OpenSVBench Scenario-driven SV leaderboard
Submission Flow

Choose what you want to submit.

OpenSVBench supports two separate submission paths: account-owned model score packages for leaderboard evaluation, and rolling dataset proposals for maintainer review.

18 systems already ranked 21 completed runs stored 26 trial sets in the benchmark
Model Scores

Upload score CSVs packaged as a submission zip. The server stores the run privately under your account; reviewed full-core runs can be approved for public ranking after an owner request.

Rolling Datasets

Upload a compact proposal package for a new dataset or trial set. Rolling proposals do not change the current core ranking by default.

Model Score Submission

Submit model score files

Use this path when you already have score files for OpenSVBench trial sets and want the server to evaluate the run.

  • Expected package: a zip with submission.json and scores/<trial_set_id>_scores.csv.
  • Score CSV columns must be utt1,utt2,score.
  • The server evaluates the run privately first; complete coverage makes it eligible for a public listing request.
Rolling Dataset Contribution

Contribute a rolling dataset

Use this path when you want to propose a new dataset or trial protocol for future rolling evaluation.

  • Upload a compact contribution package, not raw large audio.
  • Include data source, license notes, preparation code, protocol construction, and expected statistics.
  • Accepted rolling datasets are reviewed separately from the current core leaderboard.