Competitions — Sable Foundry — Light

Competitions

Public benchmarks on the open data commons. Solutions, scores and leaderboards flow back into the corpus — the competition is a dataset.

1 live 3 upcoming 214 entries 61 teams
LIVE · 18 DAYS LEFT SEASON 1 · OPEN COMMONS

Shona ASR Benchmark

Transcribe held-out full-duplex Shona conversation. Train on the open subset or anything you like — the test set stays sealed and scoring runs server-side. Lowest word error rate wins.

$2,500 prize pool + dataset credits Enter competition
Overview Data Leaderboard Rules Discussion 42
#TEAMWER ↓ENTRIESLAST
{{ row.rank }}
{{ row.initials }} {{ row.team }} {{ row.affil }} {{ row.delta }}
{{ row.wer }} {{ row.entries }} {{ row.last }}
baseline: whisper-large-v3 zero-shot · 41.2 WER scoring: server-side · 2 submissions/day
TIMELINE
{{ t.label }} {{ t.date }}
EVALUATION Word error rate on a sealed 48-minute test set spanning zezuru, karanga, manyika and korekore speakers, studio and remote channels.
WHAT FLOWS BACK Winning solutions are published; all scores join the benchmark corpus. Your work becomes national infrastructure — credited.

Upcoming seasons

institutions can sponsor a season under a data-sharing agreement
{{ u.tag }} {{ u.when }}
{{ u.title }} {{ u.desc }} metric: {{ u.metric }}