.20–1.72sconf 1.00 A high-pitched, sustained vocalization, resembling a wail or scream, is present. The sound is loud and piercing, with a distinct, sharp quality, and is repeated several times.
0.50–2.32sconf 1.00 A high-pitched, sustained vocalization, resembling a scream or shriek, is heard. The sound is loud and piercing, with a distinct, sharp quality, and is repeated several times.
6.56–8.98sconf 0.90 The audio features a distinct, deep sigh, characterized by a slow, audible exhalation. The sound is prolonged and conveys a sense of relief or resignation.
5.80–13.92sconf 1.00 The audio features a series of distinct, low-pitched vocalizations, resembling soft grunts or soft sighs. These sounds are repetitive and have a somewhat guttural quality, suggesting a relaxed or contemplative state.
52.32–56.58sconf 1.00 The audio features a series of distinct, forceful coughs, characteristic of a human clearing their throat. The coughs are sharp and somewhat wet, suggesting the presence of mucus.
48.10–54.92sconf 0.99 The audio features distinct, wet smacking sounds, consistent with the act of drinking. The sounds are sharp and close, suggesting a direct interaction with a liquid.
47.74–54.14sconf 1.00 The audio features a series of distinct, high-pitched vocalizations, resembling laughter or giggling, produced by a person with a perceived female voice. The laughter is light and airy, with a clear, joyful quality.
4.20–13.10sconf 1.00 The audio features a series of distinct, low-pitched vocalizations that resemble humming or humming. These sounds are sustained and have a resonant quality, suggesting a human voice.
49.24–54.56sconf 1.00 The audio features a female voice, perceived as adult, emitting a series of high-pitched, lighthearted giggles. The giggles are brief and punctuated, suggesting a playful or amused emotional state.
44.26–50.86sconf 0.96 and laughter. The audio contains distinct human laughter, characterized by a high-pitched, almost squealing quality, suggesting a strong expression of amusement or mirth.
0.18–2.92sconf 0.99 The audio features a distinct, high-pitched vocalization, resembling a wail or a cry, produced by a person with a perceived female voice. The sound is sustained and appears to be a continuous expression of distress or pain.
0.48–1.38sconf 0.99 The audio features a high-pitched, sustained vocalization, resembling a wail or cry, produced by a person with a perceived female voice. The sound is continuous and conveys a sense of distress or intense emotion.
0.06–2.44sconf 0.99 A high-pitched, sustained vocalization, resembling a scream or shriek, is present. The sound is loud and piercing, with a distinct, sharp quality, and is repeated several times.
0.26–2.98sconf 0.99 A high-pitched, sustained vocalization, resembling a wail or scream, is present. The sound is loud and appears to be produced by a human, with a distinct, piercing quality.
0.48–2.00sconf 0.99 A high-pitched, sustained vocalization, resembling a wail or a cry, is present. The sound is continuous and wavering in pitch, conveying a sense of distress or urgency.
0.08–1.94sconf 0.98 The audio features a distinct, high-pitched vocalization, resembling a sustained wail or cry, produced by a person with a perceived female voice. The sound is continuous and appears to be produced with significant vocal effort.
0.32–2.66sconf 0.99 A low-pitched, sustained vocalization, perceived as male, is present. The sound is a continuous, deep hum with a resonant quality, resembling a prolonged moan or a deep sigh.
0.16–2.72sconf 0.95 The audio features a distinct, deep inhalation followed by a prolonged, audible exhalation. The sound is characterized by a slow, deliberate inhale and exhale, indicating a significant release of air.
0.04–2.78sconf 0.99 The audio features a series of sharp, quick, and distinct gasping sounds, characteristic of a person struggling to catch their breath. The gasps are quick and shallow, indicating a sudden intake of air.
0.02–2.50sconf 1.00 The audio features a series of distinct, sharp, and forceful exhalations, characteristic of panting. These sounds are consistent with rapid, shallow breaths, indicating a state of physical exertion or distress.
0.06–2.44sconf 1.00 The audio features a series of distinct, sharp, and forceful exhalations, characteristic of heavy breathing or panting. The sounds are consistent in their intensity and rhythm, suggesting a continuous physiological process.
42.22–48.14sconf 1.00 The audio features a series of distinct, low-pitched vocalizations, resembling grunts or throat-clearing sounds. These sounds are short, guttural, and occur intermittently, suggesting a deliberate effort to clear the throat.
52.86–60.00sconf 1.00 The audio features a series of distinct, low-pitched chuckles from a male voice, interspersed with brief pauses. The laughter is light and airy, suggesting amusement.
25.78–29.78sconf 0.96 into the audio are high-pitched, rapid vocalizations resembling laughter. The laughter is light and airy, with a distinct, almost childlike quality, and is interspersed with short, sharp inhalations. The overall impression is one of lighthearted amusement.
52.66–56.14sconf 0.96 The audio features a series of distinct, high-pitched vocalizations resembling squeaks or squeaks. These sounds are short, sharp, and repetitive, occurring in quick succession.
56.56–59.66sconf 0.97 The audio features a high-pitched, childlike giggle, followed by a distinct, sharp, and somewhat strained laugh. The giggle is brief and punctuated.
21.56–28.28sconf 0.95 The audio features a series of distinct, wet, and guttural vocalizations, consistent with a burp or a deep, wet cough. The sounds are short, sharp, and occur in quick succession.
39.96–47.48sconf 1.00 The audio features a series of distinct, high-pitched vocalizations, resembling laughter or giggling, produced by a person with a perceived female voice. The sounds are short, sharp, and convey a sense of lightheartedness and amusement.
16.62–22.70sconf 0.99 The audio features a series of high-pitched, somewhat strained vocalizations, resembling laughter or giggling, produced by a person perceived as female. The sounds are short, sharp, and repetitive, with a light, airy quality.
33.08–37.40sconf 0.98 The audio features a person laughing, with a distinct, high-pitched, and somewhat strained quality. The laughter is brief and appears to be a single, isolated vocalization.
46.54–47.04sconf 0.93 The audio features a single, distinct vocalization from a perceived male speaker. The sound is a guttural, low-pitched grunt, characterized by a deep, resonant quality.
47.40–49.86sconf 0.95 The audio features a series of distinct, wet, and guttural vocalizations, consistent with a burp or retch. The sounds are short, sharp, and occur in quick succession.
0.36–3.54sconf 1.00 A high-pitched, sustained vocalization, resembling a wail or scream, is present. The sound is loud and piercing, with a distinct, sharp quality, and is repeated several times.
0.50–2.32sconf 1.00 A high-pitched, sustained vocalization, resembling a scream or shriek, is heard. The sound is loud and piercing, with a distinct, sharp quality, and is repeated several times.
11.40–12.40sconf 0.87 The audio features a distinct, high-pitched, and sustained whistling sound. The whistle is clear and consistent in pitch, with a slight wavering quality, suggesting a deliberate and continuous emission.
5.80–13.92sconf 1.00 The audio features a series of distinct, low-pitched vocalizations, resembling soft grunts or soft sighs. These sounds are repetitive and have a somewhat guttural quality, suggesting a relaxed or contemplative state.
52.32–56.66sconf 0.99 The audio features a series of distinct, forceful coughs, characteristic of a human clearing their throat. The coughs are sharp and somewhat wet, suggesting the presence of mucus.
42.22–48.16sconf 1.00 The audio features a series of distinct, low-pitched vocalizations, resembling grunts or throat-clearing sounds. These sounds are short, guttural, and occur intermittently, suggesting a deliberate effort to clear the throat.
52.62–60.00sconf 0.99 The audio features a series of distinct, high-pitched vocalizations, resembling laughter or giggling, produced by a person perceived as female. The laughter is light and airy, with a clear, joyful quality.
4.18–13.10sconf 1.00 The audio features a series of distinct, low-pitched vocalizations that resemble humming or humming. These sounds are sustained and have a resonant quality, suggesting a human voice.
49.24–54.58sconf 1.00 The audio features a female voice, perceived as adult, emitting a series of high-pitched, lighthearted giggles. The giggles are brief and punctuated, suggesting a playful or amused emotional state.
33.18–37.26sconf 0.99 a brief, high-pitched vocalization resembling a giggle or a soft, suppressed laugh. The sound is a light, airy laugh, possibly from a child or a person with a higher vocal register.
42.42–49.52sconf 0.98 The audio features a series of distinct, low-pitched vocalizations that resemble a deep, guttural hum or a soft, sustained groan. These sounds are produced at a slow, deliberate pace, with a resonant quality.
0.16–2.94sconf 0.99 The audio features a distinct, high-pitched vocalization, resembling a wail or a cry, produced by a person with a perceived female voice. The sound is sustained and appears to be a continuous expression of distress or pain.
0.34–3.54sconf 1.00 A high-pitched, sustained vocalization, resembling a wail or scream, is present. The sound is loud and piercing, with a clear, piercing quality.
0.48–1.40sconf 0.99 The audio features a high-pitched, sustained vocalization, resembling a wail or cry, produced by a person with a perceived female voice. The sound is continuous and conveys a sense of distress or intense emotion.
0.26–2.98sconf 0.99 A high-pitched, sustained vocalization, resembling a wail or scream, is present. The sound is loud and appears to be produced by a human, with a distinct, piercing quality.
0.48–2.00sconf 0.99 A high-pitched, sustained vocalization, resembling a wail or a cry, is present. The sound is continuous and wavering in pitch, conveying a sense of distress or urgency.
0.08–2.00sconf 0.98 The audio features a distinct, high-pitched vocalization, resembling a sustained wail or cry, produced by a person with a perceived female voice. The sound is continuous and appears to be produced with significant vocal effort.
0.16–2.72sconf 0.95 The audio features a distinct, deep inhalation followed by a prolonged, audible exhalation. The sound is characterized by a slow, deliberate inhale and exhale, indicating a significant release of air.
0.06–3.32sconf 0.93 The audio features a distinct, deep, and prolonged vocalization, characteristic of a sigh. The sound is drawn out and has a low pitch, suggesting a release of air.
0.04–2.54sconf 0.96 The audio features a distinct, sharp, and wet-sounding inhalation followed by a brief, forceful exhalation. The sound is clear and close, suggesting a direct and forceful expulsion of air.
0.04–2.78sconf 0.99 The audio features a series of sharp, quick, and distinct gasping sounds, characteristic of a person struggling to catch their breath. The gasps are quick and shallow, indicating a sudden intake of air.
0.02–2.50sconf 1.00 The audio features a series of distinct, sharp, and forceful exhalations, characteristic of panting. These sounds are consistent with rapid, shallow breaths, indicating a state of physical exertion or distress.
0.06–2.44sconf 1.00 The audio features a series of distinct, sharp, and forceful exhalations, characteristic of heavy breathing or panting. The sounds are consistent in their intensity and rhythm, suggesting a continuous physiological process.
6.56–8.98sconf 0.90 The audio features a distinct, deep sigh, characterized by a slow, audible exhalation. The sound is prolonged and conveys a sense of relief or resignation.
17.88–19.26sconf 0.96 The audio features a distinct, sharp intake of breath, characteristic of a gasp. This is immediately followed by a short, low-pitched vocalization, perceived as a grunt or a soft groan, which then transitions into a more sustained, low-pitched vocalization.
47.74–54.14sconf 1.00 The audio features a series of distinct, high-pitched vocalizations, resembling laughter or giggling, produced by a person with a perceived female voice. The laughter is light and airy, with a clear, joyful quality.
25.78–29.80sconf 0.96 into the audio are high-pitched, rapid vocalizations resembling laughter. The laughter is light and airy, with a distinct, almost childlike quality, and is interspersed with short, sharp inhalations. The overall impression is one of lighthearted amusement.
44.26–50.88sconf 0.96 The audio features a series of high-pitched, somewhat strained vocalizations, resembling laughter or giggling, produced by a person perceived as female. The laughter is rapid and continuous, with a light, airy quality.
52.66–59.68sconf 0.96 The audio features multiple instances of high-pitched, rapid vocalizations, resembling squeaks or squeaks. These sounds are short, sharp, and repetitive, occurring in quick succession.
6.08–9.90sconf 0.93 and laughter. The audio captures a burst of high-pitched, unrestrained laughter, indicating amusement or joy.
26.16–29.92sconf 0.94 The audio features a distinct, high-pitched vocalization, perceived as a male voice, characterized by a strained, somewhat forced laugh. The laughter is brief and appears to be a single, isolated burst.
36.74–41.58sconf 1.00 The audio features a series of distinct, wet, and guttural vocalizations, resembling a series of coughs. These sounds are sharp and forceful, occurring in rapid succession.
Audio embedded as base64 mp3.
Judge: Gemini 3.1 Pro (HyprLab). Detector: laion/vocalburst-locator. Captioner: laion/vocalburst-captioning-whisper.