laion/synthetic_vocal_burts_dramabox

Synthetic Vocal Bursts (DramaBox)

A synthetic dataset of 2,000 vocal burst audio samples generated using DramaBox TTS on 8x A100 GPUs. The dataset covers 197 distinct vocal burst types across 16 age/gender groups, providing a comprehensive taxonomy of human non-speech vocalizations.

Dataset Summary

Property	Value
Total samples	2,000
Total duration	~4.2 hours (14,992 seconds)
Average duration	7.5 seconds
Duration range	3.0 – 12.0 seconds
Sample rate	44.1 kHz (WAV)
Gender split	1,000 female / 1,000 male
Age groups	16 (8 female, 8 male)
Samples per group	125
Unique vocal burst types	197
Generation model	DramaBox TTS
Hardware	8x NVIDIA A100 GPUs

Age Groups

The dataset is balanced across 16 age/gender groups (125 samples each):

Female:

• toddler_girl — Toddler girl
• pre_puberty_girl — Pre-puberty girl
• teenage_girl — Teenage girl
• young_woman — Young woman
• mature_woman — Mature woman
• middle_aged_woman — Middle-aged woman
• elderly_woman — Elderly woman
• senescent_woman — Senescent woman

Male:

• toddler_boy — Toddler boy
• pre_puberty_boy — Pre-puberty boy
• teenage_boy — Teenage boy
• young_man — Young man
• mature_man — Mature man
• middle_aged_man — Middle-aged man
• elderly_man — Elderly man
• senescent_man — Senescent man

Vocal Burst Taxonomy

The dataset uses an extended taxonomy of 197 vocal burst types (from an extended set of 202 entries), derived from multiple sources:

• Original LAION taxonomy — ~120 entries from LAION Voice-Acting-Pipeline
• Additional categories:
  • NSFW vocalizations
  • Exercise / primal sounds
  • Singing
  • Humming
  • Whistling

A SFW (safe-for-work) subset of 180 entries is also provided, filtering out NSFW content.

Both taxonomy files are included in this repository:

• vocal_bursts_taxonomy_extended.json — Full extended taxonomy (202 entries + metadata)
• vocal_bursts_taxonomy_sfw.json — SFW subset (180 entries + metadata)

Example Vocal Burst Types

Belly Laugh, Chuckle, Giggle, Snicker, Cackle, Gentle Sob, Wail, Whimper, Gasp, Sigh of Relief, Huff of Annoyance, Growl of Frustration, Startle Grunt, Nervous Whistling in the Dark, Battle Cry, Lullaby Humming, and many more.

Format

The dataset is stored as WebDataset tar shards in the data/ directory:

data/
  shard-0000.tar  (samples 0001–0500)
  shard-0001.tar  (samples 0501–1000)
  shard-0002.tar  (samples 1001–1500)
  shard-0003.tar  (samples 1501–2000)

Each shard is a standard tar archive. Each sample within a shard consists of two files sharing the same zero-padded 4-digit key:

• {NNNN}.wav — Audio file (WAV, 44.1 kHz)
• {NNNN}.json — Metadata JSON

Metadata Fields

{
  "id": 1,
  "prompt": "A toddler girl performing Nervous Whistling in the Dark, A shaky, slightly off-key whistling used to mask fear in an unsettling environment.",
  "duration_s": 5.2,
  "gender": "female",
  "age_group": "toddler_girl",
  "age_descriptor": "toddler girl",
  "vocal_burst_key": "Nervous Whistling in the Dark",
  "vocal_burst_description": "A shaky, slightly off-key whistling used to mask fear in an unsettling environment."
}

Usage

With webdataset (recommended for streaming/training)

import webdataset as wds
import json

url = "https://huggingface.co/datasets/laion/synthetic_vocal_burts_dramabox/resolve/main/data/shard-{0000..0003}.tar"

dataset = (
    wds.WebDataset(url, shardshuffle=True)
    .decode("pil")  # decodes common formats
    .to_tuple("wav", "json")
)

for wav_bytes, metadata_bytes in dataset:
    metadata = json.loads(metadata_bytes)
    print(metadata["vocal_burst_key"], metadata["age_descriptor"], len(wav_bytes), "bytes")
    break

With huggingface_hub (download individual shards)

from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="laion/synthetic_vocal_burts_dramabox",
    filename="data/shard-0000.tar",
    repo_type="dataset",
)

Direct tar extraction

import tarfile
import json

with tarfile.open("shard-0000.tar") as tar:
    members = tar.getmembers()
    for m in members[:4]:  # first 2 samples (wav + json each)
        print(m.name, m.size)

Additional Files

• prompts_manifest.csv — Full generation manifest with columns: id, prompt, output_path, gender, age_group, vocal_burst_key, vocal_burst_description, duration_s
• vocal_bursts_taxonomy_extended.json — Extended taxonomy (202 entries)
• vocal_bursts_taxonomy_sfw.json — SFW taxonomy (180 entries)

Generation Process

1. An extended vocal burst taxonomy was curated from the LAION Voice-Acting-Pipeline taxonomy plus additional categories
2. For each of the 16 age/gender groups, 125 vocal burst types were sampled from the taxonomy
3. Text prompts were generated in the format: "A {age_descriptor} performing {burst_key}, {burst_description}"
4. Audio was synthesized using DramaBox TTS distributed across 8x NVIDIA A100 GPUs
5. Each sample was generated with a randomized duration between 3–12 seconds

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Citation

If you use this dataset, please cite:

@dataset{synthetic_vocal_bursts_dramabox_2026,
  title={Synthetic Vocal Bursts (DramaBox)},
  author={LAION},
  year={2026},
  url={https://huggingface.co/datasets/laion/synthetic_vocal_burts_dramabox},
  license={CC-BY-4.0}
}