OST-DiagBench is released for non-commercial academic research only. The video clips are short, transformed excerpts derived from public academic audio-visual corpora (AVE, VGGSound, MUSIC-AVQA-R, OmniPro, Daily-Omni, StreamingBench, OmniVideoBench, Video-Holmes) whose material originates from user-uploaded YouTube content; the donor sounds come from ESC-50 under CC BY-NC 3.0. We are not the rights holders of the underlying media. By requesting access you agree to (1) use the media solely for non-commercial academic research, teaching, and evaluation; (2) not use it for any commercial purpose, including commercial model training or productization; (3) not re-upload, mirror, or redistribute the media, and not attempt to reconstruct or publish the full source videos; (4) respect the terms of every upstream corpus and the YouTube Terms of Service; and (5) cite OST-DiagBench and the relevant upstream corpora in any resulting publication. See LICENSES.md for the full notice and the takedown procedure.

OST-DiagBench

OST-DiagBench contains exactly four diagnostic operations with 510 cases per operation: 2,040 cases in total. Every case includes an MP4 input and a paired M4A audio file.

Academic research use only, no commercial use. The clips are short, transformed excerpts derived from public academic audio-visual corpora whose material originates from user-uploaded YouTube content; the donor sounds come from ESC-50 under CC BY-NC 3.0. We are not the rights holders of the underlying media. See License and intended use and LICENSES.md.

Operation	IDs	Construction	What it tests
MUTE	00001–00510	Keep the video unchanged and remove the visually suggested source sound from the soundtrack; non-target background audio may remain.	Whether a model claims to hear a sound suggested by the video even when that sound is absent.
SWAP	00511–01020	Keep the video unchanged, remove the original source sound, and insert an unrelated donor sound.	Whether a model follows the sound that is actually audible instead of the visual suggestion.
MIX	01021–01530	Keep the video unchanged, retain the original source audio, and add a donor sound.	Whether a model recognizes both simultaneously audible source and donor sounds.
CLASH	01531–02040	Keep the on-screen subtitle/visual fact, but replace the corresponding spoken fact with a mutually exclusive value.	Whether a model answers from the spoken audio or copies the conflicting on-screen subtitle.

Files

data/
  mute.jsonl             510 MUTE rows
  swap.jsonl             510 SWAP rows
  mix.jsonl              510 MIX rows
  clash.jsonl            510 CLASH rows
  audio_event.jsonl      MUTE + SWAP + MIX (1,530 rows)
  ost_diagbench.jsonl    all four operations (2,040 rows)
media/{mute,swap,mix,clash}/
  0001.mp4 ... 0510.mp4
  0001.m4a ... 0510.m4a
eval/                    scoring scripts

The public id is unique across the full dataset. operation_local_id runs from 0001 to 0510 inside each operation. See data/SCHEMA.md for the manifest fields and LICENSES.md for upstream media terms.

Download

pip install -U huggingface_hub
hf auth login
python - <<'PY'
from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="EnjunDu/OST-Diagbench",
    repo_type="dataset",
    local_dir="OST-DiagBench",
)
PY

Install the evaluation dependencies from the downloaded directory:

cd OST-DiagBench
pip install -r requirements.txt

MUTE, SWAP, and MIX evaluation

Run each MP4 in data/audio_event.jsonl with the exact prompt:

Describe all salient sounds you hear in this clip.

Write one prediction per line:

{"id":"00001","description":"I hear wind and distant traffic.","model":"my-model"}

Then score the predictions:

python eval/evaluate.py \
  --data data/audio_event.jsonl \
  --predictions predictions_audio_event.jsonl \
  --output-dir outputs/audio_event

The scorer reports MUTE visual-hallucination recall, SWAP audible-donor and visually suggested source recall, and MIX source/donor/joint recall.

CLASH evaluation

For each CLASH row, ask its question twice:

condition="av": run media_path (the MP4 with conflicting speech and subtitle).
condition="audio_only": run audio_path (the paired M4A without the visual subtitle).

Write predictions without copying the ground-truth fields into the prediction file:

{"id":"01531","condition":"av","answer":"64 percent","model":"my-model"}
{"id":"01531","condition":"audio_only","answer":"64 percent","model":"my-model"}

Score them by joining against the CLASH manifest:

python eval/score_clash.py \
  --data data/clash.jsonl \
  --predictions predictions_clash.jsonl \
  --output outputs/clash_scored.jsonl

Each answer is labeled audio, subtitle, both, or neither, with separate summaries for the AV and audio-only conditions.

License and intended use

OST-DiagBench is distributed for non-commercial academic research only. It is an aggregate research benchmark rather than a media distribution service, and there is no single blanket license covering all bundled media.

Media provenance. Every visual clip is a short excerpt taken from a public academic audio-visual corpus: AVE, VGGSound, MUSIC-AVQA-R, OmniPro, Daily-Omni, StreamingBench, OmniVideoBench, and Video-Holmes. Those corpora index user-uploaded YouTube videos, so our clips inherit that provenance. We are neither the author nor the rights holder of that footage, and each source corpus remains governed by its own terms. The donor sound bank is derived from the official ESC-50 release under CC BY-NC 3.0, with per-item attribution kept in data/donor_assets.jsonl. Annotations, manifests, and evaluation code are ours and are released for research use; they do not override any upstream media license or grant additional rights.

Nature of the excerpts. Clips are deliberately short and deliberately altered — muted (MUTE), sound-replaced (SWAP), donor-mixed (MIX), or with speech made to conflict with the on-screen subtitle (CLASH) — purely so that model behaviour can be measured. They are not usable as substitutes for the original works and are not intended to compete with them in any market.

By requesting access you agree to:

use the media solely for non-commercial academic research, teaching, and evaluation;
not use it for any commercial purpose, including commercial model training, productization, resale, advertising, or redistribution as a media product;
not re-upload, mirror, or redistribute the media outside this repository, and not attempt to reconstruct or publish the full source videos;
respect the terms of every upstream corpus, of the original hosting platform (including the YouTube Terms of Service), and of each original uploader;
cite OST-DiagBench and the relevant upstream corpora in any publication that uses this benchmark.

Takedown requests. If you are a rights holder and believe a clip should not be included, please open a discussion on this repository and identify the record by public id, source_dataset, and source identifier/fingerprint. We will review and remove affected records promptly. Local checkout paths are intentionally not published, so requests can be handled through these public identifiers alone.

The dataset is provided "as is", without warranty of any kind. Users are responsible for ensuring their own use complies with applicable law and with the terms above. The full notice is in LICENSES.md.

Downloads last month: 55

Total file size:

3.49 GB